WEKA Manual for Version 3-6-13
Remco R. Bouckaert Eibe Frank Mark Hall Richard Kirkby Peter Reutemann Alex Seewald David Scuse
September 9, 2015
⃝c 2002-2015
University of Waikato, Hamilton, New Zealand Alex Seewald (original Commnd-line primer) David Scuse (original Experimenter tutorial)
This manual is licensed under the GNU General Public License version 2. More information about this license can be found at http://www.gnu.org/copyleft/gpl.html
Contents
I
1
II
2 3
4
The Command-line 5
A command-line primer 7
1.1 Introduction………………………… 7
1.2 Basicconcepts ………………………. 8 1.2.1 Dataset ………………………. 8 1.2.2 Classifier ……………………… 10 1.2.3 weka.filters …………………….. 11 1.2.4 weka.classifiers …………………… 13
1.3 Examples …………………………. 17
The Graphical User Interface 19
Launching WEKA 21
Simple CLI 25
3.1 Commands ………………………… 25 3.2 Invocation…………………………. 26 3.3 Commandredirection …………………… 26 3.4 Commandcompletion …………………… 27
Explorer 29
4.1 Theuserinterface …………………….. 29 4.1.1 SectionTabs ……………………. 29 4.1.2 StatusBox …………………….. 29 4.1.3 LogButton …………………….. 30 4.1.4 WEKAStatusIcon…………………. 30 4.1.5 Graphicaloutput………………….. 30
4.2 Preprocessing……………………….. 31 4.2.1 LoadingData……………………. 31 4.2.2 TheCurrentRelation ……………….. 31 4.2.3 WorkingWithAttributes ……………… 32 4.2.4 WorkingWithFilters ……………….. 33
4.3 Classification ……………………….. 35 4.3.1 SelectingaClassifier………………… 35 4.3.2 TestOptions ……………………. 35 4.3.3 TheClassAttribute ………………… 36 4.3.4 TrainingaClassifier ………………… 37
3
4
CONTENTS
4.4
4.5 4.6
4.7
4.3.5 TheClassifierOutputText …………….. 37 4.3.6 TheResultList…………………… 37 Clustering…………………………. 39 4.4.1 SelectingaClusterer………………… 39 4.4.2 ClusterModes …………………… 39 4.4.3 IgnoringAttributes…………………. 39 4.4.4 WorkingwithFilters………………… 40 4.4.5 LearningClusters………………….. 40 Associating ………………………… 41 4.5.1 SettingUp …………………….. 41 4.5.2 LearningAssociations ……………….. 41 SelectingAttributes ……………………. 42 4.6.1 SearchingandEvaluating ……………… 42 4.6.2 Options ………………………. 42 4.6.3 PerformingSelection………………… 42 Visualizing ………………………… 44 4.7.1 Thescatterplotmatrix ………………. 44 4.7.2 Selectinganindividual2Dscatterplot . . . . . . . . . . . 44 4.7.3 SelectingInstances…………………. 45
5 Experimenter 47
5.1 Introduction………………………… 47
5.2 StandardExperiments …………………… 48
5.2.1 Simple……………………….. 48 5.2.1.1 Newexperiment ……………… 48 5.2.1.2 Resultsdestination ……………. 48 5.2.1.3 Experimenttype……………… 50 5.2.1.4 Datasets …………………. 52 5.2.1.5 Iterationcontrol ……………… 53 5.2.1.6 Algorithms ………………… 53 5.2.1.7 Savingthesetup……………… 55 5.2.1.8 RunninganExperiment………….. 56
5.2.2 Advanced ……………………… 57
5.2.2.1 DefininganExperiment………….. 57
5.2.2.2 RunninganExperiment………….. 60
5.2.2.3 Changing the Experiment Parameters . . . . . . 62
5.2.2.4 OtherResultProducers………….. 69
5.3 ClusterExperiments ……………………. 74
5.4 RemoteExperiments……………………. 77 5.4.1 Preparation…………………….. 77 5.4.2 DatabaseServerSetup……………….. 77 5.4.3 RemoteEngineSetup ……………….. 78 5.4.4 ConfiguringtheExperimenter……………. 79 5.4.5 Multi-coresupport…………………. 80 5.4.6 Troubleshooting ………………….. 80
5.5 AnalysingResults……………………… 82 5.5.1 Setup ……………………….. 82 5.5.2 SavingtheResults …………………. 85 5.5.3 ChangingtheBaselineScheme …………… 85 5.5.4 StatisticalSignificance……………….. 86
CONTENTS 5 5.5.5 SummaryTest …………………… 86
5.5.6 RankingTest ……………………. 87
6 KnowledgeFlow 89
6.1 Introduction………………………… 89
6.2 Features………………………….. 91
6.3 Components………………………… 92
6.3.1 DataSources…………………….. 92 6.3.2 DataSinks……………………… 92 6.3.3 Filters……………………….. 92 6.3.4 Classifiers……………………… 92 6.3.5 Clusterers……………………… 92 6.3.6 Evaluation …………………….. 93 6.3.7 Visualization ……………………. 94
6.4 Examples …………………………. 95 6.4.1 Cross-validatedJ48 ………………… 95 6.4.2 PlottingmultipleROCcurves……………. 97 6.4.3 Processingdataincrementally……………. 99
6.5 PluginFacility ……………………….101
7 ArffViewer 103
7.1 Menus……………………………104 7.2 Editing …………………………..106
8 Bayesian Network Classifiers 109
8.1 Introduction…………………………109
8.2 Localscorebasedstructurelearning …………….113 8.2.1 Localscoremetrics …………………113 8.2.2 Searchalgorithms ………………….114
8.3 Conditional independence test based structure learning . . . . . . 117
8.4 Globalscoremetricbasedstructurelearning. . . . . . . . . . . . 119
8.5 Fixedstructure’learning’ ………………….120
8.6 Distributionlearning…………………….120
8.7 Runningfromthecommandline ………………122
8.8 InspectingBayesiannetworks………………..132
8.9 BayesNetworkGUI …………………….135
8.10Bayesiannetsintheexperimenter ……………..147 8.11 AddingyourownBayesiannetworklearners . . . . . . . . . . . . 147 8.12 FAQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 8.13Futuredevelopment …………………….150
III Data 153
9 ARFF 155
9.1 Overview ………………………….155
9.2 Examples ………………………….156 9.2.1 TheARFFHeaderSection ……………..156 9.2.2 TheARFFDataSection……………….158
9.3 SparseARFFfiles ……………………..159
6 CONTENTS 9.4 InstanceweightsinARFFfiles ……………….160
10 XRFF 161
10.1Fileextensions ……………………….161
10.2 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 10.2.1 ARFF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 10.2.2 XRFF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.3 Sparse format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
10.4Compression ………………………..164 10.5Usefulfeatures ……………………….164 10.5.1 Class attribute specification . . . . . . . . . . . . . . . . . 164 10.5.2 Attribute weights . . . . . . . . . . . . . . . . . . . . . . . 164 10.5.3Instanceweights …………………..165
11 Converters 167
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 11.2Usage ……………………………168 11.2.1Fileconverters ……………………168 11.2.2 Database converters . . . . . . . . . . . . . . . . . . . . . 168
12 Stemmers 171
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
12.2 Snowball stemmers . . . . . . . . . . . . . . . . . . . . . . . . . . 171
12.3 Using stemmers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
12.3.1 Commandline . . . . . . . . . . . . . . . . . . . . . . . . . 172
12.3.2 StringToWordVector . . . . . . . . . . . . . . . . . . . . . 172
12.4Addingnewstemmers ……………………172
13 Databases 173
13.1Configurationfiles ……………………..173 13.2Setup ……………………………174 13.3MissingDatatypes ……………………..175 13.4StoredProcedures ……………………..176 13.5 Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
14 Windows databases 179
IV Appendix 183
15 Research 185
15.1 Citing Weka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 15.2Paperreferences ………………………185
16 Using the API 189
16.1Optionhandling ………………………190 16.2Loadingdata ………………………..192 16.2.1Loadingdatafromfiles ……………….192 16.2.2 Loading data from databases . . . . . . . . . . . . . . . . 193 16.3Creatingdatasetsinmemory ………………..196 16.3.1Definingtheformat …………………196
CONTENTS 7
16.3.2Addingdata …………………….197 16.4Generatingartificialdata ………………….199 16.4.1GenerateARFFfile …………………199 16.4.2 Generate Instances . . . . . . . . . . . . . . . . . . . . . . 199 16.5Randomizingdata ……………………..200 16.6 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 16.6.1Batchfiltering ……………………202 16.6.2 Filtering on-the-fly . . . . . . . . . . . . . . . . . . . . . . 203 16.7Classification ………………………..204 16.7.1Buildingaclassifier …………………204 16.7.2Evaluatingaclassifier ………………..206 16.7.3 Classifying instances . . . . . . . . . . . . . . . . . . . . . 209
16.8 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 16.8.1Buildingaclusterer …………………211 16.8.2 Evaluating a clusterer . . . . . . . . . . . . . . . . . . . . 213 16.8.3Clusteringinstances …………………215
16.9 Selecting attributes . . . . . . . . . . . . . . . . . . . . . . . . . . 216 16.9.1 Using the meta-classifier . . . . . . . . . . . . . . . . . . . 217 16.9.2 Using the filter . . . . . . . . . . . . . . . . . . . . . . . . 218 16.9.3 Using the API directly . . . . . . . . . . . . . . . . . . . . 219
16.10Saving data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 16.10.1Saving data to files . . . . . . . . . . . . . . . . . . . . . . 220 16.10.2Savingdatatodatabases ………………220 16.11Visualization ………………………..222 16.11.1ROC curves . . . . . . . . . . . . . . . . . . . . . . . . . . 222 16.11.2Graphs ……………………….223 16.11.2.1 Tree . . . . . . . . . . . . . . . . . . . . . . . . . 223 16.11.2.2BayesNet . . . . . . . . . . . . . . . . . . . . . . 224 16.12Serialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
17 Extending WEKA 227
17.1WritinganewClassifier …………………..229 17.1.1Choosingthebaseclass ……………….229
17.1.2 Additional interfaces . . . . . . . . . . . . . . . . . . . . . 229
17.1.3 Packages. . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
17.1.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . 231
17.1.4.1 Methods ………………….231
17.1.4.2 Guidelines …………………235
17.2 Writing a new Filter . . . . . . . . . . . . . . . . . . . . . . . . . 241
17.2.1 Default approach . . . . . . . . . . . . . . . . . . . . . . . 241 17.2.1.1 Implementation ………………241 17.2.1.2 Examples………………….244
17.2.2 Simple approach . . . . . . . . . . . . . . . . . . . . . . . 248 17.2.2.1 SimpleBatchFilter . . . . . . . . . . . . . . . . . 248 17.2.2.2 SimpleStreamFilter …………….250 17.2.2.3 Internals ………………….252
17.2.3 Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . 252
17.2.4 Packages. . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
17.2.5Revisions ………………………252 17.2.6Testing ……………………….253
8
CONTENTS
17.3
17.2.6.1 Option handling . . . . . . . . . . . . . . . . . . 253 17.2.6.2 GenericObjectEditor…………….253 17.2.6.3 Sourcecode…………………253 17.2.6.4 Unittests………………….253
Writing other algorithms . . . . . . . . . . . . . . . . . . . . . . . 254 17.3.1 Clusterers . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 17.3.2 Attribute selection . . . . . . . . . . . . . . . . . . . . . . 256 17.3.3 Associators . . . . . . . . . . . . . . . . . . . . . . . . . . 258
17.4ExtendingtheExplorer …………………..260 17.4.1 Adding tabs . . . . . . . . . . . . . . . . . . . . . . . . . . 260 17.4.1.1 Requirements………………..260 17.4.1.2 Examples………………….260 17.4.2Addingvisualizationplugins …………….268
18 Technical documentation 271
18.1 ANT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 18.1.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 18.1.2 Weka and ANT . . . . . . . . . . . . . . . . . . . . . . . . 271
18.2 CLASSPATH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 18.2.1SettingtheCLASSPATH ………………272 18.2.2 RunWeka.bat . . . . . . . . . . . . . . . . . . . . . . . . . 273 18.2.3 java -jar . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
18.3 Subversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 18.3.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 18.3.2 Source code . . . . . . . . . . . . . . . . . . . . . . . . . . 274 18.3.3JUnit ………………………..275 18.3.4 Specific version . . . . . . . . . . . . . . . . . . . . . . . . 275 18.3.5 Clients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
18.4 GenericObjectEditor . . . . . . . . . . . . . . . . . . . . . . . . . 276 18.4.1Introduction …………………….276 18.4.2 File Structure . . . . . . . . . . . . . . . . . . . . . . . . . 277 18.4.3 Exclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 18.4.4 Class Discovery . . . . . . . . . . . . . . . . . . . . . . . . 279 18.4.5 Multiple Class Hierarchies . . . . . . . . . . . . . . . . . . 279 18.4.6 Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . 281
18.5 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 18.5.1Precedence ……………………..282 18.5.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
18.6 XML . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 18.6.1 Command Line . . . . . . . . . . . . . . . . . . . . . . . . 283 18.6.2 Serialization of Experiments . . . . . . . . . . . . . . . . . 286 18.6.3 Serialization of Classifiers . . . . . . . . . . . . . . . . . . 287 18.6.4 Bayesian Networks . . . . . . . . . . . . . . . . . . . . . . 288 18.6.5XRFFfiles ……………………..288
19 Other resources 289
19.1Mailinglist …………………………289 19.2 Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 19.2.1Wekadownloadproblems ………………289 19.2.2 OutOfMemoryException . . . . . . . . . . . . . . . . . . . 289
CONTENTS 9
19.2.2.1 Windows ………………….290 19.2.3 Mac OSX . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 19.2.4StackOverflowError …………………290 19.2.5just-in-time(JIT)compiler ……………..291 19.2.6 CSV file conversion . . . . . . . . . . . . . . . . . . . . . . 291 19.2.7 ARFF file doesn’t load . . . . . . . . . . . . . . . . . . . . 291 19.2.8 Spaces in labels of ARFF files . . . . . . . . . . . . . . . . 291 19.2.9CLASSPATHproblems ……………….291 19.2.10Instance ID . . . . . . . . . . . . . . . . . . . . . . . . . . 292
19.2.10.1 Adding the ID . . . . . . . . . . . . . . . . . . . 292
19.2.10.2 Removing the ID . . . . . . . . . . . . . . . . . . 292 19.2.11Visualization …………………….293 19.2.12Memory consumption and Garbage collector . . . . . . . 293 19.2.13GUIChooser starts but not Experimenter or Explorer ..293 19.2.14KnowledgeFlow toolbars are empty . . . . . . . . . . . . . 294 19.2.15Links ………………………..294
Bibliography 296
10 CONTENTS
Part I
The Command-line
11
Chapter 1
A command-line primer
1.1 Introduction
While for initial experiments the included graphical user interface is quite suf- ficient, for in-depth usage the command line interface is recommended, because it offers some functionality which is not available via the GUI – and uses far less memory. Should you get Out of Memory errors, increase the maximum heap size for your java engine, usually via -Xmx1024M or -Xmx1024m for 1GB – the default setting of 16 to 64MB is usually too small. If you get errors that classes are not found, check your CLASSPATH: does it include weka.jar? You can explicitly set CLASSPATH via the -cp command line option as well.
We will begin by describing basic concepts and ideas. Then, we will describe the weka.filters package, which is used to transform input data, e.g. for preprocessing, transformation, feature generation and so on.
Then we will focus on the machine learning algorithms themselves. These are called Classifiers in WEKA. We will restrict ourselves to common settings for all classifiers and shortly note representatives for all main approaches in machine learning.
Afterwards, practical examples are given.
Finally, in the doc directory of WEKA you find a documentation of all java classes within WEKA. Prepare to use it since this overview is not intended to be complete. If you want to know exactly what is going on, take a look at the mostly well-documented source code, which can be found in weka-src.jar and can be extracted via the jar utility from the Java Development Kit (or any archive program that can handle ZIP files).
13
14 CHAPTER 1. A COMMAND-LINE PRIMER
1.2 Basic concepts
1.2.1 Dataset
A set of data items, the dataset, is a very basic concept of machine learning. A dataset is roughly equivalent to a two-dimensional spreadsheet or database table. In WEKA, it is implemented by the weka.core.Instances class. A dataset is a collection of examples, each one of class weka.core.Instance. Each Instance consists of a number of attributes, any of which can be nominal (= one of a predefined list of values), numeric (= a real or integer number) or a string (= an arbitrary long list of characters, enclosed in ”double quotes”). Additional types are date and relational, which are not covered here but in the ARFF chapter. The external representation of an Instances class is an ARFF file, which consists of a header describing the attribute types and the data as comma-separated list. Here is a short, commented example. A complete description of the ARFF file format can be found here.
% This is a toy example, the UCI weather dataset.
% Any relation to real weather is purely coincidental.
@relation golfWeatherMichigan_1988/02/10_14days
@attribute outlook {sunny, overcast, rainy} @attribute windy {TRUE, FALSE}
@attribute temperature real @attribute humidity real
@attribute play {yes, no}
Comment lines at the beginning of the dataset should give an in- dication of its source, context and meaning.
Here we state the internal name of the dataset. Try to be as com- prehensive as possible.
Here we define two nominal at- tributes, outlook and windy. The former has three values: sunny, overcast and rainy ; the latter two: TRUE and FALSE. Nom- inal values with special charac- ters, commas or spaces are en- closed in ’single quotes’.
These lines define two numeric attributes. Instead of real, inte- ger or numeric can also be used. While double floating point val- ues are stored internally, only seven decimal digits are usually processed.
The last attribute is the default target or class variable used for prediction. In our case it is a nominal attribute with two val- ues, making this a binary classi- fication problem.
1.2. BASIC CONCEPTS
15
@data sunny,FALSE,85,85,no sunny,TRUE,80,90,no overcast,FALSE,83,86,yes rainy,FALSE,70,96,yes rainy,FALSE,68,80,yes
The rest of the dataset consists of the token @data, followed by comma-separated values for the attributes – one line per exam- ple. In our case there are five ex- amples.
In our example, we have not mentioned the attribute type string, which defines ”double quoted” string attributes for text mining. In recent WEKA versions, date/time attribute types are also supported.
By default, the last attribute is considered the class/target variable, i.e. the attribute which should be predicted as a function of all other attributes. If this is not the case, specify the target variable via -c. The attribute numbers are one-based indices, i.e. -c 1 specifies the first attribute.
Some basic statistics and validation of given ARFF files can be obtained via the main() routine of weka.core.Instances:
java weka.core.Instances data/soybean.arff
weka.core offers some other useful routines, e.g. converters.C45Loader and converters.CSVLoader, which can be used to import C45 datasets and comma/tab- separated datasets respectively, e.g.:
java weka.core.converters.CSVLoader data.csv > data.arff java weka.core.converters.C45Loader c45_filestem > data.arff
16 CHAPTER 1. A COMMAND-LINE PRIMER 1.2.2 Classifier
Any learning algorithm in WEKA is derived from the abstract weka.classifiers.Classifier class. Surprisingly little is needed for a basic classifier: a routine which gen-
erates a classifier model from a training dataset (= buildClassifier) and
another routine which evaluates the generated model on an unseen test dataset
(= classifyInstance), or generates a probability distribution for all classes (= distributionForInstance).
A classifier model is an arbitrary complex mapping from all-but-one dataset attributes to the class attribute. The specific form and creation of this map- ping, or model, differs from classifier to classifier. For example, ZeroR’s (= weka.classifiers.rules.ZeroR) model just consists of a single value: the most common class, or the median of all numeric values in case of predicting a numeric value (= regression learning). ZeroR is a trivial classifier, but it gives a lower bound on the performance of a given dataset which should be significantly improved by more complex classifiers. As such it is a reasonable test on how well the class can be predicted without considering the other attributes.
Later, we will explain how to interpret the output from classifiers in detail – for now just focus on the Correctly Classified Instances in the section Stratified cross-validation and notice how it improves from ZeroR to J48:
java weka.classifiers.rules.ZeroR -t weather.arff java weka.classifiers.trees.J48 -t weather.arff
There are various approaches to determine the performance of classifiers. The performance can most simply be measured by counting the proportion of cor- rectly predicted examples in an unseen test dataset. This value is the accuracy, which is also 1-ErrorRate. Both terms are used in literature.
The simplest case is using a training set and a test set which are mutually independent. This is referred to as hold-out estimate. To estimate variance in these performance estimates, hold-out estimates may be computed by repeatedly resampling the same dataset – i.e. randomly reordering it and then splitting it into training and test sets with a specific proportion of the examples, collecting all estimates on test data and computing average and standard deviation of accuracy.
A more elaborate method is cross-validation. Here, a number of folds n is specified. The dataset is randomly reordered and then split into n folds of equal size. In each iteration, one fold is used for testing and the other n-1 folds are used for training the classifier. The test results are collected and averaged over all folds. This gives the cross-validation estimate of the accuracy. The folds can be purely random or slightly modified to create the same class distributions in each fold as in the complete dataset. In the latter case the cross-validation is called stratified. Leave-one-out (= loo) cross-validation signifies that n is equal to the number of examples. Out of necessity, loo cv has to be non-stratified, i.e. the class distributions in the test set are not related to those in the training data. Therefore loo cv tends to give less reliable results. However it is still quite useful in dealing with small datasets since it utilizes the greatest amount of training data from the dataset.
1.2. BASIC CONCEPTS 17 1.2.3 weka.filters
The weka.filters package is concerned with classes that transform datasets – by removing or adding attributes, resampling the dataset, removing examples and so on. This package offers useful support for data preprocessing, which is an important step in machine learning.
All filters offer the options -i for specifying the input dataset, and -o for specifying the output dataset. If any of these parameters is not given, standard input and/or standard output will be read from/written to. Other parameters are specific to each filter and can be found out via -h, as with any other class. The weka.filters package is organized into supervised and unsupervised filtering, both of which are again subdivided into instance and attribute filtering. We will discuss each of the four subsections separately.
weka.filters.supervised
Classes below weka.filters.supervised in the class hierarchy are for super- vised filtering, i.e., taking advantage of the class information. A class must be assigned via -c, for WEKA default behaviour use -c last.
weka.filters.supervised.attribute
Discretize is used to discretize numeric attributes into nominal ones, based on the class information, via Fayyad & Irani’s MDL method, or optionally with Kononeko’s MDL method. At least some learning schemes or classifiers can only process nominal data, e.g. weka.classifiers.rules.Prism; in some cases discretization may also reduce learning time.
java weka.filters.supervised.attribute.Discretize -i data/iris.arff \ -o iris-nom.arff -c last
java weka.filters.supervised.attribute.Discretize -i data/cpu.arff \ -o cpu-classvendor-nom.arff -c first
NominalToBinary encodes all nominal attributes into binary (two-valued) at- tributes, which can be used to transform the dataset into a purely numeric representation, e.g. for visualization via multi-dimensional scaling.
java weka.filters.supervised.attribute.NominalToBinary \
-i data/contact-lenses.arff -o contact-lenses-bin.arff -c last
Keep in mind that most classifiers in WEKA utilize transformation filters in- ternally, e.g. Logistic and SMO, so you will usually not have to use these filters explicity. However, if you plan to run a lot of experiments, pre-applying the filters yourself may improve runtime performance.
weka.filters.supervised.instance
Resample creates a stratified subsample of the given dataset. This means that overall class distributions are approximately retained within the sample. A bias towards uniform class distribution can be specified via -B.
java weka.filters.supervised.instance.Resample -i data/soybean.arff \ -o soybean-5%.arff -c last -Z 5
java weka.filters.supervised.instance.Resample -i data/soybean.arff \ -o soybean-uniform-5%.arff -c last -Z 5 -B 1
18 CHAPTER 1. A COMMAND-LINE PRIMER
StratifiedRemoveFolds creates stratified cross-validation folds of the given dataset. This means that by default the class distributions are approximately retained within each fold. The following example splits soybean.arff into strat- ified training and test datasets, the latter consisting of 25% (= 1/4) of the data.
java weka.filters.supervised.instance.StratifiedRemoveFolds \ -i data/soybean.arff -o soybean-train.arff \
-c last -N 4 -F 1 -V
java weka.filters.supervised.instance.StratifiedRemoveFolds \ -i data/soybean.arff -o soybean-test.arff \
-c last -N 4 -F 1
weka.filters.unsupervised
Classes below weka.filters.unsupervised in the class hierarchy are for unsu- pervised filtering, e.g. the non-stratified version of Resample. A class attribute should not be assigned here.
weka.filters.unsupervised.attribute
StringToWordVector transforms string attributes into word vectors, i.e. creat- ing one attribute for each word which either encodes presence or word count (= -C) within the string. -W can be used to set an approximate limit on the number of words. When a class is assigned, the limit applies to each class separately. This filter is useful for text mining.
Obfuscate renames the dataset name, all attribute names and nominal attribute values. This is intended for exchanging sensitive datasets without giving away restricted information.
Remove is intended for explicit deletion of attributes from a dataset, e.g. for removing attributes of the iris dataset:
java weka.filters.unsupervised.attribute.Remove -R 1-2 \ -i data/iris.arff -o iris-simplified.arff
java weka.filters.unsupervised.attribute.Remove -V -R 3-last \ -i data/iris.arff -o iris-simplified.arff
weka.filters.unsupervised.instance
Resample creates a non-stratified subsample of the given dataset, i.e. random sampling without regard to the class information. Otherwise it is equivalent to its supervised variant.
java weka.filters.unsupervised.instance.Resample -i data/soybean.arff \ -o soybean-5%.arff -Z 5
RemoveFoldscreates cross-validation folds of the given dataset. The class distri- butions are not retained. The following example splits soybean.arff into training and test datasets, the latter consisting of 25% (= 1/4) of the data.
java weka.filters.unsupervised.instance.RemoveFolds -i data/soybean.arff \ -o soybean-train.arff -c last -N 4 -F 1 -V
java weka.filters.unsupervised.instance.RemoveFolds -i data/soybean.arff \ -o soybean-test.arff -c last -N 4 -F 1
RemoveWithValues filters instances according to the value of an attribute.
java weka.filters.unsupervised.instance.RemoveWithValues -i data/soybean.arff \ -o soybean-without_herbicide_injury.arff -V -C last -L 19
1.2. BASIC CONCEPTS 19 1.2.4 weka.classifiers
Classifiers are at the core of WEKA. There are a lot of common options for classifiers, most of which are related to evaluation purposes. We will focus on the most important ones. All others including classifier-specific parameters can be found via -h, as usual.
-t specifies the training file (ARFF format)
specifies the test file in (ARFF format). If this parameter is miss- ing, a crossvalidation will be performed (default: ten-fold cv)
This parameter determines the number of folds for the cross- validation. A cv will only be performed if -T is missing.
As we already know from the weka.filters section, this parameter sets the class variable with a one-based index.
The model after training can be saved via this parameter. Each
classifier has a different binary format for the model, so it can -d only be read back by the exact same classifier on a compatible dataset. Only the model on the training set is saved, not the
multiple models generated via cross-validation.
Loads a previously saved model, usually for testing on new, pre- -l viously unseen data. In that case, a compatible test file should be
specified, i.e. the same attributes in the same order.
If a test file is specified, this parameter shows you the predictions and one attribute (0 for none) for all test instances.
A more detailed performance description via precision, recall, true- and false positive rate is additionally output with this pa- rameter. All these values can also be computed from the confusion matrix.
This parameter switches the human-readable output of the model description off. In case of support vector machines or NaiveBayes, this makes some sense unless you want to parse and visualize a lot of information.
We now give a short list of selected classifiers in WEKA. Other classifiers below weka.classifiers may also be used. This is more easy to see in the Explorer GUI.
• trees.J48 A clone of the C4.5 decision tree learner
• bayes.NaiveBayes A Naive Bayesian learner. -K switches on kernel den-
sity estimation for numerical attributes which often improves performance.
• meta.ClassificationViaRegression-W functions.LinearRegression Multi-response linear regression.
• functions.Logistic Logistic Regression.
-T -x -c
-p #
-i
-o
20
CHAPTER 1. A COMMAND-LINE PRIMER
•
•
•
•
functions.SMOSupport Vector Machine (linear, polynomial and RBF ker- nel) with Sequential Minimal Optimization Algorithm due to [4]. Defaults to SVM with linear kernel, -E 5 -C 10 gives an SVM with polynomial kernel of degree 5 and lambda of 10.
lazy.KStar Instance-Based learner. -E sets the blend entropy automati- cally, which is usually preferable.
lazy.IBk Instance-Based learner with fixed neighborhood. -K sets the number of neighbors to use. IB1 is equivalent to IBk -K 1
rules.JRip A clone of the RIPPER rule learner.
Based on a simple example, we will now explain the output of a typical classifier, weka.classifiers.trees.J48. Consider the following call from the command line, or start the WEKA explorer and train J48 on weather.arff :
java weka.classifiers.trees.J48 -t data/weather.arff -i
J48 pruned tree ——————
outlook = sunny
| humidity <= 75: yes (2.0)
| humidity > 75: no (3.0) outlook = overcast: yes (4.0) outlook = rainy
| windy = TRUE: no (2.0)
| windy = FALSE: yes (3.0)
Number of Leaves : 5 Size of the tree : 8
Time taken to build model: 0.05 seconds
Time taken to test model on training data: 0 seconds
The first part, unless you specify -o, is a human-readable form of the training set model. In this case, it is a decision tree. out- look is at the root of the tree and determines the first decision. In case it is overcast, we’ll al- ways play golf. The numbers in (parentheses) at the end of each leaf tell us the number of exam- ples in this leaf. If one or more leaves were not pure (= all of the same class), the number of mis- classified examples would also be given, after a /slash/
As you can see, a decision tree learns quite fast and is evalu- ated even faster. E.g. for a lazy learner, testing would take far longer than training.
1.2. BASIC CONCEPTS
21
== Error on training data ===
Correctly Classified Instance 14 Incorrectly Classified Instances 0 0% Kappa statistic 1
Mean absolute error 0
Root mean squared error 0
Relative absolute error 0 % Root relative squared error 0 % Total Number of Instances 14
=== Detailed Accuracy By Class ===
TP Rate FP Rate Precision Recall F-Measure 1 0 1 1 1
1 0 1 1 1
===
ab <-- classified as
9 0 | 0 5 |
a = yes b = no
===
Stratified cross-validation ===
Confusion Matrix ===
Correctly Classified Instances 9 Incorrectly Classified Instances 5 Kappa statistic 0.186 Mean absolute error 0.2857 Root mean squared error 0.4818 Relative absolute error 60
Root relative squared error 97.6586 Total Number of Instances 14
=== Detailed Accuracy By Class ===
64.2857 % 35.7143 %
% %
TP Rate FP Rate Precision Recall F-Measure Class 0.778 0.6 0.7 0.778 0.737 yes
0.4
===
ab 7 2 3 2
0.222 0.5 0.4 0.444 no
Confusion Matrix ===
<-- classified as | a = yes
| b = no
100 %
The confusion matrix is more commonly named contingency table. In our case we have two classes, and therefore a 2x2 confusion matrix, the matrix could be arbitrarily large. The number of correctly classified instances is the sum of diagonals in the matrix; all others are incorrectly classified (class ”a” gets misclassified as ”b” exactly twice, and class ”b” gets misclassified as ”a” three times).
The True Positive (TP) rate is the proportion of examples which were clas- sified as class x, among all examples which truly have class x, i.e. how much part of the class was captured. It is equivalent to Recall. In the confusion ma- trix, this is the diagonal element divided by the sum over the relevant row, i.e. 7/(7+2)=0.778 for class yes and 2/(3+2)=0.4 for class no in our example.
The False Positive (FP) rate is the proportion of examples which were classi- fied as class x, but belong to a different class, among all examples which are not of class x. In the matrix, this is the column sum of class x minus the diagonal element, divided by the rows sums of all other classes; i.e. 3/5=0.6 for class yes and 2/9=0.222 for class no.
The Precision is the proportion of the examples which truly have class x
Class yes no
This is quite boring: our clas- sifier is perfect, at least on the training data – all instances were classified correctly and all errors are zero. As is usually the case, the training set accuracy is too optimistic. The detailed accu- racy by class, which is output via -i, and the confusion matrix is similarily trivial.
The stratified cv paints a more realistic picture. The accuracy is around 64%. The kappa statis- tic measures the agreement of prediction with the true class – 1.0 signifies complete agreement. The following error values are not very meaningful for classifi- cation tasks, however for regres- sion tasks e.g. the root of the mean squared error per exam- ple would be a reasonable cri- terion. We will discuss the re- lation between confusion matrix and other measures in the text.
22 CHAPTER 1. A COMMAND-LINE PRIMER
among all those which were classified as class x. In the matrix, this is the diagonal element divided by the sum over the relevant column, i.e. 7/(7+3)=0.7 for class yes and 2/(2+2)=0.5 for class no.
The F-Measure is simply 2*Precision*Recall/(Precision+Recall), a combined measure for precision and recall.
These measures are useful for comparing classifiers. However, if more de- tailed information about the classifier’s predictions are necessary, -p # out- puts just the predictions for each test instance, along with a range of one- based attribute ids (0 for none). Let’s look at the following example. We shall assume soybean-train.arff and soybean-test.arff have been constructed via weka.filters.supervised.instance.StratifiedRemoveFolds as in a previous example.
java weka.classifiers.bayes.NaiveBayes -K -t soybean-train.arff \ -T soybean-test.arff -p 0
The values in each line are sep-
arated by a single space. The 0 diaporthe-stem-canker 0.9999672587892333 diaporthe-stem-canker fields are the zero-based test in-
1 diaporthe-stem-canker 0.9999992614503429 diaporthe-stem-canker
2 diaporthe-stem-canker 0.999998948559035 diaporthe-stem-canker stance id, followed by the pre- 3 diaporthe-stem-canker 0.9999998441238833 diaporthe-stem-cankerdicted class value, the confi- 4 diaporthe-stem-canker 0.9999989997681132 diaporthe-stem-canker
5 rhizoctonia-root-rot 0.9999999395928124 rhizoctonia-root-rot
6 rhizoctonia-root-rot 0.999998912860593 rhizoctonia-root-rot
7 rhizoctonia-root-rot 0.9999994386283236 rhizoctonia-root-rot
...
32 phyllosticta-leaf-spot 0.7789710144361445 brown-spot ...
39 alternarialeaf-spot 0.6403333824349896 brown-spot ...
44 phyllosticta-leaf-spot 0.893568420641914 brown-spot ...
46 alternarialeaf-spot 0.5788190397739439 brown-spot ...
73 brown-spot 0.4943768155314637 alternarialeaf-spot ...
dence for the prediction (esti- mated probability of predicted class), and the true class. All these are correctly classified, so let’s look at a few erroneous ones.
In each of these cases, a misclas- sification occurred, mostly be- tween classes alternarialeaf-spot and brown-spot. The confidences seem to be lower than for correct classification, so for a real-life ap- plication it may make sense to output don’t know below a cer- tain threshold. WEKA also out- puts a trailing newline.
If we had chosen a range of attributes via -p, e.g. -p first-last, the mentioned attributes would have been output afterwards as comma-separated values, in (parentheses). However, the zero-based instance id in the first column offers a safer way to determine the test instances.
If we had saved the output of -p in soybean-test.preds, the following call would compute the number of correctly classified instances:
cat soybean-test.preds | awk ’$2==$4&&$0!=""’ | wc -l
Dividing by the number of instances in the test set, i.e. wc -l < soybean-test.preds minus one (= trailing newline), we get the training set accuracy.
1.3. EXAMPLES 23 1.3 Examples
Usually, if you evaluate a classifier for a longer experiment, you will do something like this (for csh):
java -Xmx1024m weka.classifiers.trees.J48 -t data.arff -i -k \ -d J48-data.model >&! J48-data.out &
The -Xmx1024m parameter for maximum heap size ensures your task will get enough memory. There is no overhead involved, it just leaves more room for the heap to grow. -i and -k gives you some additional information, which may be useful, e.g. precision and recall for all classes. In case your model performs well, it makes sense to save it via -d – you can always delete it later! The implicit cross-validation gives a more reasonable estimate of the expected accuracy on unseen data than the training set accuracy. The output both of standard error and output should be redirected, so you get both errors and the normal output of your classifier. The last & starts the task in the background. Keep an eye on your task via top and if you notice the hard disk works hard all the time (for linux), this probably means your task needs too much memory and will not finish in time for the exam. In that case, switch to a faster classifier or use filters, e.g. for Resample to reduce the size of your dataset or StratifiedRemoveFolds to create training and test sets – for most classifiers, training takes more time than testing.
So, now you have run a lot of experiments – which classifier is best? Try
cat *.out | grep -A 3 “Stratified” | grep “^Correctly”
…this should give you all cross-validated accuracies. If the cross-validated ac- curacy is roughly the same as the training set accuracy, this indicates that your classifiers is presumably not overfitting the training set.
Now you have found the best classifier. To apply it on a new dataset, use e.g.
java weka.classifiers.trees.J48 -l J48-data.model -T new-data.arff
You will have to use the same classifier to load the model, but you need not set any options. Just add the new test file via -T. If you want, -p first-last will output all test instances with classifications and confidence, followed by all attribute values, so you can look at each error separately.
The following more complex csh script creates datasets for learning curves, i.e. creating a 75% training set and 25% test set from a given dataset, then successively reducing the test set by factor 1.2 (83%), until it is also 25% in size. All this is repeated thirty times, with different random reorderings (-S) and the results are written to different directories. The Experimenter GUI in WEKA can be used to design and run similar experiments.
#!/bin/csh foreach f ($*)
set run=1
while ( $run <= 30 )
mkdir $run >&! /dev/null
java weka.filters.supervised.instance.StratifiedRemoveFolds -N 4 -F 1 -S $run -c last -i ../$f -o $run/t_$f java weka.filters.supervised.instance.StratifiedRemoveFolds -N 4 -F 1 -S $run -V -c last -i ../$f -o $run/t0$f foreach nr (0 1 2 3 4 5)
set nrp1=$nr @ nrp1++
24 CHAPTER 1. A COMMAND-LINE PRIMER
java weka.filters.supervised.instance.Resample -S 0 -Z 83 -c last -i $run/t$nr$f -o $run/t$nrp1$f
end
echo Run $run of $f done.
@ run++ end
end
If meta classifiers are used, i.e. classifiers whose options include classi- fier specifications – for example, StackingC or ClassificationViaRegression, care must be taken not to mix the parameters. E.g.:
java weka.classifiers.meta.ClassificationViaRegression \ -W weka.classifiers.functions.LinearRegression -S 1 \ -t data/iris.arff -x 2
gives us an illegal options exception for -S 1. This parameter is meant for LinearRegression, not for ClassificationViaRegression, but WEKA does not know this by itself. One way to clarify this situation is to enclose the classifier specification, including all parameters, in ”double” quotes, like this:
java weka.classifiers.meta.ClassificationViaRegression \ -W “weka.classifiers.functions.LinearRegression -S 1” \ -t data/iris.arff -x 2
However this does not always work, depending on how the option handling was implemented in the top-level classifier. While for Stacking this approach would work quite well, for ClassificationViaRegression it does not. We get the dubious error message that the class weka.classifiers.functions.LinearRegression -S 1 cannot be found. Fortunately, there is another approach: All parameters given after — are processed by the first sub-classifier; another — lets us specify parameters for the second sub-classifier and so on.
java weka.classifiers.meta.ClassificationViaRegression \ -W weka.classifiers.functions.LinearRegression \
-t data/iris.arff -x 2 — -S 1
In some cases, both approaches have to be mixed, for example:
java weka.classifiers.meta.Stacking -B “weka.classifiers.lazy.IBk -K 10” \
-M “weka.classifiers.meta.ClassificationViaRegression -W weka.classifiers.functions.LinearRegression — -S 1” \ -t data/iris.arff -x 2
Notice that while ClassificationViaRegression honors the — parameter, Stacking itself does not. Sadly the option handling for sub-classifier specifi- cations is not yet completely unified within WEKA, but hopefully one or the other approach mentioned here will work.
Part II
The Graphical User Interface
25
Chapter 2 Launching WEKA
The Weka GUI Chooser (class weka.gui.GUIChooser) provides a starting point for launching Weka’s main GUI applications and supporting tools. If one prefers a MDI (“multiple document interface”) appearance, then this is provided by an alternative launcher called “Main” (class weka.gui.Main).
The GUI Chooser consists of four buttons—one for each of the four major Weka applications—and four menus.
The buttons can be used to start the following applications:
• Explorer An environment for exploring data with WEKA (the rest of this documentation deals with this application in more detail).
• Experimenter An environment for performing experiments and conduct- ing statistical tests between learning schemes.
• KnowledgeFlow This environment supports essentially the same func- tions as the Explorer but with a drag-and-drop interface. One advantage is that it supports incremental learning.
• SimpleCLI Provides a simple command-line interface that allows direct execution of WEKA commands for operating systems that do not provide their own command line interface.
The menu consists of four sections: 1. Program
27
28 CHAPTER 2. LAUNCHING WEKA
• LogWindow Opens a log window that captures all that is printed to stdout or stderr. Useful for environments like MS Windows, where WEKA is normally not started from a terminal.
• Exit Closes WEKA.
2. Tools Other useful applications.
• ArffViewer An MDI application for viewing ARFF files in spread- sheet format.
• SqlViewer Represents an SQL worksheet, for querying databases via JDBC.
• Bayes net editor An application for editing, visualizing and learn- ing Bayes nets.
3. Visualization Ways of visualizing data with WEKA.
• Plot For plotting a 2D plot of a dataset.
• ROC Displays a previously saved ROC curve.
• TreeVisualizer For displaying directed graphs, e.g., a decision tree.
• GraphVisualizer Visualizes XML BIF or DOT format graphs, e.g., for Bayesian networks.
• BoundaryVisualizer Allows the visualization of classifier decision boundaries in two dimensions.
4. Help Online resources for WEKA can be found here.
• Weka homepage Opens a browser window with WEKA’s home- page.
• HOWTOs, code snippets, etc. The general WekaWiki [2], con- taining lots of examples and HOWTOs around the development and use of WEKA.
• WekaonSourceforgeWEKA’sprojecthomepageonSourceforge.net.
• SystemInfo Lists some internals about the Java/WEKA environ-
ment, e.g., the CLASSPATH.
To make it easy for the user to add new functionality to the menu with- out having to modify the code of WEKA itself, the GUI now offers a plugin mechanism for such add-ons. Due to the inherent dynamic class discovery, plu- gins only need to implement the weka.gui.MainMenuExtension interface and
WEKA notified of the package they reside in to be displayed in the menu un- der “Extensions” (this extra menu appears automatically as soon as extensions are discovered). More details can be found in the Wiki article “Extensions for Weka’s main GUI” [6].
If you launch WEKA from a terminal window, some text begins scrolling in the terminal. Ignore this text unless something goes wrong, in which case it can help in tracking down the cause (the LogWindow from the Program menu displays that information as well).
This User Manual focuses on using the Explorer but does not explain the individual data preprocessing tools and learning algorithms in WEKA. For more information on the various filters and learning methods in WEKA, see the book Data Mining [1].
29
30 CHAPTER 2. LAUNCHING WEKA
Chapter 3 Simple CLI
The Simple CLI provides full access to all Weka classes, i.e., classifiers, filters, clusterers, etc., but without the hassle of the CLASSPATH (it facilitates the one, with which Weka was started).
It offers a simple Weka shell with separated commandline and output.
3.1 Commands
The following commands are available in the Simple CLI:
• java
invokes a java class with the given arguments (if any)
• break
stops the current thread, e.g., a running classifier, in a friendly manner
31
32
CHAPTER 3.
SIMPLE CLI
• • • •
3.2
kill
stops the current thread in an unfriendly fashion
cls
clears the output area
exit
exits the Simple CLI
help [
provides an overview of the available commands if without a command name as argument, otherwise more help on the specified command
Invocation
In order to invoke a Weka class, one has only to prefix the class with ”java”. This command tells the Simple CLI to load a class and execute it with any given parameters. E.g., the J48 classifier can be invoked on the iris dataset with the following command:
java weka.classifiers.trees.J48 -t c:/temp/iris.arff
This results in the following output:
3.3 Command redirection
Starting with this version of Weka one can perform a basic redirection:
java weka.classifiers.trees.J48 -t test.arff > j48.txt
Note: the > must be preceded and followed by a space, otherwise it is not recognized as redirection, but part of another parameter.
3.4. COMMAND COMPLETION 33 3.4 Command completion
Commands starting with java support completion for classnames and filenames via Tab (Alt+BackSpace deletes parts of the command again). In case that there are several matches, Weka lists all possible matches.
• package name completion
java weka.cl
results in the following output of possible matches of package names:
Possible matches: weka.classifiers weka.clusterers
• classname completion
java weka.classifiers.meta.A
Possible matches:
weka.classifiers.meta.AdaBoostM1 weka.classifiers.meta.AdditiveRegression weka.classifiers.meta.AttributeSelectedClassifier
• filename completion
In order for Weka to determine whether a the string under the cursor
is a classname or a filename, filenames need to be absolute (Unix/Linx: /some/path/file; Windows: C:\Some\Path\file) or relative and starting
with a dot (Unix/Linux: ./some/other/path/file; Windows: .\Some\Other\Path\file).
34 CHAPTER 3. SIMPLE CLI
Chapter 4 Explorer
4.1 The user interface 4.1.1 Section Tabs
At the very top of the window, just below the title bar, is a row of tabs. When the Explorer is first started only the first tab is active; the others are greyed out. This is because it is necessary to open (and potentially pre-process) a data set before starting to explore the data.
The tabs are as follows:
1. Preprocess. Choose and modify the data being acted on.
2. Classify. Train and test learning schemes that classify or perform regres- sion.
3. Cluster. Learn clusters for the data.
4. Associate. Learn association rules for the data.
5. Select attributes. Select the most relevant attributes in the data.
6. Visualize. View an interactive 2D plot of the data.
Once the tabs are active, clicking on them flicks between different screens, on which the respective actions can be performed. The bottom area of the window (including the status box, the log button, and the Weka bird) stays visible regardless of which section you are in.
The Explorer can be easily extended with custom tabs. The Wiki article “Adding tabs in the Explorer” [7] explains this in detail.
4.1.2 Status Box
The status box appears at the very bottom of the window. It displays messages that keep you informed about what’s going on. For example, if the Explorer is busy loading a file, the status box will say that.
TIP—right-clicking the mouse anywhere inside the status box brings up a little menu. The menu gives two options:
35
36
CHAPTER 4. EXPLORER
1. Memory information. Display in the log box the amount of memory available to WEKA.
2. Run garbage collector. Force the Java garbage collector to search for memory that is no longer needed and free it up, allowing more memory for new tasks. Note that the garbage collector is constantly running as a background task anyway.
4.1.3 Log Button
Clicking on this button brings up a separate window containing a scrollable text field. Each line of text is stamped with the time it was entered into the log. As you perform actions in WEKA, the log keeps a record of what has happened. For people using the command line or the SimpleCLI, the log now also contains the full setup strings for classification, clustering, attribute selection, etc., so that it is possible to copy/paste them elsewhere. Options for dataset(s) and, if applicable, the class attribute still have to be provided by the user (e.g., -t for classifiers or -i and -o for filters).
4.1.4 WEKA Status Icon
To the right of the status box is the WEKA status icon. When no processes are running, the bird sits down and takes a nap. The number beside the × symbol gives the number of concurrent processes running. When the system is idle it is zero, but it increases as the number of processes increases. When any process is started, the bird gets up and starts moving around. If it’s standing but stops moving for a long time, it’s sick: something has gone wrong! In that case you should restart the WEKA Explorer.
4.1.5 Graphical output
Most graphical displays in WEKA, e.g., the GraphVisualizer or the TreeVisu- alizer, support saving the output to a file. A dialog for saving the output can be brought up with Alt+Shift+left-click. Supported formats are currently Win- dows Bitmap, JPEG, PNG and EPS (encapsulated Postscript). The dialog also allows you to specify the dimensions of the generated image.
4.2. PREPROCESSING 37 4.2 Preprocessing
4.2.1 Loading Data
The first four buttons at the top of the preprocess section enable you to load data into WEKA:
1. Open file…. Brings up a dialog box allowing you to browse for the data file on the local file system.
2. Open URL…. Asks for a Uniform Resource Locator address for where the data is stored.
3. Open DB…. Reads data from a database. (Note that to make this work you might have to edit the file in weka/experiment/DatabaseUtils.props.)
4. Generate…. Enables you to generate artificial data from a variety of DataGenerators.
Using the Open file… button you can read files in a variety of formats: WEKA’s ARFF format, CSV format, C4.5 format, or serialized Instances for- mat. ARFF files typically have a .arff extension, CSV files a .csv extension, C4.5 files a .data and .names extension, and serialized Instances objects a .bsi extension.
NB: This list of formats can be extended by adding custom file converters to the weka.core.converters package.
4.2.2 The Current Relation
Once some data has been loaded, the Preprocess panel shows a variety of in- formation. The Current relation box (the “current relation” is the currently loaded data, which can be interpreted as a single relational table in database terminology) has three entries:
38 1.
2. 3.
CHAPTER 4. EXPLORER
Relation. The name of the relation, as given in the file it was loaded from. Filters (described below) modify the name of a relation.
Instances. The number of instances (data points/records) in the data. Attributes. The number of attributes (features) in the data.
4.2.3 Working With Attributes
Below the Current relation box is a box titled Attributes. There are four buttons, and beneath them is a list of the attributes in the current relation. The list has three columns:
1. No.. A number that identifies the attribute in the order they are specified in the data file.
2. Selection tick boxes. These allow you select which attributes are present in the relation.
3. Name. The name of the attribute, as it was declared in the data file.
When you click on different rows in the list of attributes, the fields change in the box to the right titled Selected attribute. This box displays the char- acteristics of the currently highlighted attribute in the list:
1. Name. The name of the attribute, the same as that given in the attribute list.
2. Type. The type of attribute, most commonly Nominal or Numeric.
3. Missing. The number (and percentage) of instances in the data for which
this attribute is missing (unspecified).
4. Distinct. The number of different values that the data contains for this attribute.
5. Unique. The number (and percentage) of instances in the data having a value for this attribute that no other instances have.
Below these statistics is a list showing more information about the values stored in this attribute, which differ depending on its type. If the attribute is nominal, the list consists of each possible value for the attribute along with the number of instances that have that value. If the attribute is numeric, the list gives four statistics describing the distribution of values in the data—the minimum, maximum, mean and standard deviation. And below these statistics there is a coloured histogram, colour-coded according to the attribute chosen as the Class using the box above the histogram. (This box will bring up a drop-down list of available selections when clicked.) Note that only nominal Class attributes will result in a colour-coding. Finally, after pressing the Visualize All button, histograms for all the attributes in the data are shown in a separate window.
Returning to the attribute list, to begin with all the tick boxes are unticked. They can be toggled on/off by clicking on them individually. The four buttons above can also be used to change the selection:
4.2. PREPROCESSING 39
1. All. All boxes are ticked.
2. None. All boxes are cleared (unticked).
3. Invert. Boxes that are ticked become unticked and vice versa.
4. Pattern. Enables the user to select attributes based on a Perl 5 Regular Expression. E.g., .* id selects all attributes which name ends with id.
Once the desired attributes have been selected, they can be removed by clicking the Remove button below the list of attributes. Note that this can be undone by clicking the Undo button, which is located next to the Edit button in the top-right corner of the Preprocess panel.
4.2.4 Working With Filters
The preprocess section allows filters to be defined that transform the data in various ways. The Filter box is used to set up the filters that are required. At the left of the Filter box is a Choose button. By clicking this button it is possible to select one of the filters in WEKA. Once a filter has been selected, its name and options are shown in the field next to the Choose button. Clicking on this box with the left mouse button brings up a GenericObjectEditor dialog box. A click with the right mouse button (or Alt+Shift+left click) brings up a menu where you can choose, either to display the properties in a GenericObjectEditor dialog box, or to copy the current setup string to the clipboard.
The GenericObjectEditor Dialog Box
The GenericObjectEditor dialog box lets you configure a filter. The same kind of dialog box is used to configure other objects, such as classifiers and clusterers (see below). The fields in the window reflect the available options.
Right-clicking (or Alt+Shift+Left-Click) on such a field will bring up a popup menu, listing the following options:
40 1.
2.
3.
CHAPTER 4. EXPLORER
Show properties… has the same effect as left-clicking on the field, i.e., a dialog appears allowing you to alter the settings.
Copy configuration to clipboard copies the currently displayed con- figuration string to the system’s clipboard and therefore can be used any- where else in WEKA or in the console. This is rather handy if you have to setup complicated, nested schemes.
Enter configuration… is the “receiving” end for configurations that got copied to the clipboard earlier on. In this dialog you can enter a classname followed by options (if the class supports these). This also allows you to transfer a filter setting from the Preprocess panel to a FilteredClassifier used in the Classify panel.
Left-Clicking on any of these gives an opportunity to alter the filters settings. For example, the setting may take a text string, in which case you type the string into the text field provided. Or it may give a drop-down box listing several states to choose from. Or it may do something else, depending on the information required. Information on the options is provided in a tool tip if you let the mouse pointer hover of the corresponding field. More information on the filter and its options can be obtained by clicking on the More button in the About panel at the top of the GenericObjectEditor window.
Some objects display a brief description of what they do in an About box, along with a More button. Clicking on the More button brings up a window describing what the different options do. Others have an additional button, Capabilities, which lists the types of attributes and classes the object can handle.
At the bottom of the GenericObjectEditor dialog are four buttons. The first two, Open… and Save… allow object configurations to be stored for future use. The Cancel button backs out without remembering any changes that have been made. Once you are happy with the object and settings you have chosen, click OK to return to the main Explorer window.
Applying Filters
Once you have selected and configured a filter, you can apply it to the data by pressing the Apply button at the right end of the Filter panel in the Preprocess panel. The Preprocess panel will then show the transformed data. The change can be undone by pressing the Undo button. You can also use the Edit… button to modify your data manually in a dataset editor. Finally, the Save… button at the top right of the Preprocess panel saves the current version of the relation in file formats that can represent the relation, allowing it to be kept for future use.
Note: Some of the filters behave differently depending on whether a class at- tribute has been set or not (using the box above the histogram, which will bring up a drop-down list of possible selections when clicked). In particular, the “supervised filters” require a class attribute to be set, and some of the “unsu- pervised attribute filters” will skip the class attribute if one is set. Note that it is also possible to set Class to None, in which case no class is set.
4.3. CLASSIFICATION 41 4.3 Classification
4.3.1 Selecting a Classifier
At the top of the classify section is the Classifier box. This box has a text field that gives the name of the currently selected classifier, and its options. Clicking on the text box with the left mouse button brings up a GenericObjectEditor dialog box, just the same as for filters, that you can use to configure the options of the current classifier. With a right click (or Alt+Shift+left click) you can once again copy the setup string to the clipboard or display the properties in a GenericObjectEditor dialog box. The Choose button allows you to choose one of the classifiers that are available in WEKA.
4.3.2 Test Options
The result of applying the chosen classifier will be tested according to the options that are set by clicking in the Test options box. There are four test modes:
1. Use training set. The classifier is evaluated on how well it predicts the class of the instances it was trained on.
2. Supplied test set. The classifier is evaluated on how well it predicts the class of a set of instances loaded from a file. Clicking the Set… button brings up a dialog allowing you to choose the file to test on.
3. Cross-validation. The classifier is evaluated by cross-validation, using the number of folds that are entered in the Folds text field.
4. Percentage split. The classifier is evaluated on how well it predicts a certain percentage of the data which is held out for testing. The amount of data held out depends on the value entered in the % field.
Note: No matter which evaluation method is used, the model that is output is always the one build from all the training data. Further testing options can be set by clicking on the More options… button:
42 1.
2. 3. 4. 5.
6.
7.
8.
9. 10. 11.
CHAPTER 4. EXPLORER
Output model. The classification model on the full training set is output so that it can be viewed, visualized, etc. This option is selected by default.
Output per-class stats. The precision/recall and true/false statistics for each class are output. This option is also selected by default.
Output entropy evaluation measures. Entropy evaluation measures are included in the output. This option is not selected by default.
Output confusion matrix. The confusion matrix of the classifier’s pre- dictions is included in the output. This option is selected by default.
Store predictions for visualization. The classifier’s predictions are remembered so that they can be visualized. This option is selected by default.
Output predictions. The predictions on the evaluation data are output. Note that in the case of a cross-validation the instance numbers do not correspond to the location in the data!
Output additional attributes. If additional attributes need to be out- put alongside the predictions, e.g., an ID attribute for tracking misclassi- fications, then the index of this attribute can be specified here. The usual Weka ranges are supported,“first” and “last” are therefore valid indices as well (example: “first-3,6,8,12-last”).
Cost-sensitive evaluation. The errors is evaluated with respect to a cost matrix. The Set… button allows you to specify the cost matrix used.
Random seed for xval / % Split. This specifies the random seed used when randomizing the data before it is divided up for evaluation purposes.
Preserve order for % Split. This suppresses the randomization of the data before splitting into train and test set.
Output source code. If the classifier can output the built model as Java source code, you can specify the class name here. The code will be printed in the “Classifier output” area.
4.3.3 The Class Attribute
The classifiers in WEKA are designed to be trained to predict a single ‘class’ attribute, which is the target for prediction. Some classifiers can only learn nominal classes; others can only learn numeric classes (regression problems); still others can learn both.
By default, the class is taken to be the last attribute in the data. If you want to train a classifier to predict a different attribute, click on the box below the Test options box to bring up a drop-down list of attributes to choose from.
4.3. CLASSIFICATION 43 4.3.4 Training a Classifier
Once the classifier, test options and class have all been set, the learning process is started by clicking on the Start button. While the classifier is busy being trained, the little bird moves around. You can stop the training process at any time by clicking on the Stop button.
When training is complete, several things happen. The Classifier output area to the right of the display is filled with text describing the results of training and testing. A new entry appears in the Result list box. We look at the result list below; but first we investigate the text that has been output.
4.3.5 The Classifier Output Text
The text in the Classifier output area has scroll bars allowing you to browse the results. Clicking with the left mouse button into the text area, while holding Alt and Shift, brings up a dialog that enables you to save the displayed output in a variety of formats (currently, BMP, EPS, JPEG and PNG). Of course, you can also resize the Explorer window to get a larger display area. The output is split into several sections:
1. Run information. A list of information giving the learning scheme op- tions, relation name, instances, attributes and test mode that were in- volved in the process.
2. Classifier model (full training set). A textual representation of the classification model that was produced on the full training data.
3. The results of the chosen test mode are broken down thus:
4. Summary. A list of statistics summarizing how accurately the classifier was able to predict the true class of the instances under the chosen test mode.
5. Detailed Accuracy By Class. A more detailed per-class break down of the classifier’s prediction accuracy.
6. Confusion Matrix. Shows how many instances have been assigned to each class. Elements show the number of test examples whose actual class is the row and whose predicted class is the column.
7. Source code (optional). This section lists the Java source code if one chose “Output source code” in the “More options” dialog.
4.3.6 The Result List
After training several classifiers, the result list will contain several entries. Left- clicking the entries flicks back and forth between the various results that have been generated. Pressing Delete removes a selected entry from the results. Right-clicking an entry invokes a menu containing these items:
1. View in main window. Shows the output in the main window (just like left-clicking the entry).
44 2.
3.
4. 5.
6.
7.
8.
9.
10.
11. 12.
CHAPTER 4. EXPLORER
View in separate window. Opens a new independent window for view- ing the results.
Save result buffer. Brings up a dialog allowing you to save a text file containing the textual output.
Load model. Loads a pre-trained model object from a binary file.
Save model. Saves a model object to a binary file. Objects are saved in
Java ‘serialized object’ form.
Re-evaluate model on current test set. Takes the model that has been built and tests its performance on the data set that has been specified with the Set.. button under the Supplied test set option.
Visualize classifier errors. Brings up a visualization window that plots the results of classification. Correctly classified instances are represented by crosses, whereas incorrectly classified ones show up as squares.
Visualize tree or Visualize graph. Brings up a graphical representation of the structure of the classifier model, if possible (i.e. for decision trees or Bayesian networks). The graph visualization option only appears if a Bayesian network classifier has been built. In the tree visualizer, you can bring up a menu by right-clicking a blank area, pan around by dragging the mouse, and see the training instances at each node by clicking on it. CTRL-clicking zooms the view out, while SHIFT-dragging a box zooms the view in. The graph visualizer should be self-explanatory.
Visualize margin curve. Generates a plot illustrating the prediction margin. The margin is defined as the difference between the probability predicted for the actual class and the highest probability predicted for the other classes. For example, boosting algorithms may achieve better performance on test data by increasing the margins on the training data.
Visualize threshold curve. Generates a plot illustrating the trade-offs in prediction that are obtained by varying the threshold value between classes. For example, with the default threshold value of 0.5, the pre- dicted probability of ‘positive’ must be greater than 0.5 for the instance to be predicted as ‘positive’. The plot can be used to visualize the pre- cision/recall trade-off, for ROC curve analysis (true positive rate vs false positive rate), and for other types of curves.
Visualize cost curve. Generates a plot that gives an explicit represen- tation of the expected cost, as described by [5].
Plugins. This menu item only appears if there are visualization plugins available (by default: none). More about these plugins can be found in the WekaWiki article “Explorer visualization plugins” [8].
Options are greyed out if they do not apply to the specific set of results.
4.4. CLUSTERING 45 4.4 Clustering
4.4.1 Selecting a Clusterer
By now you will be familiar with the process of selecting and configuring objects. Clicking on the clustering scheme listed in the Clusterer box at the top of the window brings up a GenericObjectEditor dialog with which to choose a new clustering scheme.
4.4.2 Cluster Modes
The Cluster mode box is used to choose what to cluster and how to evaluate the results. The first three options are the same as for classification: Use train- ing set, Supplied test set and Percentage split (Section 4.3.1)—except that now the data is assigned to clusters instead of trying to predict a specific class. The fourth mode, Classes to clusters evaluation, compares how well the chosen clusters match up with a pre-assigned class in the data. The drop-down box below this option selects the class, just as in the Classify panel.
An additional option in the Cluster mode box, the Store clusters for visualization tick box, determines whether or not it will be possible to visualize the clusters once training is complete. When dealing with datasets that are so large that memory becomes a problem it may be helpful to disable this option.
4.4.3 Ignoring Attributes
Often, some attributes in the data should be ignored when clustering. The Ignore attributes button brings up a small window that allows you to select which attributes are ignored. Clicking on an attribute in the window highlights it, holding down the SHIFT key selects a range of consecutive attributes, and holding down CTRL toggles individual attributes on and off. To cancel the selection, back out with the Cancel button. To activate it, click the Select button. The next time clustering is invoked, the selected attributes are ignored.
46 CHAPTER 4. EXPLORER 4.4.4 Working with Filters
The FilteredClusterer meta-clusterer offers the user the possibility to apply filters directly before the clusterer is learned. This approach eliminates the manual application of a filter in the Preprocess panel, since the data gets processed on the fly. Useful if one needs to try out different filter setups.
4.4.5 Learning Clusters
The Cluster section, like the Classify section, has Start/Stop buttons, a result text area and a result list. These all behave just like their classifica- tion counterparts. Right-clicking an entry in the result list brings up a similar menu, except that it shows only two visualization options: Visualize cluster assignments and Visualize tree. The latter is grayed out when it is not applicable.
4.5. ASSOCIATING 47 4.5 Associating
4.5.1 Setting Up
This panel contains schemes for learning association rules, and the learners are chosen and configured in the same way as the clusterers, filters, and classifiers in the other panels.
4.5.2 Learning Associations
Once appropriate parameters for the association rule learner bave been set, click the Start button. When complete, right-clicking on an entry in the result list allows the results to be viewed or saved.
48
CHAPTER 4. EXPLORER
4.6
Selecting Attributes
4.6.1
Searching and Evaluating
Attribute selection involves searching through all possible combinations of at- tributes in the data to find which subset of attributes works best for prediction. To do this, two objects must be set up: an attribute evaluator and a search method. The evaluator determines what method is used to assign a worth to each subset of attributes. The search method determines what style of search is performed.
4.6.2 Options
The Attribute Selection Mode box has two options:
1. Use full training set. The worth of the attribute subset is determined
using the full set of training data.
2. Cross-validation. The worth of the attribute subset is determined by a process of cross-validation. The Fold and Seed fields set the number of folds to use and the random seed used when shuffling the data.
As with Classify (Section 4.3.1), there is a drop-down box that can be used to specify which attribute to treat as the class.
4.6.3 Performing Selection
Clicking Start starts running the attribute selection process. When it is fin- ished, the results are output into the result area, and an entry is added to the result list. Right-clicking on the result list gives several options. The first three, (View in main window, View in separate window and Save result buffer), are the same as for the classify panel. It is also possible to Visualize
4.6. SELECTING ATTRIBUTES 49
reduced data, or if you have used an attribute transformer such as Principal- Components, Visualize transformed data. The reduced/transformed data can be saved to a file with the Save reduced data… or Save transformed data… option.
In case one wants to reduce/transform a training and a test at the same time and not use the AttributeSelectedClassifier from the classifier panel, it is best to use the AttributeSelection filter (a supervised attribute filter) in batch mode (’-b’) from the command line or in the SimpleCLI. The batch mode allows one to specify an additional input and output file pair (options -r and -s), that is processed with the filter setup that was determined based on the training data (specified by options -i and -o).
Here is an example for a Unix/Linux bash:
java weka.filters.supervised.attribute.AttributeSelection \ -E “weka.attributeSelection.CfsSubsetEval ” \
-S “weka.attributeSelection.BestFirst -D 1 -N 5” \ -b \
-i
Notes:
• The “backslashes” at the end of each line tell the bash that the command is not finished yet. Using the SimpleCLI one has to use this command in one line without the backslashes.
• It is assumed that WEKA is available in the CLASSPATH, otherwise one has to use the -classpath option.
• The full filter setup is output in the log, as well as the setup for running regular attribute selection.
50 CHAPTER 4. EXPLORER 4.7 Visualizing
WEKA’s visualization section allows you to visualize 2D plots of the current relation.
4.7.1 The scatter plot matrix
When you select the Visualize panel, it shows a scatter plot matrix for all the attributes, colour coded according to the currently selected class. It is possible to change the size of each individual 2D plot and the point size, and to randomly jitter the data (to uncover obscured points). It also possible to change the attribute used to colour the plots, to select only a subset of attributes for inclusion in the scatter plot matrix, and to sub sample the data. Note that changes will only come into effect once the Update button has been pressed.
4.7.2 Selecting an individual 2D scatter plot
When you click on a cell in the scatter plot matrix, this will bring up a separate window with a visualization of the scatter plot you selected. (We described above how to visualize particular results in a separate window—for example, classifier errors—the same visualization controls are used here.)
Data points are plotted in the main area of the window. At the top are two drop-down list buttons for selecting the axes to plot. The one on the left shows which attribute is used for the x-axis; the one on the right shows which is used for the y-axis.
Beneath the x-axis selector is a drop-down list for choosing the colour scheme. This allows you to colour the points based on the attribute selected. Below the plot area, a legend describes what values the colours correspond to. If the values are discrete, you can modify the colour used for each one by clicking on them and making an appropriate selection in the window that pops up.
To the right of the plot area is a series of horizontal strips. Each strip represents an attribute, and the dots within it show the distribution of values
4.7. VISUALIZING 51
of the attribute. These values are randomly scattered vertically to help you see concentrations of points. You can choose what axes are used in the main graph by clicking on these strips. Left-clicking an attribute strip changes the x-axis to that attribute, whereas right-clicking changes the y-axis. The ‘X’ and ‘Y’ written beside the strips shows what the current axes are (‘B’ is used for ‘both X and Y’).
Above the attribute strips is a slider labelled Jitter, which is a random displacement given to all points in the plot. Dragging it to the right increases the amount of jitter, which is useful for spotting concentrations of points. Without jitter, a million instances at the same point would look no different to just a single lonely instance.
4.7.3 Selecting Instances
There may be situations where it is helpful to select a subset of the data us- ing the visualization tool. (A special case of this is the UserClassifier in the Classify panel, which lets you build your own classifier by interactively selecting instances.)
Below the y-axis selector button is a drop-down list button for choosing a selection method. A group of data points can be selected in four ways:
1. Select Instance. Clicking on an individual data point brings up a window listing its attributes. If more than one point appears at the same location, more than one set of attributes is shown.
2. Rectangle. You can create a rectangle, by dragging, that selects the points inside it.
3. Polygon. You can build a free-form polygon that selects the points inside it. Left-click to add vertices to the polygon, right-click to complete it. The polygon will always be closed off by connecting the first point to the last.
4. Polyline. You can build a polyline that distinguishes the points on one side from those on the other. Left-click to add vertices to the polyline, right-click to finish. The resulting shape is open (as opposed to a polygon, which is always closed).
Once an area of the plot has been selected using Rectangle, Polygon or Polyline, it turns grey. At this point, clicking the Submit button removes all instances from the plot except those within the grey selection area. Clicking on the Clear button erases the selected area without affecting the graph.
Once any points have been removed from the graph, the Submit button changes to a Reset button. This button undoes all previous removals and returns you to the original graph with all points included. Finally, clicking the Save button allows you to save the currently visible instances to a new ARFF file.
52 CHAPTER 4. EXPLORER
Chapter 5 Experimenter
5.1 Introduction
The Weka Experiment Environment enables the user to create, run, modify, and analyse experiments in a more convenient manner than is possible when processing the schemes individually. For example, the user can create an exper- iment that runs several schemes against a series of datasets and then analyse the results to determine if one of the schemes is (statistically) better than the other schemes.
The Experiment Environment can be run from the command line using the Simple CLI. For example, the following commands could be typed into the CLI to run the OneR scheme on the Iris dataset using a basic train and test process. (Note that the commands would be typed on one line into the CLI.)
java weka.experiment.Experiment -r -T data/iris.arff -D weka.experiment.InstancesResultListener
-P weka.experiment.RandomSplitResultProducer — -W weka.experiment.ClassifierSplitEvaluator —
-W weka.classifiers.rules.OneR
While commands can be typed directly into the CLI, this technique is not particularly convenient and the experiments are not easy to modify.
The Experimenter comes in two flavours, either with a simple interface that provides most of the functionality one needs for experiments, or with an interface with full access to the Experimenter’s capabilities. You can choose between those two with the Experiment Configuration Mode radio buttons:
• Simple
• Advanced
Both setups allow you to setup standard experiments, that are run locally on a single machine, or remote experiments, which are distributed between several hosts. The distribution of experiments cuts down the time the experiments will take until completion, but on the other hand the setup takes more time.
The next section covers the standard experiments (both, simple and ad- vanced), followed by the remote experiments and finally the analysing of the results.
53
54 CHAPTER 5. EXPERIMENTER
5.2 Standard Experiments
5.2.1 Simple
5.2.1.1 New experiment
After clicking New default parameters for an Experiment are defined.
5.2.1.2 Results destination
By default, an ARFF file is the destination for the results output. But you can choose between
• ARFF file
• CSV file
• JDBC database
ARFF file and JDBC database are discussed in detail in the following sec- tions. CSV is similar to ARFF, but it can be used to be loaded in an external spreadsheet application.
ARFF file
If the file name is left empty a temporary file will be created in the TEMP directory of the system. If one wants to specify an explicit results file, click on Browse and choose a filename, e.g., Experiment1.arff.
5.2. STANDARD EXPERIMENTS 55
Click on Save and the name will appear in the edit field next to ARFF file.
The advantage of ARFF or CSV files is that they can be created without any additional classes besides the ones from Weka. The drawback is the lack of the ability to resume an experiment that was interrupted, e.g., due to an error or the addition of dataset or algorithms. Especially with time-consuming experiments, this behavior can be annoying.
JDBC database
With JDBC it is easy to store the results in a database. The necessary jar archives have to be in the CLASSPATH to make the JDBC functionality of a particular database available.
After changing ARFF file to JDBC database click on User… to specify JDBC URL and user credentials for accessing the database.
56 CHAPTER 5. EXPERIMENTER
After supplying the necessary data and clicking on OK, the URL in the main window will be updated.
Note: at this point, the database connection is not tested; this is done when the experiment is started.
The advantage of a JDBC database is the possibility to resume an in- terrupted or extended experiment. Instead of re-running all the other algo- rithm/dataset combinations again, only the missing ones are computed.
5.2.1.3 Experiment type
The user can choose between the following three different types
• Cross-validation (default)
performs stratified cross-validation with the given number of folds
• Train/Test Percentage Split (data randomized)
splits a dataset according to the given percentage into a train and a test file (one cannot specify explicit training and test files in the Experimenter), after the order of the data has been randomized and stratified
5.2. STANDARD EXPERIMENTS 57
• Train/Test Percentage Split (order preserved)
because it is impossible to specify an explicit train/test files pair, one can abuse this type to un-merge previously merged train and test file into the two original files (one only needs to find out the correct percentage)
Additionally, one can choose between Classification and Regression, depend- ing on the datasets and classifiers one uses. For decision trees like J48 (Weka’s implementation of Quinlan’s C4.5 [10]) and the iris dataset, Classification is necessary, for a numeric classifier like M5P, on the other hand, Regression. Clas- sification is selected by default.
Note: if the percentage splits are used, one has to make sure that the cor- rected paired T-Tester still produces sensible results with the given ratio [9].
58 CHAPTER 5. EXPERIMENTER 5.2.1.4 Datasets
One can add dataset files either with an absolute path or with a relative one. The latter makes it often easier to run experiments on different machines, hence one should check Use relative paths, before clicking on Add new….
In this example, open the data directory and choose the iris.arff dataset.
After clicking Open the file will be displayed in the datasets list. If one selects a directory and hits Open, then all ARFF files will be added recursively. Files can be deleted from the list by selecting them and then clicking on Delete selected.
ARFF files are not the only format one can load, but all files that can be converted with Weka’s “core converters”. The following formats are currently supported:
• ARFF (+ compressed) • C4.5
• CSV
• libsvm
• binary serialized instances • XRFF (+ compressed)
5.2. STANDARD EXPERIMENTS 59
By default, the class attribute is assumed to be the last attribute. But if a data format contains information about the class attribute, like XRFF or C4.5, this attribute will be used instead.
5.2.1.5 Iteration control
• Number of repetitions
In order to get statistically meaningful results, the default number of it- erations is 10. In case of 10-fold cross-validation this means 100 calls of one classifier with training data and tested against test data.
• Data sets first/Algorithms first
As soon as one has more than one dataset and algorithm, it can be useful to switch from datasets being iterated over first to algorithms. This is the case if one stores the results in a database and wants to complete the results for all the datasets for one algorithm as early as possible.
5.2.1.6 Algorithms
New algorithms can be added via the Add new… button. Opening this dialog for the first time, ZeroR is presented, otherwise the one that was selected last.
With the Choose button one can open the GenericObjectEditor and choose another classifier.
60 CHAPTER 5. EXPERIMENTER
The Filter… button enables one to highlight classifiers that can handle certain attribute and class types. With the Remove filter button all the selected capabilities will get cleared and the highlighting removed again.
Additional algorithms can be added again with the Add new… button, e.g., the J48 decision tree.
After setting the classifier parameters, one clicks on OK to add it to the list of algorithms.
5.2. STANDARD EXPERIMENTS 61
With the Load options… and Save options… buttons one can load and save the setup of a selected classifier from and to XML. This is especially useful for highly configured classifiers (e.g., nested meta-classifiers), where the manual setup takes quite some time, and which are used often.
One can also paste classifier settings here by right-clicking (or Alt-Shift-left- clicking) and selecting the appropriate menu point from the popup menu, to either add a new classifier or replace the selected one with a new setup. This is rather useful for transferring a classifier setup from the Weka Explorer over to the Experimenter without having to setup the classifier from scratch.
5.2.1.7 Saving the setup
For future re-use, one can save the current setup of the experiment to a file by clicking on Save… at the top of the window.
By default, the format of the experiment files is the binary format that Java serialization offers. The drawback of this format is the possible incompatibility between different versions of Weka. A more robust alternative to the binary format is the XML format.
Previously saved experiments can be loaded again via the Open… button.
62 CHAPTER 5. EXPERIMENTER 5.2.1.8 Running an Experiment
To run the current experiment, click the Run tab at the top of the Experiment Environment window. The current experiment performs 10 runs of 10-fold strat- ified cross-validation on the Iris dataset using the ZeroR and J48 scheme.
Click Start to run the experiment.
If the experiment was defined correctly, the 3 messages shown above will be displayed in the Log panel. The results of the experiment are saved to the dataset Experiment1.arff.
5.2. STANDARD EXPERIMENTS 63 5.2.2 Advanced
5.2.2.1 Defining an Experiment
When the Experimenter is started in Advanced mode, the Setup tab is displayed. Click New to initialize an experiment. This causes default parameters to be defined for the experiment.
To define the dataset to be processed by a scheme, first select Use relative paths in the Datasets panel of the Setup tab and then click on Add new… to open a dialog window.
Double click on the data folder to view the available datasets or navigate to an alternate location. Select iris.arff and click Open to select the Iris dataset.
64 CHAPTER 5. EXPERIMENTER
The dataset name is now displayed in the Datasets panel of the Setup tab. Saving the Results of the Experiment
To identify a dataset to which the results are to be sent, click on the Instances- ResultListener entry in the Destination panel. The output file parameter is near the bottom of the window, beside the text outputFile. Click on this parameter to display a file selection window.
5.2. STANDARD EXPERIMENTS 65
Type the name of the output file and click Select. The file name is displayed in the outputFile panel. Click on OK to close the window.
The dataset name is displayed in the Destination panel of the Setup tab.
Saving the Experiment Definition
The experiment definition can be saved at any time. Select Save… at the top of the Setup tab. Type the dataset name with the extension exp (or select the dataset name if the experiment definition dataset already exists) for binary files or choose Experiment configuration files (*.xml) from the file types combobox (the XML files are robust with respect to version changes).
66 CHAPTER 5. EXPERIMENTER
The experiment can be restored by selecting Open in the Setup tab and then selecting Experiment1.exp in the dialog window.
5.2.2.2 Running an Experiment
To run the current experiment, click the Run tab at the top of the Experiment Environment window. The current experiment performs 10 randomized train and test runs on the Iris dataset, using 66% of the patterns for training and 34% for testing, and using the ZeroR scheme.
Click Start to run the experiment.
5.2. STANDARD EXPERIMENTS 67
If the experiment was defined correctly, the 3 messages shown above will be displayed in the Log panel. The results of the experiment are saved to the dataset Experiment1.arff. The first few lines in this dataset are shown below.
@relation InstanceResultListener
@attribute Key_Dataset {iris}
@attribute Key_Run {1,2,3,4,5,6,7,8,9,10}
@attribute Key_Scheme {weka.classifiers.rules.ZeroR,weka.classifiers.trees.J48} @attribute Key_Scheme_options {,’-C 0.25 -M 2’}
@attribute Key_Scheme_version_ID {48055541465867954,-217733168393644444} @attribute Date_time numeric
@attribute Number_of_training_instances numeric
@attribute Number_of_testing_instances numeric
@attribute Number_correct numeric
@attribute Number_incorrect numeric
@attribute Number_unclassified numeric
@attribute Percent_correct numeric
@attribute Percent_incorrect numeric
@attribute Percent_unclassified numeric
@attribute Kappa_statistic numeric
@attribute Mean_absolute_error numeric
@attribute Root_mean_squared_error numeric
@attribute Relative_absolute_error numeric
@attribute Root_relative_squared_error numeric
@attribute SF_prior_entropy numeric
@attribute SF_scheme_entropy numeric
@attribute SF_entropy_gain numeric
@attribute SF_mean_prior_entropy numeric
@attribute SF_mean_scheme_entropy numeric
@attribute SF_mean_entropy_gain numeric
@attribute KB_information numeric
68 CHAPTER 5.
EXPERIMENTER
@attribute KB_mean_information numeric @attribute KB_relative_information numeric @attribute True_positive_rate numeric @attribute Num_true_positives numeric @attribute False_positive_rate numeric @attribute Num_false_positives numeric @attribute True_negative_rate numeric @attribute Num_true_negatives numeric @attribute False_negative_rate numeric @attribute Num_false_negatives numeric @attribute IR_precision numeric @attribute IR_recall numeric
@attribute F_measure numeric @attribute Area_under_ROC numeric @attribute Time_training numeric @attribute Time_testing numeric @attribute Summary {’Number of leaves:
3\nSize of the tree: 5\n’, ’Number of leaves: 5\nSize of the tree: 9\n’,
’Number of leaves: 4\nSize of the tree: 7\n’} @attribute measureTreeSize numeric
@attribute measureNumLeaves numeric
@attribute measureNumRules numeric
@data
iris,1,weka.classifiers.rules.ZeroR,,48055541465867954,20051221.033,99,51, 17,34,0,33.333333,66.666667,0,0,0.444444,0.471405,100,100,80.833088,80.833088, 0,1.584963,1.584963,0,0,0,0,1,17,1,34,0,0,0,0,0.333333,1,0.5,0.5,0,0,?,?,?,?
5.2.2.3 Changing the Experiment Parameters Changing the Classifier
The parameters of an experiment can be changed by clicking on the Result generator panel.
The RandomSplitResultProducer performs repeated train/test runs. The number of instances (expressed as a percentage) used for training is given in the
5.2. STANDARD EXPERIMENTS 69
trainPercent box. (The number of runs is specified in the Runs panel in the Setup tab.)
A small help file can be displayed by clicking More in the About panel.
Click on the splitEvaluator entry to display the SplitEvaluator properties.
Click on the classifier entry (ZeroR) to display the scheme properties.
This scheme has no modifiable properties (besides debug mode on/off) but most other schemes do have properties that can be modified by the user. The Capabilities button opens a small dialog listing all the attribute and class types this classifier can handle. Click on the Choose button to select a different scheme. The window below shows the parameters available for the J48 decision- tree scheme. If desired, modify the parameters and then click OK to close the window.
70 CHAPTER 5. EXPERIMENTER
The name of the new scheme is displayed in the Result generator panel.
Adding Additional Schemes
Additional schemes can be added in the Generator properties panel. To begin, change the drop-down list entry from Disabled to Enabled in the Generator properties panel.
5.2. STANDARD EXPERIMENTS 71
Click Select property and expand splitEvaluator so that the classifier entry is visible in the property list; click Select.
The scheme name is displayed in the Generator properties panel.
72 CHAPTER 5. EXPERIMENTER
To add another scheme, click on the Choose button to display the Generic- ObjectEditor window.
The Filter… button enables one to highlight classifiers that can handle certain attribute and class types. With the Remove filter button all the selected capabilities will get cleared and the highlighting removed again.
To change to a decision-tree scheme, select J48 (in subgroup trees).
5.2. STANDARD EXPERIMENTS 73
The new scheme is added to the Generator properties panel. Click Add to add the new scheme.
Now when the experiment is run, results are generated for both schemes.
To add additional schemes, repeat this process. To remove a scheme, select the scheme by clicking on it and then click Delete.
Adding Additional Datasets
The scheme(s) may be run on any number of datasets at a time. Additional datasets are added by clicking Add new… in the Datasets panel. Datasets are deleted from the experiment by selecting the dataset and then clicking Delete Selected.
74 CHAPTER 5. EXPERIMENTER Raw Output
The raw output generated by a scheme during an experiment can be saved to a file and then examined at a later time. Open the ResultProducer window by clicking on the Result generator panel in the Setup tab.
Click on rawOutput and select the True entry from the drop-down list. By default, the output is sent to the zip file splitEvaluatorOut.zip. The output file can be changed by clicking on the outputFile panel in the window. Now when the experiment is run, the result of each processing run is archived, as shown below.
The contents of the first run are:
ClassifierSplitEvaluator: weka.classifiers.trees.J48 -C 0.25 -M 2(version -217733168393644444)Classifier model:
J48 pruned tree ——————
petalwidth <= 0.6: Iris-setosa (33.0) petalwidth > 0.6
| petalwidth <= 1.5: Iris-versicolor (31.0/1.0)
| petalwidth > 1.5: Iris-virginica (35.0/3.0)
Number of Leaves : 3 Sizeofthetree: 5
5.2. STANDARD EXPERIMENTS
75
Correctly Classified Instances Incorrectly Classified Instances Kappa statistic
Mean absolute error
Root mean squared error Relative absolute error Root relative squared error Total Number of Instances measureTreeSize : 5.0 measureNumLeaves : 3.0 measureNumRules : 3.0
5.2.2.4 Other Result Producers
Cross-Validation Result Producer
47 4
0.8824
0.0723
0.2191
16.2754
46.4676
51
92.1569 %
7.8431 %
To change from random train and test experiments to cross-validation exper- iments, click on the Result generator entry. At the top of the window, click on the drop-down list and select CrossValidationResultProducer. The window now contains parameters specific to cross-validation such as the number of par- titions/folds. The experiment performs 10-fold cross-validation instead of train and test in the given example.
The Result generator panel now indicates that cross-validation will be per- formed. Click on More to generate a brief description of the CrossValidation- ResultProducer.
% %
76 CHAPTER 5. EXPERIMENTER As with the RandomSplitResultProducer, multiple schemes can be run during
cross-validation by adding them to the Generator properties panel.
The number of runs is set to 1 in the Setup tab in this example, so that only one run of cross-validation for each scheme and dataset is executed.
When this experiment is analysed, the following results are generated. Note that there are 30 (1 run times 10 folds times 3 schemes) result lines processed.
Averaging Result Producer
An alternative to the CrossValidationResultProducer is the AveragingResultPro- ducer. This result producer takes the average of a set of runs (which are typ- ically cross-validation runs). This result producer is identified by clicking the Result generator panel and then choosing the AveragingResultProducer from the GenericObjectEditor.
5.2. STANDARD EXPERIMENTS 77
The associated help file is shown below.
Clicking the resultProducer panel brings up the following window.
As with the other ResultProducers, additional schemes can be defined. When the AveragingResultProducer is used, the classifier property is located deeper in the Generator properties hierarchy.
78 CHAPTER 5. EXPERIMENTER
In this experiment, the ZeroR, OneR, and J48 schemes are run 10 times with 10-fold cross-validation. Each set of 10 cross-validation folds is then averaged, producing one result line for each run (instead of one result line for each fold as in the previous example using the CrossValidationResultProducer) for a total of 30 result lines. If the raw output is saved, all 300 results are sent to the archive.
5.2. STANDARD EXPERIMENTS 79
80 CHAPTER 5. EXPERIMENTER 5.3 Cluster Experiments
Using the advanced mode of the Experimenter you can now run experiments on clustering algorithms as well as classifiers (Note: this is a new feature available with Weka 3.6.0). The main evaluation metric for this type of experiment is the log likelihood of the clusters found by each clusterer. Here is an example of setting up a cross-validation experiment using clusterers.
Choose CrossValidationResultProducer from the Result generator panel.
5.3. CLUSTER EXPERIMENTS 81 Next, choose DensityBasedClustererSplitEvaluator as the split evaluator to use.
If you click on DensityBasedClustererSplitEvaluator you will see its options. Note that there is an option for removing the class column from the data. In the Experimenter, the class column is set to be the last column by default. Turn this off if you want to keep this column in the data.
Once DensityBasedClustererSplitEvaluator has been selected, you will notice that the Generator properties have become disabled. Enable them again and expand splitEvaluator. Select the clusterer node.
Now you will see that EM becomes the default clusterer and gets added to the list of schemes. You can now add/delete other clusterers.
IMPORTANT: in order to any clusterer that does not produce density esti- mates (i.e. most other clusterers in Weka), they will have to wrapped in the MakeDensityBasedClusterer.
82 CHAPTER 5. EXPERIMENTER
Once and experiment has been run, you can analyze results in the Analyse panel. In the Comparison field you will need to scroll down and select ”Log likelihood”.
5.4. REMOTE EXPERIMENTS 83 5.4 Remote Experiments
Remote experiments enable you to distribute the computing load across multiple computers. In the following we will discuss the setup and operation for HSQLDB [12] and MySQL [13].
5.4.1 Preparation
To run a remote experiment you will need:
• A database server.
• A number of computers to run remote engines on.
• To edit the remote engine policy file included in the Weka distribution to allow Java class and dataset loading from your home directory.
• An invocation of the Experimenter on a machine somewhere (any will do). For the following examples, we assume a user called johndoe with this setup:
• Access to a set of computers running a flavour of Unix (pathnames need to be changed for Windows).
• The home directory is located at /home/johndoe.
• Weka is found in /home/johndoe/weka.
• Additionaljararchives,i.e.,JDBCdrivers,arestoredin/home/johndoe/jars.
• The directory for the datasets is /home/johndoe/datasets.
Note: The example policy file remote.policy.example is using this setup (available in weka/experiment1).
5.4.2 Database Server Setup
• HSQLDB
– Download the JDBC driver for HSQLDB, extract the hsqldb.jar
and place it in the directory /home/johndoe/jars.
– To set up the database server, choose or create a directory to run the
database server from, and start the server with:
java -classpath /home/johndoe/jars/hsqldb.jar \ org.hsqldb.Server \
-database.0 experiment -dbname.0 experiment
Note: This will start up a database with the alias “experiment” (-dbname.0
1Weka’s source code can be found in the weka-src.jar archive or obtained from Subversion [11].
84
CHAPTER 5. EXPERIMENTER
5.4.3
• MySQL
We won’t go into the details of setting up a MySQL server, but this is
rather straightforward and includes the following steps:
– Download a suitable version of MySQL for your server machine.
– Install and start the MySQL server.
– Create a database – for our example we will use experiment as database name.
– Download the appropriate JDBC driver, extract the JDBC jar and place it as mysql.jar in /home/johndoe/jars.
Remote Engine Setup
• First, set up a directory for scripts and policy files: /home/johndoe/remote_engine
• Unzip the remoteExperimentServer.jar (from the Weka distribution; or build it from the sources2 with ant remotejar) into a temporary direc- tory.
• Next, copy remoteEngine.jar and remote.policy.example to the /home/johndoe/remote engine directory.
• Createascript,called/home/johndoe/remoteengine/startRemoteEngine, with the following content (don’t forget to make it executable with chmod a+x startRemoteEngine when you are on Linux/Unix):
– HSQLDB
java -Xmx256m \
-classpath /home/johndoe/jars/hsqldb.jar:remoteEngine.jar \ -Djava.security.policy=remote.policy \ weka.experiment.RemoteEngine &
– MySQL
java -Xmx256m \
-classpath /home/johndoe/jars/mysql.jar:remoteEngine.jar \ -Djava.security.policy=remote.policy \ weka.experiment.RemoteEngine &
• Now we will start the remote engines that run the experiments on the remote computers (note that the same version of Java must be used for the Experimenter and remote engines):
– Rename the remote.policy.example file to remote.policy. – For each machine you want to run a remote engine on:
∗ ssh to the machine.
2Weka’s source code can be found in the weka-src.jar archive or obtained from Subversion [11].
5.4. REMOTE EXPERIMENTS 85 ∗ cd to /home/johndoe/remote engine.
∗ Run/home/johndoe/startRemoteEngine(toenabletheremote engines to use more memory, modify the -Xmx option in the startRemoteEngine script) .
5.4.4 Configuring the Experimenter
Now we will run the Experimenter:
• HSQLDB
– Copy the DatabaseUtils.props.hsql file from weka/experiment in the weka.jar archive to the /home/johndoe/remote engine directory and rename it to DatabaseUtils.props.
– Editthisfileandchangethe”jdbcURL=jdbc:hsqldb:hsql://servername/databasename” entry to include the name of the machine that is running your database
server (e.g., jdbcURL=jdbc:hsqldb:hsql://dodo.company.com/experiment).
– Now start the Experimenter (inside this directory):
java \
-cp /home/johndoe/jars/hsqldb.jar:remoteEngine.jar:/home/johndoe/weka/weka.jar \ -Djava.rmi.server.codebase=file:/home/johndoe/weka/weka.jar \ weka.gui.experiment.Experimenter
• MySQL
– Copy the DatabaseUtils.props.mysql file from weka/experiment in the weka.jar archive to the /home/johndoe/remote engine direc- tory and rename it to DatabaseUtils.props.
– Editthisfileandchangethe”jdbcURL=jdbc:mysql://servername:3306/databasename” entry to include the name of the machine that is running your database
server and the name of the database the result will be stored in (e.g., jdbcURL=jdbc:mysql://dodo.company.com:3306/experiment).
– Now start the Experimenter (inside this directory):
java \
-cp /home/johndoe/jars/mysql.jar:remoteEngine.jar:/home/johndoe/weka/weka.jar \ -Djava.rmi.server.codebase=file:/home/johndoe/weka/weka.jar \ weka.gui.experiment.Experimenter
Note: the database name experiment can still be modified in the Exper- imenter, this is just the default setup.
Now we will configure the experiment:
• First of all select the Advanced mode in the Setup tab
• Now choose the DatabaseResultListener in the Destination panel. Config- ure this result producer:
– HSQLDB
Supply the value sa for the username and leave the password empty.
86
CHAPTER 5. EXPERIMENTER
•
• •
• • • •
– MySQL
Provide the username and password that you need for connecting to the database.
From the Result generator panel choose either the CrossValidationResult- Producer or the RandomSplitResultProducer (these are the most com- monly used ones) and then configure the remaining experiment details (e.g., datasets and classifiers).
Now enable the Distribute Experiment panel by checking the tick box. Click on the Hosts button and enter the names of the machines that you
started remote engines on (
Save your experiment configuration.
Now start your experiment as you would do normally.
Check your results in the Analyse tab by clicking either the Database or Experiment buttons.
5.4.5 Multi-core support
If you want to utilize all the cores on a multi-core machine, then you can do so with Weka. All you have to do, is define the port alongside the hostname in the Experimenter (format: hostname:port) and then start the RemoteEngine with the -p option, specifying the port to listen on.
5.4.6 Troubleshooting
• If you get an error at the start of an experiment that looks a bit like this:
01:13:19: RemoteExperiment(//blabla.company.com/RemoteEngine) (sub)experiment (datataset vineyard.arff) failed : java.sql.SQLException: Table already exists: EXPERIMENT INDEX in statement [CREATE TABLE Experiment index ( Experiment type LONGVARCHAR, Experiment setup LONGVARCHAR, Result table INT )]
01:13:19: dataset :vineyard.arff RemoteExperiment (//blabla.company.com/RemoteEngine) (sub)experiment (datataset vineyard.arff) failed : java.sql.SQLException: Table already exists: EXPERIMENTINDEXinstatement[CREATETABLE
Experiment index ( Experiment type LONGVARCHAR, Experiment setup LONGVARCHAR, Result table INT )]. Scheduling for execution on another host.
then do not panic – this happens because multiple remote machines are trying to create the same table and are temporarily locked out – this will resolve itself so just leave your experiment running – in fact, it is a sign that the experiment is working!
5.4. REMOTE EXPERIMENTS 87
• If you serialized an experiment and then modify your DatabaseUtils.props file due to an error (e.g., a missing type-mapping), the Experimenter will use the DatabaseUtils.props you had at the time you serialized the ex- periment. Keep in mind that the serialization process also serializes the DatabaseUtils class and therefore stored your props-file! This is another reason for storing your experiments as XML and not in the properietary binary format the Java serialization produces.
• Using a corrupt or incomplete DatabaseUtils.props file can cause peculiar interface errors, for example disabling the use of the ”User” button along- side the database URL. If in doubt copy a clean DatabaseUtils.props from Subversion [11].
• If you get NullPointerException at java.util.Hashtable.get() in the Remote Engine do not be alarmed. This will have no effect on the results of your experiment.
88 CHAPTER 5. EXPERIMENTER 5.5 Analysing Results
5.5.1 Setup
Weka includes an experiment analyser that can be used to analyse the results of experiments (in this example, the results were sent to an InstancesResultLis- tener ). The experiment shown below uses 3 schemes, ZeroR, OneR, and J48, to classify the Iris data in an experiment using 10 train and test runs, with 66% of the data used for training and 34% used for testing.
After the experiment setup is complete, run the experiment. Then, to anal- yse the results, select the Analyse tab at the top of the Experiment Environment window.
Click on Experiment to analyse the results of the current experiment.
5.5. ANALYSING RESULTS 89
The number of result lines available (Got 30 results) is shown in the Source panel. This experiment consisted of 10 runs, for 3 schemes, for 1 dataset, for a total of 30 result lines. Results can also be loaded from an earlier experiment file by clicking File and loading the appropriate .arff results file. Similarly, results sent to a database (using the DatabaseResultListener) can be loaded from the database.
Select the Percent correct attribute from the Comparison field and click Perform test to generate a comparison of the 3 schemes.
The schemes used in the experiment are shown in the columns and the datasets used are shown in the rows.
The percentage correct for each of the 3 schemes is shown in each dataset row: 33.33% for ZeroR, 94.31% for OneR, and 94.90% for J48. The annotation v or * indicates that a specific result is statistically better (v) or worse (*) than the baseline scheme (in this case, ZeroR) at the significance level specified (currently 0.05). The results of both OneR and J48 are statistically better than the baseline established by ZeroR. At the bottom of each column after the first column is a count (xx/ yy/ zz) of the number of times that the scheme was better than (xx), the same as (yy), or worse than (zz), the baseline scheme on the datasets used in the experiment. In this example, there was only one dataset and OneR was better than ZeroR once and never equivalent to or worse than ZeroR (1/0/0); J48 was also better than ZeroR on the dataset.
The standard deviation of the attribute being evaluated can be generated by selecting the Show std. deviations check box and hitting Perform test again. The value (10) at the beginning of the iris row represents the number of esti- mates that are used to calculate the standard deviation (the number of runs in this case).
90 CHAPTER 5. EXPERIMENTER
Selecting Number correct as the comparison field and clicking Perform test generates the average number correct (out of 50 test patterns – 33% of 150 patterns in the Iris dataset).
Clicking on the button for the Output format leads to a dialog that lets you choose the precision for the mean and the std. deviations, as well as the format of the output. Checking the Show Average checkbox adds an additional line to the output listing the average of each column. With the Remove filter classnames checkbox one can remove the filter name and options from processed datasets (filter names in Weka can be quite lengthy).
The following formats are supported: • CSV
• GNUPlot • HTML
5.5. ANALYSING RESULTS 91
• LaTeX
• Plain text (default) • Significance only
5.5.2 Saving the Results
The information displayed in the Test output panel is controlled by the currently- selected entry in the Result list panel. Clicking on an entry causes the results corresponding to that entry to be displayed.
The results shown in the Test output panel can be saved to a file by clicking Save output. Only one set of results can be saved at a time but Weka permits the user to save all results to the same file by saving them one at a time and using the Append option instead of the Overwrite option for the second and subsequent saves.
5.5.3 Changing the Baseline Scheme
The baseline scheme can be changed by clicking Select base… and then selecting the desired scheme. Selecting the OneR scheme causes the other schemes to be compared individually with the OneR scheme.
92 CHAPTER 5. EXPERIMENTER
If the test is performed on the Percent correct field with OneR as the base scheme, the system indicates that there is no statistical difference between the results for OneR and J48. There is however a statistically significant difference between OneR and ZeroR.
5.5.4 Statistical Significance
The term statistical significance used in the previous section refers to the re- sult of a pair-wise comparison of schemes using either a standard T-Test or the corrected resampled T-Test [9]. The latter test is the default, because the stan- dard T-Test can generate too many significant differences due to dependencies in the estimates (in particular when anything other than one run of an x-fold cross-validation is used). For more information on the T-Test, consult the Weka book [1] or an introductory statistics text. As the significance level is decreased, the confidence in the conclusion increases.
In the current experiment, there is not a statistically significant difference between the OneR and J48 schemes.
5.5.5 Summary Test
Selecting Summary from Test base and performing a test causes the following information to be generated.
5.5. ANALYSING RESULTS 93
In this experiment, the first row (- 1 1) indicates that column b (OneR) is better than row a (ZeroR) and that column c (J48) is also better than row a. The number in brackets represents the number of significant wins for the column with regard to the row. A 0 means that the scheme in the corresponding column did not score a single (significant) win with regard to the scheme in the row.
5.5.6 Ranking Test
Selecting Ranking from Test base causes the following information to be gener- ated.
The ranking test ranks the schemes according to the total number of sig- nificant wins (>) and losses (<) against the other schemes. The first column (> − <) is the difference between the number of wins and the number of losses. This difference is used to generate the ranking.
94 CHAPTER 5. EXPERIMENTER
Chapter 6 KnowledgeFlow
6.1 Introduction
The KnowledgeFlow provides an alternative to the Explorer as a graphical front end to WEKA’s core algorithms. The KnowledgeFlow is a work in progress so some of the functionality from the Explorer is not yet available. On the other hand, there are things that can be done in the KnowledgeFlow but not in the Explorer.
The KnowledgeFlow presents a data-flow inspired interface to WEKA. The user can select WEKA components from a tool bar, place them on a layout can- vas and connect them together in order to form a knowledge flow for processing and analyzing data. At present, all of WEKA’s classifiers, filters, clusterers, loaders and savers are available in the KnowledgeFlow along with some extra tools.
The KnowledgeFlow can handle data either incrementally or in batches (the Explorer handles batch data only). Of course learning from data incremen-
95
96 CHAPTER 6. KNOWLEDGEFLOW tally requires a classifier that can be updated on an instance by instance basis.
Currently in WEKA there are ten classifiers that can handle data incrementally:
• AODE
• IB1
• IBk
• KStar
• NaiveBayesMultinomialUpdateable • NaiveBayesUpdateable
• NNge
• Winnow
And two of them are meta classifiers:
• RacedIncrementalLogitBoost - that can use of any regression base learner to learn from discrete class data incrementally.
• LWL - locally weighted learning.
6.2. FEATURES 97
6.2 Features
The KnowledgeFlow offers the following features:
• intuitive data flow style layout
• process data in batches or incrementally
• process multiple batches or streams in parallel (each separate flow executes in its own thread)
• chain filters together
• view models produced by classifiers for each fold in a cross validation
• visualizeperformanceofincrementalclassifiersduringprocessing(scrolling plots of classification accuracy, RMS error, predictions etc.)
• plugin facility for allowing easy addition of new components to the Knowl- edgeFlow
98 CHAPTER 6. 6.3 Components
Components available in the KnowledgeFlow:
6.3.1 DataSources
All of WEKA’s loaders are available.
6.3.2 DataSinks
All of WEKA’s savers are available.
6.3.3 Filters
All of WEKA’s filters are available.
6.3.4 Classifiers
All of WEKA’s classifiers are available.
6.3.5 Clusterers
All of WEKA’s clusterers are available.
KNOWLEDGEFLOW
6.3. COMPONENTS 99 6.3.6 Evaluation
• TrainingSetMaker - make a data set into a training set.
• TestSetMaker - make a data set into a test set.
• CrossValidationFoldMaker - split any data set, training set or test set into folds.
• TrainTestSplitMaker - split any data set, training set or test set into a training set and a test set.
• ClassAssigner - assign a column to be the class for any data set, training set or test set.
• ClassValuePicker - choose a class value to be considered as the “posi- tive” class. This is useful when generating data for ROC style curves (see ModelPerformanceChart below and example 6.4.2).
• ClassifierPerformanceEvaluator - evaluate the performance of batch trained/tested classifiers.
• IncrementalClassifierEvaluator - evaluate the performance of incremen- tally trained classifiers.
• ClustererPerformanceEvaluator - evaluate the performance of batch trained/tested clusterers.
• PredictionAppender - append classifier predictions to a test set. For dis- crete class problems, can either append predicted class labels or probabil- ity distributions.
100 CHAPTER 6. KNOWLEDGEFLOW 6.3.7 Visualization
• DataVisualizer - component that can pop up a panel for visualizing data in a single large 2D scatter plot.
• ScatterPlotMatrix - component that can pop up a panel containing a ma- trix of small scatter plots (clicking on a small plot pops up a large scatter plot).
• AttributeSummarizer - component that can pop up a panel containing a matrix of histogram plots - one for each of the attributes in the input data.
• ModelPerformanceChart - component that can pop up a panel for visual- izing threshold (i.e. ROC style) curves.
• TextViewer - component for showing textual data. Can show data sets, classification performance statistics etc.
• GraphViewer - component that can pop up a panel for visualizing tree based models.
• StripChart - component that can pop up a panel that displays a scrolling plot of data (used for viewing the online performance of incremental clas- sifiers).
6.4. EXAMPLES 101 6.4 Examples
6.4.1 Cross-validated J48
Setting up a flow to load an ARFF file (batch mode) and perform a cross- validation using J48 (WEKA’s C4.5 implementation).
• Click on the DataSources tab and choose ArffLoader from the toolbar (the mouse pointer will change to a cross hairs).
• Next place the ArffLoader component on the layout area by clicking some- where on the layout (a copy of the ArffLoader icon will appear on the layout area).
• Next specify an ARFF file to load by first right clicking the mouse over the ArffLoader icon on the layout. A pop-up menu will appear. Select Configure under Edit in the list from this menu and browse to the location of your ARFF file.
• Next click the Evaluation tab at the top of the window and choose the ClassAssigner (allows you to choose which column to be the class) com- ponent from the toolbar. Place this on the layout.
• Now connect the ArffLoader to the ClassAssigner: first right click over the ArffLoader and select the dataSet under Connections in the menu. A rubber band line will appear. Move the mouse over the ClassAssigner component and left click - a red line labeled dataSet will connect the two components.
• Next right click over the ClassAssigner and choose Configure from the menu. This will pop up a window from which you can specify which column is the class in your data (last is the default).
• Next grab a CrossValidationFoldMaker component from the Evaluation toolbar and place it on the layout. Connect the ClassAssigner to the CrossValidationFoldMaker by right clicking over ClassAssigner and se- lecting dataSet from under Connections in the menu.
• Next click on the Classifiers tab at the top of the window and scroll along the toolbar until you reach the J48 component in the trees section. Place a J48 component on the layout.
102 CHAPTER 6. KNOWLEDGEFLOW
• Connect the CrossValidationFoldMaker to J48 TWICE by first choosing trainingSet and then testSet from the pop-up menu for the CrossValida- tionFoldMaker.
• Next go back to the Evaluation tab and place a ClassifierPerformanceE- valuator component on the layout. Connect J48 to this component by selecting the batchClassifier entry from the pop-up menu for J48.
• Next go to the Visualization toolbar and place a TextViewer compo- nent on the layout. Connect the ClassifierPerformanceEvaluator to the TextViewer by selecting the text entry from the pop-up menu for Classi- fierPerformanceEvaluator.
• Now start the flow executing by selecting Start loading from the pop-up menu for ArffLoader. Depending on how big the data set is and how long cross-validation takes you will see some animation from some of the icons in the layout (J48’s tree will grow in the icon and the ticks will animate on the ClassifierPerformanceEvaluator). You will also see some progress information in the Status bar and Log at the bottom of the window.
When finished you can view the results by choosing Show results from the pop-up menu for the TextViewer component.
Other cool things to add to this flow: connect a TextViewer and/or a GraphViewer to J48 in order to view the textual or graphical representations of the trees produced for each fold of the cross validation (this is something that is not possible in the Explorer).
6.4. EXAMPLES 103 6.4.2 Plotting multiple ROC curves
The KnowledgeFlow can draw multiple ROC curves in the same plot window, something that the Explorer cannot do. In this example we use J48 and Ran- domForest as classifiers. This example can be found on the WekaWiki as well [14].
• Click on the DataSources tab and choose ArffLoader from the toolbar (the mouse pointer will change to a cross hairs).
• Next place the ArffLoader component on the layout area by clicking some- where on the layout (a copy of the ArffLoader icon will appear on the layout area).
• Next specify an ARFF file to load by first right clicking the mouse over the ArffLoader icon on the layout. A pop-up menu will appear. Select Configure under Edit in the list from this menu and browse to the location of your ARFF file.
• Next click the Evaluation tab at the top of the window and choose the ClassAssigner (allows you to choose which column to be the class) com- ponent from the toolbar. Place this on the layout.
• Now connect the ArffLoader to the ClassAssigner: first right click over the ArffLoader and select the dataSet under Connections in the menu. A rubber band line will appear. Move the mouse over the ClassAssigner component and left click - a red line labeled dataSet will connect the two components.
• Next right click over the ClassAssigner and choose Configure from the menu. This will pop up a window from which you can specify which column is the class in your data (last is the default).
• Next choose the ClassValuePicker (allows you to choose which class label to be evaluated in the ROC) component from the toolbar. Place this on the layout and right click over ClassAssigner and select dataSet from under Connections in the menu and connect it with the ClassValuePicker.
• Next grab a CrossValidationFoldMaker component from the Evaluation toolbar and place it on the layout. Connect the ClassAssigner to the CrossValidationFoldMaker by right clicking over ClassAssigner and se- lecting dataSet from under Connections in the menu.
104 CHAPTER 6. KNOWLEDGEFLOW
• Next click on the Classifiers tab at the top of the window and scroll along the toolbar until you reach the J48 component in the trees section. Place a J48 component on the layout.
• Connect the CrossValidationFoldMaker to J48 TWICE by first choosing trainingSet and then testSet from the pop-up menu for the CrossValida- tionFoldMaker.
• Repeat these two steps with the RandomForest classifier.
• Next go back to the Evaluation tab and place a ClassifierPerformanceE- valuator component on the layout. Connect J48 to this component by selecting the batchClassifier entry from the pop-up menu for J48. Add another ClassifierPerformanceEvaluator for RandomForest and connect them via batchClassifier as well.
• Next go to the Visualization toolbar and place a ModelPerformanceChart component on the layout. Connect both ClassifierPerformanceEvaluators to the ModelPerformanceChart by selecting the thresholdData entry from the pop-up menu for ClassifierPerformanceEvaluator.
• Now start the flow executing by selecting Start loading from the pop-up menu for ArffLoader. Depending on how big the data set is and how long cross validation takes you will see some animation from some of the icons in the layout. You will also see some progress information in the Status bar and Log at the bottom of the window.
• Select Show plot from the popup-menu of the ModelPerformanceChart under the Actions section.
Here are the two ROC curves generated from the UCI dataset credit-g, eval- uated on the class label good:
6.4. EXAMPLES 105 6.4.3 Processing data incrementally
Some classifiers, clusterers and filters in Weka can handle data incrementally in a streaming fashion. Here is an example of training and testing naive Bayes incrementally. The results are sent to a TextViewer and predictions are plotted by a StripChart component.
• Click on the DataSources tab and choose ArffLoader from the toolbar (the mouse pointer will change to a cross hairs).
• Next place the ArffLoader component on the layout area by clicking some- where on the layout (a copy of the ArffLoader icon will appear on the layout area).
• Next specify an ARFF file to load by first right clicking the mouse over the ArffLoader icon on the layout. A pop-up menu will appear. Select Configure under Edit in the list from this menu and browse to the location of your ARFF file.
• Next click the Evaluation tab at the top of the window and choose the ClassAssigner (allows you to choose which column to be the class) com- ponent from the toolbar. Place this on the layout.
• Now connect the ArffLoader to the ClassAssigner: first right click over the ArffLoader and select the dataSet under Connections in the menu. A rubber band line will appear. Move the mouse over the ClassAssigner component and left click - a red line labeled dataSet will connect the two components.
• Next right click over the ClassAssigner and choose Configure from the menu. This will pop up a window from which you can specify which column is the class in your data (last is the default).
• Now grab a NaiveBayesUpdateable component from the bayes section of the Classifiers panel and place it on the layout.
• Next connect the ClassAssigner to NaiveBayesUpdateable using a instance connection.
• Next place an IncrementalClassiferEvaluator from the Evaluation panel onto the layout and connect NaiveBayesUpdateable to it using a incre- mentalClassifier connection.
106 CHAPTER 6. KNOWLEDGEFLOW
• Next place a TextViewer component from the Visualization panel on the Layout. Connect the IncrementalClassifierEvaluator to it using a text connection.
• Next place a StripChart component from the Visualization panel on the layout and connect IncrementalClassifierEvaluator to it using a chart con- nection.
• Display the StripChart’s chart by right-clicking over it and choosing Show chart from the pop-up menu. Note: the StripChart can be configured with options that control how often data points and labels are displayed.
• Finally, start the flow by right-clicking over the ArffLoader and selecting Start loading from the pop-up menu.
Note that, in this example, a prediction is obtained from naive Bayes for each incoming instance before the classifier is trained (updated) with the instance. If you have a pre-trained classifier, you can specify that the classifier not be updated on incoming instances by unselecting the check box in the configuration dialog for the classifier. If the pre-trained classifier is a batch classifier (i.e. it is not capable of incremental training) then you will only be able to test it in an incremental fashion.
6.5. PLUGIN FACILITY 107 6.5 Plugin Facility
The KnowledgeFlow offers the ability to easily add new components via a plugin mechanism. Plugins are installed in a directory called .knowledgeFlow/plugins in the user’s home directory. If this directory does not exist you must create it in order to install plugins. Plugins are installed in subdirectories of the .knowledgeFlow/plugins directory. More than one plugin component may reside in the same subdirectory. Each subdirectory should contain jar file(s) that contain and support the plugin components. The KnowledgeFlow will dy- namically load jar files and add them to the classpath. In order to tell the KnowledgeFlow which classes in the jar files to instantiate as components, a second file called Beans.props needs to be created and placed into each plu- gin subdirectory. This file contains a list of fully qualified class names to be instantiated. Successfully instantiated components will appear in a “Plugins” tab in the KnowledgeFlow user interface.Below is an example plugin directory listing, the listing of the contents of the jar file and the contents of the associated Beans.props file:
cygnus:~ mhall$ ls -l $HOME/.knowledgeFlow/plugins/kettle/ total 24
-rw-r--r-- 1 mhall mhall 117 20 Feb 10:56 Beans.props -rw-r--r-- 1 mhall mhall 8047 20 Feb 14:01 kettleKF.jar
cygnus:~ mhall$ jar tvf /Users/mhall/.knowledgeFlow/plugins/kettle/kettleKF.jar 0 Wed Feb 20 14:01:34 NZDT 2008 META-INF/
70 Wed Feb 20 14:01:34 NZDT 2008 META-INF/MANIFEST.MF 0 Tue Feb 19 14:59:08 NZDT 2008 weka/
0 Tue Feb 19 14:59:08 NZDT 2008 weka/gui/
0 Wed Feb 20 13:55:52 NZDT 2008 weka/gui/beans/
0 Wed Feb 20 13:56:36 NZDT 2008 weka/gui/beans/icons/
2812 Wed Feb 20 14:01:20 NZDT 2008 weka/gui/beans/icons/KettleInput.gif
2812 Wed Feb 20 14:01:18 NZDT 2008 weka/gui/beans/icons/KettleInput_animated.gif 1839 Wed Feb 20 13:59:08 NZDT 2008 weka/gui/beans/KettleInput.class
174 Tue Feb 19 15:27:24 NZDT 2008 weka/gui/beans/KettleInputBeanInfo.class
cygnus:~ mhall$ more /Users/mhall/.knowledgeFlow/plugins/kettle/Beans.props # Specifies the tools to go into the Plugins toolbar weka.gui.beans.KnowledgeFlow.Plugins=weka.gui.beans.KettleInput
108 CHAPTER 6. KNOWLEDGEFLOW
Chapter 7 ArffViewer
The ArffViewer is a little tool for viewing ARFF files in a tabular format. The advantage of this kind of display over the file representation is, that attribute name, type and data are directly associated in columns and not separated in defintion and data part. But the viewer is not only limited to viewing multiple files at once, but also provides simple editing functionality, like sorting and deleting.
109
110 CHAPTER 7. ARFFVIEWER 7.1 Menus
The ArffViewer offers most of its functionality either through the main menu or via popups (table header and table cells).
Short description of the available menus:
• File
contains options for opening and closing files, as well as viewing properties about the current file.
• Edit
allows one to delete attributes/instances, rename attributes, choose a new class attribute, search for certain values in the data and of course undo the modifications.
• View
brings either the chosen attribute into view or displays all the values of an attribute.
After opening a file, by default, the column widths are optimized based on the attribute name and not the content. This is to ensure that overlong cells do not force an enormously wide table, which the user has to reduce with quite some effort.
7.1. MENUS 111 In the following, screenshots of the table popups:
112 CHAPTER 7. ARFFVIEWER 7.2 Editing
Besides the first column, which is the instance index, all cells in the table are editable. Nominal values can be easily modified via dropdown lists, numeric values are edited directly.
7.2. EDITING 113
For convenience, it is possible to sort the view based on a column (the underlying data is NOT changed; via Edit/Sort data one can sort the data permanently). This enables one to look for specific values, e.g., missing values. To better distinguish missing values from empty cells, the background of cells with missing values is colored grey.
114 CHAPTER 7. ARFFVIEWER
Chapter 8
Bayesian Network Classifiers
8.1 Introduction
Let U = {x1,...,xn}, n ≥ 1 be a set of variables. A Bayesian network B over a set of variables U is a network structure BS, which is a directed acyclic graph (DAG) over U and a set of probability tables BP = {p(u|pa(u))|u ∈ U} where pa(u) is the set of parents of u in BS. A Bayesian network represents a probability distributions P (U ) = u∈U p(u|pa(u)).
Below, a Bayesian network is shown for the variables in the iris data set. Note that the links between the nodes class, petallength and petalwidth do not form a directed cycle, so the graph is a proper DAG.
This picture just shows the network structure of the Bayes net, but for each of the nodes a probability distribution for the node given its parents are specified as well. For example, in the Bayes net above there is a conditional distribution
115
116 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS for petallength given the value of class. Since class has no parents, there is an
unconditional distribution for sepalwidth.
Basic assumptions
The classification task consist of classifying a variable y = x0 called the class variable given a set of variables x = x1 . . . xn , called attribute variables. A classifier h : x → y is a function that maps an instance of x to a value of y. The classifier is learned from a dataset D consisting of samples over (x, y). The learning task consists of finding an appropriate Bayesian network given a data set D over U.
All Bayes network algorithms implemented in Weka assume the following for the data set:
• all variables are discrete finite variables. If you have a data set with continuous variables, you can use the following filter to discretize them: weka.filters.unsupervised.attribute.Discretize
• no instances have missing values. If there are missing values in the data set, values are filled in using the following filter: weka.filters.unsupervised.attribute.ReplaceMissingValues
The first step performed by buildClassifier is checking if the data set fulfills those assumptions. If those assumptions are not met, the data set is automatically filtered and a warning is written to STDERR.1
Inference algorithm
To use a Bayesian network as a classifier, one simply calculates argmaxyP(y|x) using the distribution P(U) represented by the Bayesian network. Now note that
P(y|x) = P(U)/P(x) ∝ P(U)
= p(u|pa(u)) (8.1) u∈U
And since all variables in x are known, we do not need complicated inference algorithms, but just calculate (8.1) for all class values.
Learning algorithms
The dual nature of a Bayesian network makes learning a Bayesian network as a two stage process a natural division: first learn a network structure, then learn the probability tables.
There are various approaches to structure learning and in Weka, the following areas are distinguished:
1If there are missing values in the test data, but not in the training data, the values are filled in in the test data with a ReplaceMissingValues filter based on the training data.
8.1. INTRODUCTION 117
• local score metrics: Learning a network structure BS can be considered an optimization problem where a quality measure of a network structure given the training data Q(BS |D) needs to be maximized. The quality mea- sure can be based on a Bayesian approach, minimum description length, information and other criteria. Those metrics have the practical property that the score of the whole network can be decomposed as the sum (or product) of the score of the individual nodes. This allows for local scoring and thus local search methods.
• conditional independence tests: These methods mainly stem from the goal of uncovering causal structure. The assumption is that there is a network structure that exactly represents the independencies in the distribution that generated the data. Then it follows that if a (conditional) indepen- dency can be identified in the data between two variables that there is no arrow between those two variables. Once locations of edges are identified, the direction of the edges is assigned such that conditional independencies in the data are properly represented.
• global score metrics: A natural way to measure how well a Bayesian net- work performs on a given data set is to predict its future performance by estimating expected utilities, such as classification accuracy. Cross- validation provides an out of sample evaluation method to facilitate this by repeatedly splitting the data in training and validation sets. A Bayesian network structure can be evaluated by estimating the network’s param- eters from the training set and the resulting Bayesian network’s perfor- mance determined against the validation set. The average performance of the Bayesian network over the validation sets provides a metric for the quality of the network.
Cross-validation differs from local scoring metrics in that the quality of a network structure often cannot be decomposed in the scores of the indi- vidual nodes. So, the whole network needs to be considered in order to determine the score.
• fixed structure: Finally, there are a few methods so that a structure can be fixed, for example, by reading it from an XML BIF file2.
For each of these areas, different search algorithms are implemented in Weka, such as hill climbing, simulated annealing and tabu search.
Once a good network structure is identified, the conditional probability ta- bles for each of the variables can be estimated.
You can select a Bayes net classifier by clicking the classifier ’Choose’ button in the Weka explorer, experimenter or knowledge flow and find BayesNet under the weka.classifiers.bayes package (see below).
2 See http://www.cs.cmu.edu/∼ fgcozman/Research/InterchangeFormat/ for details on XML BIF.
118 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
The Bayes net classifier has the following options:
The BIFFile option can be used to specify a Bayes network stored in file in BIF format. When the toString() method is called after learning the Bayes network, extra statistics (like extra and missing arcs) are printed comparing the network learned with the one on file.
The searchAlgorithm option can be used to select a structure learning algorithm and specify its options.
The estimator option can be used to select the method for estimating the conditional probability distributions (Section 8.6).
When setting the useADTree option to true, counts are calculated using the ADTree algorithm of Moore [24]. Since I have not noticed a lot of improvement for small data sets, it is set off by default. Note that this ADTree algorithm is dif- ferent from the ADTree classifier algorithm from weka.classifiers.tree.ADTree.
The debug option has no effect.
8.2. LOCAL SCORE BASED STRUCTURE LEARNING 119 8.2 Local score based structure learning
Distinguish score metrics (Section 2.1) and search algorithms (Section 2.2). A local score based structure learning can be selected by choosing one in the weka.classifiers.bayes.net.search.local package.
Local score based algorithms have the following options in common: initAsNaiveBayes if set true (default), the initial network structure used for starting the traversal of the search space is a naive Bayes network structure. That is, a structure with arrows from the class variable to each of the attribute variables.
If set false, an empty network structure will be used (i.e., no arrows at all). markovBlanketClassifier (false by default) if set true, at the end of the traversal of the search space, a heuristic is used to ensure each of the attributes are in the Markov blanket of the classifier node. If a node is already in the Markov blanket (i.e., is a parent, child of sibling of the classifier node) nothing happens, otherwise an arrow is added.
If set to false no such arrows are added.
scoreType determines the score metric used (see Section 2.1 for details). Cur- rently, K2, BDe, AIC, Entropy and MDL are implemented.
maxNrOfParents is an upper bound on the number of parents of each of the nodes in the network structure learned.
8.2.1 Local score metrics
We use the following conventions to identify counts in the database D and a network structure BS. Let ri (1 ≤ i ≤ n) be the cardinality of xi. We use qi to denote the cardinality of the parent set of xi in BS, that is, the number of different values to which the parents of xi can be instantiated. So, qi can be calculated as the product of cardinalities of nodes in pa(xi), qi = xj∈pa(xi) rj.
120 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
Notepa(xi)=∅impliesqi =1. WeuseNij (1≤i≤n,1≤j≤qi)todenote the number of records in D for which pa(xi) takes its jth value.We use Nijk (1 ≤ i ≤ n, 1 ≤ j ≤ qi, 1 ≤ k ≤ ri) to denote the number of records in D for which pa(xi) takes its jth value and for which xi takes its kth value. So, Nij = ri Nijk. We use N to denote the number of records in D.
k=1
Let the entropy metric H(BS,D) of a network structure and database be
defined as
and the number of parameters K as
n qi ri
H(BS,D) = −N Nijk log Nijk
(8.2)
i=1 j=1 k=1 N Nij
n i=1
(ri −1)·qi
AIC metric The AIC metric QAIC(BS,D) of a Bayesian network structure
BS for a database D is
QAIC(BS,D) = H(BS,D) + K (8.4)
A term P(BS) can be added [15] representing prior information over network structures, but will be ignored for simplicity in the Weka implementation.
MDL metric The minimum description length metric QMDL(BS,D) of a Bayesian network structure BS for a database D is is defined as
QMDL(BS,D) = H(BS,D)+ K logN (8.5) 2
Bayesian metric The Bayesian metric of a Bayesian network structure BD for a database D is
i=0 j=1 ij
where P(BS) is the prior on the network structure (taken to be constant hence
K =
(8.3)
QBayes(BS,D)=P(BS) Γ(Nij) Γ(N′ +Nij)
Γ(Nijk +Nijk) Γ(N′ )
nqi ′ ri ′
k=1 ijk
ignored in the Weka implementation) and Γ(.) the gamma-function. N′ and
ij
N′ represent choices of priors on counts restricted by N′ = ri N′ . With
ijk ij k=1 ijk N′ =1(andthusN′ =r),weobtaintheK2metric[19]
ijk
With N′ ijk
ij i
QK2(BS,D)=P(BS) (ri −1)! Nijk!
n qi ri i=0j=1 (ri −1+Nij)!k=1
= 1/ri · qi (and thus N′ = 1/qi), we obtain the BDe metric [22]. ij
8.2.2 Search algorithms
The following search algorithms are implemented for local score metrics;
• K2 [19]: hill climbing add arcs with a fixed ordering of variables. Specific option: randomOrder if true a random ordering of the nodes is made at the beginning of the search. If false (default) the ordering in the data set is used. The only exception in both cases is that in case the initial network is a naive Bayes network (initAsNaiveBayes set true) the class variable is made first in the ordering.
8.2. LOCAL SCORE BASED STRUCTURE LEARNING 121
• Hill Climbing [16]: hill climbing adding and deleting arcs with no fixed ordering of variables.
useArcReversal if true, also arc reversals are consider when determining the next step to make.
• Repeated Hill Climber starts with a randomly generated network and then applies hill climber to reach a local optimum. The best network found is returned.
useArcReversal option as for Hill Climber.
• LAGD Hill Climbing does hill climbing with look ahead on a limited set of best scoring steps, implemented by Manuel Neubach. The number of look ahead steps and number of steps considered for look ahead are configurable.
• TAN [17, 21]: Tree Augmented Naive Bayes where the tree is formed by calculating the maximum weight spanning tree using Chow and Liu algorithm [18].
No specific options.
• Simulated annealing [15]: using adding and deleting arrows.
The algorithm randomly generates a candidate network BS′ close to the current network BS . It accepts the network if it is better than the current,
i.e., Q(BS′ , D) > Q(BS , D). Otherwise, it accepts the candidate with probability
eti·(Q(BS′ ,D)−Q(BS,D))
where ti is the temperature at iteration i. The temperature starts at t0
and is slowly decreases with each iteration.
Specific options:
TStart start temperature t0.
delta is the factor δ used to update the temperature, so ti+1 = ti · δ. runs number of iterations used to traverse the search space.
seed is the initialization value for the random number generator.
• Tabu search [15]: using adding and deleting arrows.
Tabu search performs hill climbing until it hits a local optimum. Then it
122
CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
steps to the least worse candidate in the neighborhood. However, it does not consider points in the neighborhood it just visited in the last tl steps. These steps are stored in a so called tabu-list.
Specific options:
runs is the number of iterations used to traverse the search space. tabuList is the length tl of the tabu list.
• Genetic search: applies a simple implementation of a genetic search algo- rithm to network structure learning. A Bayes net structure is represented by a array of n·n (n = number of nodes) bits where bit i·n+j represents whether there is an arrow from node j → i.
Specific options:
populationSize is the size of the population selected in each generation. descendantPopulationSize is the number of offspring generated in each
8.3.
CONDITIONAL INDEPENDENCE TEST BASED STRUCTURE LEARNING123
8.3
generation.
runs is the number of generation to generate.
seed is the initialization value for the random number generator. useMutation flag to indicate whether mutation should be used. Mutation is applied by randomly adding or deleting a single arc.
useCrossOver flag to indicate whether cross-over should be used. Cross- over is applied by randomly picking an index k in the bit representation and selecting the first k bits from one and the remainder from another network structure in the population. At least one of useMutation and useCrossOver should be set to true.
useTournamentSelection when false, the best performing networks are selected from the descendant population to form the population of the next generation. When true, tournament selection is used. Tournament selection randomly chooses two individuals from the descendant popula- tion and selects the one that performs best.
Conditional independence test based struc- ture learning
Conditional independence tests in Weka are slightly different from the standard tests described in the literature. To test whether variables x and y are condi- tionally independent given a set of variables Z, a network structure with arrows ∀z∈Zz → y is compared with one with arrows {x → y} ∪ ∀z∈Zz → y. A test is performed by using any of the score metrics described in Section 2.1.
At the moment, only the ICS [25]and CI algorithm are implemented.
The ICS algorithm makes two steps, first find a skeleton (the undirected graph with edges iff there is an arrow in network structure) and second direct
124 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
all the edges in the skeleton to get a DAG.
Starting with a complete undirected graph, we try to find conditional inde-
pendencies ⟨x,y|Z⟩ in the data. For each pair of nodes x, y, we consider sets Z starting with cardinality 0, then 1 up to a user defined maximum. Further- more, the set Z is a subset of nodes that are neighbors of both x and y. If an independency is identified, the edge between x and y is removed from the skeleton.
The first step in directing arrows is to check for every configuration x − −z − −y where x and y not connected in the skeleton whether z is in the set Z of variables that justified removing the link between x and y (cached in the first step). If z is not in Z, we can assign direction x → z ← y.
Finally, a set of graphical rules is applied [25] to direct the remaining arrows.
Rule 1: i->j–k & i-/-k => j->k Rule 2: i->j->k & i–k => i->k Rule3 m
/|\
i|k =>m->j
i->j<-k \|/ j
Rule4 m
/\
i---k => i->m & k->m
i->j \/ j
Rule 5: if no edges are directed then take a random one (first we can find)
The ICS algorithm comes with the following options.
Since the ICS algorithm is focused on recovering causal structure, instead of finding the optimal classifier, the Markov blanket correction can be made afterwards.
Specific options:
The maxCardinality option determines the largest subset of Z to be considered in conditional independence tests ⟨x, y|Z⟩.
The scoreType option is used to select the scoring metric.
8.4. GLOBAL SCORE METRIC BASED STRUCTURE LEARNING 125 8.4 Global score metric based structure learning
Common options for cross-validation based algorithms are: initAsNaiveBayes, markovBlanketClassifier and maxNrOfParents (see Sec- tion 8.2 for description).
Further, for each of the cross-validation based algorithms the CVType can be chosen out of the following:
• Leave one out cross-validation (loo-cv) selects m = N training sets simply by taking the data set D and removing the ith record for training set Dit. The validation set consist of just the ith single record. Loo-cv does not always produce accurate performance estimates.
• K-fold cross-validation (k-fold cv) splits the data D in m approximately equal parts D1, . . . , Dm. Training set Dit is obtained by removing part Di from D. Typical values for m are 5, 10 and 20. With m = N, k-fold cross-validation becomes loo-cv.
• Cumulative cross-validation (cumulative cv) starts with an empty data set and adds instances item by item from D. After each time an item is added the next item to be added is classified using the then current state of the Bayes network.
Finally, the useProb flag indicates whether the accuracy of the classifier should be estimated using the zero-one loss (if set to false) or using the esti- mated probability of the class.
126 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
The following search algorithms are implemented: K2, HillClimbing, Repeat- edHillClimber, TAN, Tabu Search, Simulated Annealing and Genetic Search. See Section 8.2 for a description of the specific options for those algorithms.
8.5 Fixed structure ’learning’
The structure learning step can be skipped by selecting a fixed network struc- ture. There are two methods of getting a fixed structure: just make it a naive Bayes network, or reading it from a file in XML BIF format.
8.6 Distribution learning
Once the network structure is learned, you can choose how to learn the prob- ability tables selecting a class in the weka.classifiers.bayes.net.estimate
8.6. DISTRIBUTION LEARNING 127 package.
The SimpleEstimator class produces direct estimates of the conditional probabilities, that is,
Nijk +N′ P (xi = k|pa(xi) = j) = ijk
where N′ ijk
is the alpha parameter that can be set and is 0.5 by default. With alpha = 0, we get maximum likelihood estimates.
With the BMAEstimator, we get estimates for the conditional probability tables based on Bayes model averaging of all network structures that are sub- structures of the network structure learned [15]. This is achieved by estimat- ing the conditional probability table of a node xi given its parents pa(xi) as a weighted average of all conditional probability tables of xi given subsets of pa(xi). The weight of a distribution P(xi|S) with S ⊆ pa(xi) used is propor- tional to the contribution of network structure ∀y∈Sy → xi to either the BDe metric or K2 metric depending on the setting of the useK2Prior option (false and true respectively).
Nij +N′ ij
128 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
8.7 Running from the command line
These are the command line options of BayesNet.
General options:
-t
-T
Sets test file. If missing, a cross-validation will be performed on the training data.
-c
Sets index of class attribute (default: last).
-x
Sets number of folds for cross-validation (default: 10).
-no-cv
Do not perform any cross validation.
-split-percentage
-preserve-order
Preserves the order in the percentage split.
-s
Sets random number seed for cross-validation or percentage split (default: 1).
-m
-l
Sets model input file. In case the filename ends with ’.xml’, the options are loaded from the XML file.
-d
Sets model output file. In case the filename ends with ’.xml’, only the options are saved to the XML file, not the model.
-v -o -i -k
Outputs no statistics for training data.
Outputs statistics only, not the classifier.
Outputs detailed information-retrieval statistics for each class.
8.7. RUNNING FROM THE COMMAND LINE 129
Outputs information-theoretic statistics. -p
Only outputs predictions for test instances (or the train instances if no test instances provided), along with attributes (0 for none).
-distribution
Outputs the distribution instead of only the prediction
in conjunction with the ’-p’ option (only nominal classes).
-r
-g
-xml filename | xml-string
-M
Only outputs cumulative margin distribution.
Only outputs the graph representation of the classifier.
Retrieves the options from the XML-data instead of the command line.
Options specific to weka.classifiers.bayes.BayesNet: -D
Do not use ADTree data structure
-B
BIF file to compare with
-Q weka.classifiers.bayes.net.search.SearchAlgorithm Search algorithm
-E weka.classifiers.bayes.net.estimate.SimpleEstimator Estimator algorithm
The search algorithm option -Q and estimator option -E options are manda- tory.
Note that it is important that the -E options should be used after the -Q option. Extra options can be passed to the search algorithm and the estimator after the class name specified following ’–’.
For example:
java weka.classifiers.bayes.BayesNet -t iris.arff -D \
-Q weka.classifiers.bayes.net.search.local.K2 — -P 2 -S ENTROPY \ -E weka.classifiers.bayes.net.estimate.SimpleEstimator — -A 1.0
Overview of options for search algorithms
• weka.classifiers.bayes.net.search.local.GeneticSearch
-L
-A
Descendant population size
-U
Number of runs
Use mutation.
130
CHAPTER 8.
BAYESIAN NETWORK CLASSIFIERS
•
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
weka.classifiers.bayes.net.search.local.HillClimber
-P
Maximum number of parents
•
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
weka.classifiers.bayes.net.search.local.K2
-C
-O
-R
-mbc
Random number seed
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-R
-N -mbc
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-N
(default true)
Use cross-over. (default true)
Use tournament selection (true) or maximum subpopulatin (false). (default false)
Initial structure is empty (instead of Naive Bayes) Maximum number of parents
Random order. (default false)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-P
-R -mbc
8.7. RUNNING FROM THE COMMAND LINE 131 -S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES]
Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.local.LAGDHillClimber
-L
-G
-P
Maximum number of parents
-R
-N -mbc
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.local.RepeatedHillClimber
-U
Number of runs
-A
Random number seed
-P
Maximum number of parents
-R
-N -mbc
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.local.SimulatedAnnealing
132
CHAPTER 8.
BAYESIAN NETWORK CLASSIFIERS
•
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
weka.classifiers.bayes.net.search.local.TabuSearch
-L
Tabu list length
-U
Number of runs
-P
Maximum number of parents
-R
Use arc reversal operation.
(default false) -P
•
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
weka.classifiers.bayes.net.search.local.TAN -mbc
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the
-A
Start temperature
-U
Number of runs
-D
Delta temperature
-R
Random number seed
-mbc
-R
-N -mbc
Maximum number of parents
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
8.7. RUNNING FROM THE COMMAND LINE 133
classifier node.
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES]
Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.ci.CISearchAlgorithm -mbc
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.ci.ICSSearchAlgorithm
-cardinality
When determining whether an edge exists a search is performed for a set Z that separates the nodes. MaxCardinality determines the maximum size of the set Z. This greatly influences the length of the search. (default 2)
-mbc
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [BAYES|MDL|ENTROPY|AIC|CROSS_CLASSIC|CROSS_BAYES] Score type (BAYES, BDeu, MDL, ENTROPY and AIC)
• weka.classifiers.bayes.net.search.global.GeneticSearch
-L
-A
Descendant population size
-U
Number of runs
-M
-C
-O
-R
Use mutation. (default true)
Use cross-over. (default true)
Use tournament selection (true) or maximum subpopulatin (false). (default false)
134
CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
•
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
weka.classifiers.bayes.net.search.global.HillClimber
-P
Maximum number of parents
•
weka.classifiers.bayes.net.search.global.K2
-mbc
Random number seed
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-R
-N -mbc
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
-N
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
Initial structure is empty (instead Maximum number of parents
Random order. (default false)
Applies a Markov Blanket correction
after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-P
-R -mbc
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
of Naive Bayes)
to the network structure,
8.7. RUNNING FROM THE COMMAND LINE 135 -Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
• weka.classifiers.bayes.net.search.global.RepeatedHillClimber
-U
Number of runs
-A
Random number seed
-P
Maximum number of parents
-R
-N -mbc
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
• weka.classifiers.bayes.net.search.global.SimulatedAnnealing
-A
Start temperature
-U
Number of runs
-D
Delta temperature
-R
Random number seed
-mbc
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
136 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS • weka.classifiers.bayes.net.search.global.TabuSearch
-L
Tabu list length
-U
Number of runs
-P
Maximum number of parents
-R
Use arc reversal operation.
(default false) -P
-R
-N -mbc
Maximum number of parents
Use arc reversal operation. (default false)
Initial structure is empty (instead of Naive Bayes)
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
• weka.classifiers.bayes.net.search.global.TAN -mbc
Applies a Markov Blanket correction to the network structure, after a network structure is learned. This ensures that all nodes in the network are part of the Markov blanket of the classifier node.
-S [LOO-CV|k-Fold-CV|Cumulative-CV]
Score type (LOO-CV,k-Fold-CV,Cumulative-CV)
-Q
Use probabilistic or 0/1 scoring. (default probabilistic scoring)
• weka.classifiers.bayes.net.search.fixed.FromFile -B
Name of file containing network structure in BIF format
• weka.classifiers.bayes.net.search.fixed.NaiveBayes
8.7. RUNNING FROM THE COMMAND LINE 137 No options.
Overview of options for estimators
• weka.classifiers.bayes.net.estimate.BayesNetEstimator -A
Initial count (alpha)
• weka.classifiers.bayes.net.estimate.BMAEstimator -k2
Whether to use K2 prior. -A
Initial count (alpha)
• weka.classifiers.bayes.net.estimate.MultiNomialBMAEstimator -k2
Whether to use K2 prior. -A
Initial count (alpha)
• weka.classifiers.bayes.net.estimate.SimpleEstimator -A
Initial count (alpha)
Generating random networks and artificial data sets
You can generate random Bayes nets and data sets using
weka.classifiers.bayes.net.BayesNetGenerator
The options are:
-B
Generate network (instead of instances) -N
Nr of nodes
-A
Nr of arcs
-M
Nr of instances
-C
Cardinality of the variables -S
Seed for random number generator -F
The BIF file to obtain the structure from.
138 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
The network structure is generated by first generating a tree so that we can ensure that we have a connected graph. If any more arrows are specified they are randomly added.
8.8 Inspecting Bayesian networks
You can inspect some of the properties of Bayesian networks that you learned in the Explorer in text format and also in graphical format.
Bayesian networks in text
Below, you find output typical for a 10 fold cross-validation run in the Weka Explorer with comments where the output is specific for Bayesian nets.
=== Run information ===
Scheme: weka.classifiers.bayes.BayesNet -D -B iris.xml -Q weka.classifiers.bayes.n
Options for BayesNet include the class names for the structure learner and for the distribution estimator.
Relation:
Instances:
Attributes:
iris-weka.filters.unsupervised.attribute.Discretize-B2-M-1.0-Rfirst-last 150
5
sepallength
sepalwidth
petallength
petalwidth
class
10-fold cross-validation
Test mode:
=== Classifier model (full training set) ===
Bayes Network Classifier not using ADTree
Indication whether the ADTree algorithm [24] for calculating counts in the data set was used.
#attributes=5 #classindex=4
This line lists the number of attribute and the number of the class variable for which the classifier was trained.
Network structure (nodes followed by parents) sepallength(2): class
sepalwidth(2): class
petallength(2): class sepallength petalwidth(2): class petallength
class(3):
e
8.8. INSPECTING BAYESIAN NETWORKS 139
This list specifies the network structure. Each of the variables is followed by a list of parents, so the petallength variable has parents sepallength and class, while class has no parents. The number in braces is the cardinality of the variable. It shows that in the iris dataset there are three class variables. All other variables are made binary by running it through a discretization filter.
LogScore Bayes: -374.9942769685747 LogScore BDeu: -351.85811477631626 LogScore MDL: -416.86897021246466 LogScore ENTROPY: -366.76261727150217 LogScore AIC: -386.76261727150217
These lines list the logarithmic score of the network structure for various meth- ods of scoring.
If a BIF file was specified, the following two lines will be produced (if no such file was specified, no information is printed).
Missing: 0 Extra: 2 Reversed: 0 Divergence: -0.0719759699700729
In this case the network that was learned was compared with a file iris.xml which contained the naive Bayes network structure. The number after “Missing” is the number of arcs that was in the network in file that is not recovered by the structure learner. Note that a reversed arc is not counted as missing. The number after “Extra” is the number of arcs in the learned network that are not in the network on file. The number of reversed arcs is listed as well.
Finally, the divergence between the network distribution on file and the one learned is reported. This number is calculated by enumerating all possible in- stantiations of all variables, so it may take some time to calculate the divergence for large networks.
The remainder of the output is standard output for all classifiers.
Time taken to build model: 0.01 seconds
=== Stratified cross-validation === === Summary ===
Correctly Classified Instances 116 Incorrectly Classified Instances 34
etc…
Bayesian networks in GUI
77.3333 %
22.6667 %
To show the graphical structure, right click the appropriate BayesNet in result list of the Explorer. A menu pops up, in which you select “Visualize graph”.
140 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
The Bayes network is automatically layed out and drawn thanks to a graph drawing algorithm implemented by Ashraf Kibriya.
When you hover the mouse over a node, the node lights up and all its children are highlighted as well, so that it is easy to identify the relation between nodes in crowded graphs.
Saving Bayes nets You can save the Bayes network to file in the graph visualizer. You have the choice to save as XML BIF format or as dot format. Select the floppy button and a file save dialog pops up that allows you to select the file name and file format.
Zoom The graph visualizer has two buttons to zoom in and out. Also, the exact zoom desired can be entered in the zoom percentage entry. Hit enter to redraw at the desired zoom level.
8.9. BAYES NETWORK GUI 141 Graph drawing options Hit the ’extra controls’ button to show extra
options that control the graph layout settings.
The Layout Type determines the algorithm applied to place the nodes. The Layout Method determines in which direction nodes are considered. The Edge Concentration toggle allows edges to be partially merged.
The Custom Node Size can be used to override the automatically deter-
mined node size.
When you click a node in the Bayesian net, a window with the probability
table of the node clicked pops up. The left side shows the parent attributes and lists the values of the parents, the right side shows the probability of the node clicked conditioned on the values of the parents listed on the left.
So, the graph visualizer allows you to inspect both network structure and probability tables.
8.9 Bayes Network GUI
The Bayesian network editor is a stand alone application with the following features
• Edit Bayesian network completely by hand, with unlimited undo/redo stack, cut/copy/paste and layout support.
• Learn Bayesian network from data using learning algorithms in Weka.
• Edit structure by hand and learn conditional probability tables (CPTs) using
142 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
learning algorithms in Weka.
• Generate dataset from Bayesian network.
• Inference (using junction tree method) of evidence through the network, in- teractively changing values of nodes.
• Viewing cliques in junction tree.
• Accelerator key support for most common operations.
The Bayes network GUI is started as
java weka.classifiers.bayes.net.GUI ¡bif file¿
The following window pops up when an XML BIF file is specified (if none is specified an empty graph is shown).
Moving a node
Click a node with the left mouse button and drag the node to the desired position.
8.9. BAYES NETWORK GUI 143
Selecting groups of nodes
Drag the left mouse button in the graph panel. A rectangle is shown and all nodes intersecting with the rectangle are selected when the mouse is released. Selected nodes are made visible with four little black squares at the corners (see screenshot above).
The selection can be extended by keeping the shift key pressed while selecting another set of nodes.
The selection can be toggled by keeping the ctrl key pressed. All nodes in the selection selected in the rectangle are de-selected, while the ones not in the selection but intersecting with the rectangle are added to the selection.
Groups of nodes can be moved by keeping the left mouse pressed on one of the selected nodes and dragging the group to the desired position.
File menu
The New, Save, Save As, and Exit menu provide functionality as expected. The file format used is XML BIF [20].
There are two file formats supported for opening
• .xml for XML BIF files. The Bayesian network is reconstructed from the information in the file. Node width information is not stored so the nodes are
shown with the default width. This can be changed by laying out the graph
(menu Tools/Layout).
• .arff Weka data files. When an arff file is selected, a new empty Bayesian net-
work is created with nodes for each of the attributes in the arff file. Continuous variables are discretized using the weka.filters.supervised.attribute.Discretize filter (see note at end of this section for more details). The network structure
can be specified and the CPTs learned using the Tools/Learn CPT menu.
The Print menu works (sometimes) as expected.
The Export menu allows for writing the graph panel to image (currently supported are bmp, jpg, png and eps formats). This can also be activated using the Alt-Shift-Left Click action in the graph panel.
144 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS Edit menu
Unlimited undo/redo support. Most edit operations on the Bayesian network are undoable. A notable exception is learning of network and CPTs.
Cut/copy/paste support. When a set of nodes is selected these can be placed on a clipboard (internal, so no interaction with other applications yet) and a paste action will add the nodes. Nodes are renamed by adding ”Copy of” before the name and adding numbers if necessary to ensure uniqueness of name. Only the arrows to parents are copied, not these of the children.
The Add Node menu brings up a dialog (see below) that allows to specify the name of the new node and the cardinality of the new node. Node values are assigned the names ’Value1’, ’Value2’ etc. These values can be renamed (right click the node in the graph panel and select Rename Value). Another option is to copy/paste a node with values that are already properly named and rename the node.
The Add Arc menu brings up a dialog to choose a child node first;
8.9. BAYES NETWORK GUI 145
Then a dialog is shown to select a parent. Descendants of the child node, parents of the child node and the node itself are not listed since these cannot be selected as child node since they would introduce cycles or already have an arc in the network.
The Delete Arc menu brings up a dialog with a list of all arcs that can be deleted.
The list of eight items at the bottom are active only when a group of at least two nodes are selected.
• Align Left/Right/Top/Bottom moves the nodes in the selection such that all nodes align to the utmost left, right, top or bottom node in the selection re- spectively.
• Center Horizontal/Vertical moves nodes in the selection halfway between left and right most (or top and bottom most respectively).
• Space Horizontal/Vertical spaces out nodes in the selection evenly between left and right most (or top and bottom most respectively). The order in which the nodes are selected impacts the place the node is moved to.
Tools menu
The Generate Network menu allows generation of a complete random Bayesian network. It brings up a dialog to specify the number of nodes, number of arcs, cardinality and a random seed to generate a network.
146 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
The Generate Data menu allows for generating a data set from the Bayesian network in the editor. A dialog is shown to specify the number of instances to be generated, a random seed and the file to save the data set into. The file format is arff. When no file is selected (field left blank) no file is written and only the internal data set is set.
The Set Data menu sets the current data set. From this data set a new Bayesian network can be learned, or the CPTs of a network can be estimated. A file choose menu pops up to select the arff file containing the data.
The Learn Network and Learn CPT menus are only active when a data set is specified either through
• Tools/Set Data menu, or
• Tools/Generate Data menu, or
• File/Open menu when an arff file is selected.
The Learn Network action learns the whole Bayesian network from the data set. The learning algorithms can be selected from the set available in Weka by selecting the Options button in the dialog below. Learning a network clears the undo stack.
The Learn CPT menu does not change the structure of the Bayesian network, only the probability tables. Learning the CPTs clears the undo stack.
The Layout menu runs a graph layout algorithm on the network and tries to make the graph a bit more readable. When the menu item is selected, the node size can be specified or left to calculate by the algorithm based on the size of the labels by deselecting the custom node size check box.
8.9. BAYES NETWORK GUI 147
The Show Margins menu item makes marginal distributions visible. These are calculated using the junction tree algorithm [23]. Marginal probabilities for nodes are shown in green next to the node. The value of a node can be set (right click node, set evidence, select a value) and the color is changed to red to indicate evidence is set for the node. Rounding errors may occur in the marginal probabilities.
The Show Cliques menu item makes the cliques visible that are used by the junction tree algorithm. Cliques are visualized using colored undirected edges. Both margins and cliques can be shown at the same time, but that makes for rather crowded graphs.
148 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
View menu
The view menu allows for zooming in and out of the graph panel. Also, it allows for hiding or showing the status and toolbars.
Help menu
The help menu points to this document.
8.9. BAYES NETWORK GUI 149 Toolbar
The toolbar allows a shortcut to many functions. Just hover the mouse over the toolbar buttons and a tooltiptext pops up that tells which function is activated. The toolbar can be shown or hidden with the View/View Toolbar menu.
Statusbar
At the bottom of the screen the statusbar shows messages. This can be helpful when an undo/redo action is performed that does not have any visible effects, such as edit actions on a CPT. The statusbar can be shown or hidden with the View/View Statusbar menu.
Click right mouse button
Clicking the right mouse button in the graph panel outside a node brings up the following popup menu. It allows to add a node at the location that was clicked, or add select a parent to add to all nodes in the selection. If no node is selected, or no node can be added as parent, this function is disabled.
Clicking the right mouse button on a node brings up a popup menu.
The popup menu shows list of values that can be set as evidence to selected node. This is only visible when margins are shown (menu Tools/Show margins). By selecting ’Clear’, the value of the node is removed and the margins calculated based on CPTs again.
150 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS A node can be renamed by right click and select Rename in the popup menu.
The following dialog appears that allows entering a new node name.
The CPT of a node can be edited manually by selecting a node, right click/Edit CPT. A dialog is shown with a table representing the CPT. When a value is edited, the values of the remainder of the table are update in order to ensure that the probabilities add up to 1. It attempts to adjust the last column first, then goes backward from there.
The whole table can be filled with randomly generated distributions by selecting the Randomize button.
The popup menu shows list of parents that can be added to selected node. CPT for the node is updated by making copies for each value of the new parent.
8.9. BAYES NETWORK GUI 151
The popup menu shows list of parents that can be deleted from selected node. CPT of the node keeps only the one conditioned on the first value of the parent node.
The popup menu shows list of children that can be deleted from selected node. CPT of the child node keeps only the one conditioned on the first value of the parent node.
Selecting the Add Value from the popup menu brings up this dialog, in which the name of the new value for the node can be specified. The distribution for the node assign zero probability to the value. Child node CPTs are updated by copying distributions conditioned on the new value.
The popup menu shows list of values that can be renamed for selected node.
152 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
Selecting a value brings up the following dialog in which a new name can be specified.
The popup menu shows list of values that can be deleted from selected node. This is only active when there are more then two values for the node (single valued nodes do not make much sense). By selecting the value the CPT of the node is updated in order to ensure that the CPT adds up to unity. The CPTs of children are updated by dropping the distributions conditioned on the value.
A note on CPT learning
Continuous variables are discretized by the Bayes network class. The discretiza- tion algorithm chooses its values based on the information in the data set.
8.10. BAYESIAN NETS IN THE EXPERIMENTER 153
However, these values are not stored anywhere. So, reading an arff file with continuous variables using the File/Open menu allows one to specify a network, then learn the CPTs from it since the discretization bounds are still known. However, opening an arff file, specifying a structure, then closing the applica- tion, reopening and trying to learn the network from another file containing continuous variables may not give the desired result since a the discretization algorithm is re-applied and new boundaries may have been found. Unexpected behavior may be the result.
Learning from a dataset that contains more attributes than there are nodes in the network is ok. The extra attributes are just ignored.
Learning from a dataset with differently ordered attributes is ok. Attributes are matched to nodes based on name. However, attribute values are matched with node values based on the order of the values.
The attributes in the dataset should have the same number of values as the corresponding nodes in the network (see above for continuous variables).
8.10 Bayesian nets in the experimenter
Bayesian networks generate extra measures that can be examined in the exper- imenter. The experimenter can then be used to calculate mean and variance for those measures.
The following metrics are generated:
• measureExtraArcs: extra arcs compared to reference network. The net- work must be provided as BIFFile to the BayesNet class. If no such network is provided, this value is zero.
• measureMissingArcs: missing arcs compared to reference network or zero if not provided.
• measureReversedArcs: reversed arcs compared to reference network or zero if not provided.
• measureDivergence: divergence of network learned compared to reference network or zero if not provided.
• measureBayesScore: log of the K2 score of the network structure.
• measureBDeuScore: log of the BDeu score of the network structure.
• measureMDLScore: log of the MDL score.
• measureAICScore: log of the AIC score.
• measureEntropyScore:log of the entropy.
8.11 Adding your own Bayesian network learn- ers
You can add your own structure learners and estimators.
154 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS Adding a new structure learner
Here is the quick guide for adding a structure learner:
1. Createaclassthatderivesfromweka.classifiers.bayes.net.search.SearchAlgorithm. If your searcher is score based, conditional independence based or cross-
validation based, you probably want to derive from ScoreSearchAlgorithm, CISearchAlgorithm or CVSearchAlgorithm instead of deriving from SearchAlgorithm directly. Let’s say it is called
weka.classifiers.bayes.net.search.local.MySearcher derived from ScoreSearchAlgorithm.
2. Implement the method
public void buildStructure(BayesNet bayesNet, Instances instances). Essentially, you are responsible for setting the parent sets in bayesNet.
You can access the parentsets using bayesNet.getParentSet(iAttribute) where iAttribute is the number of the node/variable.
To add a parent iParent to node iAttribute, use bayesNet.getParentSet(iAttribute).AddParent(iParent, instances) where instances need to be passed for the parent set to derive properties
of the attribute.
Alternatively, implement public void search(BayesNet bayesNet, Instances instances). The implementation of buildStructure in the base class.
This method is called by the SearchAlgorithm will call search after ini- tializing parent sets and if the initAsNaiveBase flag is set, it will start
with a naive Bayes network structure. After calling search in your cus- tom class, it will add arrows if the markovBlanketClassifier flag is set to ensure all attributes are in the Markov blanket of the class node.
3. If the structure learner has options that are not default options, you want to implement public Enumeration listOptions(), public void setOptions(String[] options), public String[] getOptions() and the get and set methods for the properties you want to be able to set.
NB 1. do not use the -E option since that is reserved for the BayesNet class to distinguish the extra options for the SearchAlgorithm class and the Estimator class. If the -E option is used, it will not be passed to your SearchAlgorithm (and probably causes problems in the BayesNet class).
NB 2. make sure to process options of the parent class if any in the get/setOpions methods.
Adding a new estimator
This is the quick guide for adding a new estimator:
1. Create a class that derives from weka.classifiers.bayes.net.estimate.BayesNetEstimator. Let’s say it is called
weka.classifiers.bayes.net.estimate.MyEstimator.
2. Implement the methods
public void initCPTs(BayesNet bayesNet)
8.12. FAQ 155
public void estimateCPTs(BayesNet bayesNet)
public void updateClassifier(BayesNet bayesNet, Instance instance), and
public double[] distributionForInstance(BayesNet bayesNet, Instance instance).
3. If the structure learner has options that are not default options, you want to implement public Enumeration listOptions(), public void setOptions(String[] options), public String[] getOptions() and the get and set methods for the properties you want to be able to set.
NB do not use the -E option since that is reserved for the BayesNet class to distinguish the extra options for the SearchAlgorithm class and the Estimator class. If the -E option is used and no extra arguments are passed to the SearchAlgorithm, the extra options to your Estimator will be passed to the SearchAlgorithm instead. In short, do not use the -E option.
8.12 FAQ
How do I use a data set with continuous variables with the BayesNet classes?
Use the class weka.filters.unsupervised.attribute.Discretize to discretize them. From the command line, you can use
java weka.filters.unsupervised.attribute.Discretize -B 3 -i infile.arff -o outfile.arff
where the -B option determines the cardinality of the discretized variables.
How do I use a data set with missing values with the BayesNet classes?
You would have to delete the entries with missing values or fill in dummy values.
How do I create a random Bayes net structure?
Running from the command line
java weka.classifiers.bayes.net.BayesNetGenerator -B -N 10 -A 9 -C 2
will print a Bayes net with 10 nodes, 9 arcs and binary variables in XML BIF format to standard output.
How do I create an artificial data set using a random Bayes nets?
Running
java weka.classifiers.bayes.net.BayesNetGenerator -N 15 -A 20 -C 3 -M 300
will generate a data set in arff format with 300 instance from a random network with 15 ternary variables and 20 arrows.
156 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS How do I create an artificial data set using a Bayes nets I
have on file?
Running
java weka.classifiers.bayes.net.BayesNetGenerator -F alarm.xml -M 1000
will generate a data set with 1000 instances from the network stored in the file alarm.xml.
How do I save a Bayes net in BIF format?
• GUI: In the Explorer
– learn the network structure,
– right click the relevant run in the result list,
– choose “Visualize graph” in the pop up menu,
– click the floppy button in the Graph Visualizer window.
– a file “save as” dialog pops up that allows you to select the file name to save to.
• Java: CreateaBayesNetandcallBayesNet.toXMLBIF03()whichreturns the Bayes network in BIF format as a String.
• Command line: use the -g option and redirect the output on stdout into a file.
How do I compare a network I learned with one in BIF format?
Specify the -B
How do I use the network I learned for general inference?
There is no general purpose inference in Weka, but you can export the network as XML BIF file (see above) and import it in other packages, for example JavaBayes available under GPL from http://www.cs.cmu.edu/∼javabayes.
8.13 Future development
If you would like to add to the current Bayes network facilities in Weka, you might consider one of the following possibilities.
• Implement more search algorithms, in particular,
– general purpose search algorithms (such as an improved implemen-
tation of genetic search).
– structure search based on equivalent model classes.
– implement those algorithms both for local and global metric based search algorithms.
8.13. FUTURE DEVELOPMENT 157 – implement more conditional independence based search algorithms.
• Implement score metrics that can handle sparse instances in order to allow for processing large datasets.
• Implement traditional conditional independence tests for conditional in- dependence based structure learning algorithms.
• Currently, all search algorithms assume that all variables are discrete. Search algorithms that can handle continuous variables would be interest- ing.
• A limitation of the current classes is that they assume that there are no missing values. This limitation can be undone by implementing score metrics that can handle missing values. The classes used for estimating the conditional probabilities need to be updated as well.
• Onlyleave-one-out,k-foldandcumulativecross-validationareimplemented. These implementations can be made more efficient and other cross-validation methods can be implemented, such as Monte Carlo cross-validation and bootstrap cross validation.
• Implement methods that can handle incremental extensions of the data set for updating network structures.
And for the more ambitious people, there are the following challenges.
• A GUI for manipulating Bayesian network to allow user intervention for adding and deleting arcs and updating the probability tables.
• General purpose inference algorithms built into the GUI to allow user defined queries.
• Allow learning of other graphical models, such as chain graphs, undirected graphs and variants of causal graphs.
• Allow learning of networks with latent variables.
• Allow learning of dynamic Bayesian networks so that time series data can be handled.
158 CHAPTER 8. BAYESIAN NETWORK CLASSIFIERS
Part III Data
159
Chapter 9 ARFF
An ARFF (= Attribute-Relation File Format) file is an ASCII text file that describes a list of instances sharing a set of attributes.
9.1 Overview
ARFF files have two distinct sections. The first section is the Header informa- tion, which is followed the Data information.
The Header of the ARFF file contains the name of the relation, a list of the attributes (the columns in the data), and their types. An example header on the standard IRIS dataset looks like this:
% 1. Title: Iris Plants Database %
% 2. Sources:
% (a) Creator: R.A. Fisher
% (b) Donor: Michael Marshall (MARSHALL%PLU@io.arc.nasa.gov)
% (c) Date: July, 1988
%
@RELATION iris
@ATTRIBUTE sepallength @ATTRIBUTE sepalwidth @ATTRIBUTE petallength @ATTRIBUTE petalwidth @ATTRIBUTE class
NUMERIC
NUMERIC
NUMERIC
NUMERIC {Iris-setosa,Iris-versicolor,Iris-virginica}
The Data of the ARFF file looks like the following:
@DATA 5.1,3.5,1.4,0.2,Iris-setosa 4.9,3.0,1.4,0.2,Iris-setosa 4.7,3.2,1.3,0.2,Iris-setosa 4.6,3.1,1.5,0.2,Iris-setosa 5.0,3.6,1.4,0.2,Iris-setosa 5.4,3.9,1.7,0.4,Iris-setosa
161
162 CHAPTER 9. ARFF
4.6,3.4,1.4,0.3,Iris-setosa 5.0,3.4,1.5,0.2,Iris-setosa 4.4,2.9,1.4,0.2,Iris-setosa 4.9,3.1,1.5,0.1,Iris-setosa
Lines that begin with a % are comments. The @RELATION, @ATTRIBUTE and @DATA declarations are case insensitive.
9.2 Examples
Several well-known machine learning datasets are distributed with Weka in the $WEKAHOME/data directory as ARFF files.
9.2.1 The ARFF Header Section
The ARFF Header section of the file contains the relation declaration and at- tribute declarations.
The @relation Declaration
The relation name is defined as the first line in the ARFF file. The format is:
@relation
where
• a character below \\u0021 • ’{’, ’}’, ’,’, or ’%’
Moreover, it can only begin with a single or double quote if there is a corre- sponding quote at the end of the name.
The @attribute Declarations
Attribute declarations take the form of an ordered sequence of @attribute statements. Each attribute in the data set has its own @attribute statement which uniquely defines the name of that attribute and it’s data type. The order the attributes are declared indicates the column position in the data section of the file. For example, if an attribute is the third one declared then Weka expects that all that attributes values will be found in the third comma delimited column.
The format for the @attribute statement is: @attribute
where the
The
9.2. EXAMPLES 163
• numeric
• integer is treated as numeric
• real is treated as numeric
•
• string
• date [
• relational for multi-instance data (for future use)
where
Numeric attributes
Numeric attributes can be real or integer numbers.
Nominal attributes
Nominal values are defined by providing an
For example, the class value of the Iris dataset can be defined as follows:
@ATTRIBUTE class {Iris-setosa,Iris-versicolor,Iris-virginica}
Values that contain spaces must be quoted.
String attributes
String attributes allow us to create attributes containing arbitrary textual val- ues. This is very useful in text-mining applications, as we can create datasets with string attributes, then write Weka Filters to manipulate strings (like String- ToWordVectorFilter). String attributes are declared as follows:
@ATTRIBUTE LCC string
Date attributes
Date attribute declarations take the form:
@attribute
where
Dates must be specified in the data section as the corresponding string rep- resentations of the date/time (see example below).
164 CHAPTER 9. ARFF Relational attributes
Relational attribute declarations take the form:
@attribute
@end
For the multi-instance dataset MUSK1 the definition would look like this (”…” denotes an omission):
@attribute molecule_name {MUSK-jf78,…,NON-MUSK-199} @attribute bag relational
@attribute f1 numeric …
@attribute f166 numeric
@end bag
@attribute class {0,1} …
9.2.2 The ARFF Data Section
The ARFF Data section of the file contains the data declaration line and the actual instance lines.
The @data Declaration
The @data declaration is a single line denoting the start of the data segment in the file. The format is:
@data
The instance data
Each instance is represented on a single line, with carriage returns denoting the end of the instance. A percent sign (%) introduces a comment, which continues to the end of the line.
Attribute values for each instance are delimited by commas. They must appear in the order that they were declared in the header section (i.e. the data corresponding to the nth @attribute declaration is always the nth field of the attribute).
Missing values are represented by a single question mark, as in:
@data 4.4,?,1.5,?,Iris-setosa
Values of string and nominal attributes are case sensitive, and any that contain space or the comment-delimiter character % must be quoted. (The code suggests that double-quotes are acceptable and that a backslash will escape individual characters.) An example follows:
9.3. SPARSE ARFF FILES 165 @relation LCCvsLCSH
@attribute LCC string @attribute LCSH string
@data
AG5, ’Encyclopedias and dictionaries.;Twentieth century.’ AS262, ’Science — Soviet Union — History.’
AE5, ’Encyclopedias and dictionaries.’
AS281, ’Astronomy, Assyro-Babylonian.;Moon — Phases.’ AS281, ’Astronomy, Assyro-Babylonian.;Moon — Tables.’
Dates must be specified in the data section using the string representation spec- ified in the attribute declaration. For example:
@RELATION Timestamps
@ATTRIBUTE timestamp DATE “yyyy-MM-dd HH:mm:ss”
@DATA
“2001-04-03 12:12:12” “2001-05-03 12:59:55″
Relational data must be enclosed within double quotes ”. For example an in- stance of the MUSK1 dataset (”…” denotes an omission):
MUSK-188,”42,…,30”,1
9.3 Sparse ARFF files
Sparse ARFF files are very similar to ARFF files, but data with value 0 are not be explicitly represented.
Sparse ARFF files have the same header (i.e @relation and @attribute tags) but the data section is different. Instead of representing each value in order, like this:
@data
0, X, 0, Y, “class A” 0, 0, W, 0, “class B”
the non-zero attributes are explicitly identified by attribute number and their value stated, like this:
@data
{1 X, 3 Y, 4 “class A”} {2 W, 4 “class B”}
Each instance is surrounded by curly braces, and the format for each entry is:
166 CHAPTER 9. ARFF
Note that the omitted values in a sparse instance are 0, they are not missing values! If a value is unknown, you must explicitly represent it with a question mark (?).
Warning: There is a known problem saving SparseInstance objects from datasets that have string attributes. In Weka, string and nominal data values are stored as numbers; these numbers act as indexes into an array of possible attribute values (this is very efficient). However, the first string value is as- signed index 0: this means that, internally, this value is stored as a 0. When a SparseInstance is written, string instances with internal value 0 are not out- put, so their string value is lost (and when the arff file is read again, the default value 0 is the index of a different string value, so the attribute value appears to change). To get around this problem, add a dummy string value at index 0 that is never used whenever you declare string attributes that are likely to be used in SparseInstance objects and saved as Sparse ARFF files.
9.4 Instance weights in ARFF files
A weight can be associated with an instance in a standard ARFF file by ap- pending it to the end of the line for that instance and enclosing the value in curly braces. E.g:
@data
0, X, 0, Y, “class A”, {5}
For a sparse instance, this example would look like:
@data
{1 X, 3 Y, 4 “class A”}, {5}
Note that any instance without a weight value specified is assumed to have a weight of 1 for backwards compatibility.
Chapter 10 XRFF
The XRFF (Xml attribute Relation File Format) is a representing the data in a format that can store comments, attribute and instance weights.
10.1 File extensions
The following file extensions are recognized as XRFF files:
• .xrff
the default extension of XRFF files
• .xrff.gz
the extension for gzip compressed XRFF files (see Compression section for more details)
10.2 Comparison 10.2.1 ARFF
In the following a snippet of the UCI dataset iris in ARFF format: @relation iris
@attribute sepallength numeric
@attribute sepalwidth numeric
@attribute petallength numeric
@attribute petalwidth numeric
@attribute class {Iris-setosa,Iris-versicolor,Iris-virginica}
@data 5.1,3.5,1.4,0.2,Iris-setosa 4.9,3,1.4,0.2,Iris-setosa …
167
168 CHAPTER 10. XRFF
10.2.2 XRFF
And the same dataset represented as XRFF file:
] >
10.3.
SPARSE FORMAT 169
…
10.3 Sparse format
The XRFF format also supports a sparse data representation. Even though the iris dataset does not contain sparse data, the above example will be used here to illustrate the sparse format:
…
…
…
170 CHAPTER 10. XRFF In contrast to the normal data format, each sparse instance tag contains a type
attribute with the value sparse:
And each value tag needs to specify the index attribute, which contains the 1-based index of this value.
10.4 Compression
Since the XML representation takes up considerably more space than the rather compact ARFF format, one can also compress the data via gzip. Weka automat- ically recognizes a file being gzip compressed, if the file’s extension is .xrff.gz instead of .xrff.
The Weka Explorer, Experimenter and command-line allow one to load/save compressed and uncompressed XRFF files (this applies also to ARFF files).
10.5 Useful features
In addition to all the features of the ARFF format, the XRFF format contains the following additional features:
• class attribute specification • attribute weights
10.5.1 Class attribute specification
Via the class=”yes” attribute in the attribute specification in the header, one can define which attribute should act as class attribute. A feature that can be used on the command line as well as in the Experimenter, which now can also load other data formats, and removing the limitation of the class attribute always having to be the last one.
Snippet from the iris dataset:
10.5.2 Attribute weights
Attribute weights are stored in an attributes meta-data tag (in the header sec- tion). Here is an example of the petalwidth attribute with a weight of 0.9:
10.5. USEFUL FEATURES 171 10.5.3 Instance weights
Instance weights are defined via the weight attribute in each instance tag. By default, the weight is 1. Here is an example:
172 CHAPTER 10. XRFF
Chapter 11 Converters
11.1 Introduction
Weka offers conversion utilities for several formats, in order to allow import from different sorts of datasources. These utilities, called converters, are all located in the following package:
weka.core.converters
For a certain kind of converter you will find two classes • one for loading (classname ends with Loader) and • one for saving (classname ends with Saver).
Weka contains converters for the following data sources:
• ARFF files (ArffLoader, ArffSaver)
• C4.5 files (C45Loader, C45Saver)
• CSV files (CSVLoader, CSVSaver)
• files containing serialized instances (SerializedInstancesLoader, Serial- izedInstancesSaver)
• JDBC databases (DatabaseLoader, DatabaseSaver)
• libsvm files (LibSVMLoader, LibSVMSaver)
• XRFF files (XRFFLoader, XRFFSaver)
• text directories for text mining (TextDirectoryLoader)
173
174 CHAPTER 11. CONVERTERS 11.2 Usage
11.2.1 File converters
File converters can be used as follows:
• Loader
They take one argument, which is the file that should be converted, and print the result to stdout. You can also redirect the output into a file:
java
Here’s an example for loading the CSV file iris.csv and saving it as iris.arff :
java weka.core.converters.CSVLoader iris.csv > iris.arff
• Saver
For a Saver you specify the ARFF input file via -i and the output file in the specific format with -o:
java
11.2. USAGE 175
• Loader
You have to specify at least a SQL query with the -Q option (there are additional options for incremental loading)
java weka.core.converters.DatabaseLoader -Q “select * from employee”
• Saver
The Saver takes an ARFF file as input like any other Saver, but then also the table where to save the data to via -T:
java weka.core.converters.DatabaseSaver -i iris.arff -T iris
176 CHAPTER 11. CONVERTERS
Chapter 12 Stemmers
12.1 Introduction
Weka now supports stemming algorithms. The stemming algorithms are located in the following package:
weka.core.stemmers
Currently, the Lovins Stemmer (+ iterated version) and support for the Snow- ball stemmers are included.
12.2 Snowball stemmers
Weka contains a wrapper class for the Snowball (homepage: http://snowball.tartarus.org/) stemmers (containing the Porter stemmer and several other stemmers for dif-
ferent languages). The relevant class is weka.core.stemmers.Snowball.
The Snowball classes are not included, they only have to be present in the classpath. The reason for this is, that the Weka team doesn’t have to watch out for new versions of the stemmers and update them.
There are two ways of getting hold of the Snowball stemmers:
1. You can add the following pre-compiled jar archive to your classpath and you’re set (based on source code from 2005-10-19, compiled 2005-10-22). http://www.cs.waikato.ac.nz/∼ml/weka/stemmers/snowball.jar
2. You can compile the stemmers yourself with the newest sources. Just download the following ZIP file, unpack it and follow the instructions in the README file (the zip contains an ANT (http://ant.apache.org/) build script for generating the jar archive). http://www.cs.waikato.ac.nz/∼ml/weka/stemmers/snowball.zip Note: the patch target is specific to the source code from 2005-10-19.
177
178 CHAPTER 12. STEMMERS
12.3 Using stemmers
The stemmers can either used • from commandline
• withintheStringToWordVector(packageweka.filters.unsupervised.attribute) 12.3.1 Commandline
All stemmers support the following options:
• -h
for displaying a brief help
• -i
• -o
The file to output the processed data to (default stdout)
• -l
Uses lowercase strings, i.e., the input is automatically converted to lower case
12.3.2 StringToWordVector
Just use the GenericObjectEditor to choose the right stemmer and the desired options (if the stemmer offers additional options).
12.4 Adding new stemmers
You can easily add new stemmers, if you follow these guidelines (for use in the GenericObjectEditor):
• they should be located in the weka.core.stemmers package (if not, then the GenericObjectEditor.props/GenericPropertiesCreator.props file need to be updated) and
• they must implement the interface weka.core.stemmers.Stemmer.
Chapter 13 Databases
13.1 Configuration files
Thanks to JDBC it is easy to connect to Databases that provide a JDBC driver. Responsible for the setup is the following properties file, located in the weka.experiment package:
DatabaseUtils.props
You can get this properties file from the weka.jar or weka-src.jar jar-archive, both part of a normal Weka release. If you open up one of those files, you’ll find the properties file in the sub-folder weka/experiment.
Weka comes with example files for a wide range of databases:
• DatabaseUtils.props.hsql – HSQLDB (>=3.4.1)
• DatabaseUtils.props.msaccess – MS Access (>3.4.14, >3.5.8, >3.6.0) see the Windows databases chapter for more information.
• DatabaseUtils.props.mssqlserver – MS SQL Server 2000 (>=3.4.9, >=3.5.4)
• DatabaseUtils.props.mssqlserver2005 – MS SQL Server 2005 (>=3.4.11, >=3.5.6)
• DatabaseUtils.props.mysql – MySQL (>=3.4.9, >=3.5.4)
• DatabaseUtils.props.odbc – ODBC access via Sun’s ODBC/JDBC bridge, e.g., for MS Sql Server (>=3.4.9, >=3.5.4)
see the Windows databases chapter for more information.
• DatabaseUtils.props.oracle – Oracle 10g (>=3.4.9, >=3.5.4)
• DatabaseUtils.props.postgresql – PostgreSQL 7.4 (>=3.4.9, >=3.5.4)
• DatabaseUtils.props.sqlite3 – sqlite 3.x (>3.4.12, >3.5.7)
179
180 CHAPTER 13. DATABASES
The easiest way is just to place the extracted properties file into your HOME directory. For more information on how property files are processed, check out the following URL:
http://weka.wikispaces.com/Properties+File
Note: Weka only looks for the DatabaseUtils.props file. If you take one of
the example files listed above, you need to rename it first.
13.2 Setup
Under normal circumstances you only have to edit the following two properties:
• jdbcDriver • jdbcURL
Driver
jdbcDriver is the classname of the JDBC driver, necessary to connect to your database, e.g.:
• HSQLDB org.hsqldb.jdbcDriver
• MS SQL Server 2000 (Desktop Edition) com.microsoft.jdbc.sqlserver.SQLServerDriver
• MS SQL Server 2005 com.microsoft.sqlserver.jdbc.SQLServerDriver
• MySQL
org.gjt.mm.mysql.Driver (or com.mysql.jdbc.Driver)
• ODBC – part of Sun’s JDKs/JREs, no external driver necessary sun.jdbc.odbc.JdbcOdbcDriver
• Oracle oracle.jdbc.driver.OracleDriver
• PostgreSQL org.postgresql.Driver
• sqlite 3.x org.sqlite.JDBC
URL
jdbcURL specifies the JDBC URL pointing to your database (can be still changed in the Experimenter/Explorer), e.g. for the database MyDatabase on the server server.my.domain:
13.3. MISSING DATATYPES 181
• HSQLDB
jdbc:hsqldb:hsql://server.my.domain/MyDatabase
• MS SQL Server 2000 (Desktop Edition) jdbc:microsoft:sqlserver://v:1433
(Note: if you add ;databasename=db-name you can connect to a different database than the default one, e.g., MyDatabase)
• MS SQL Server 2005 jdbc:sqlserver://server.my.domain:1433
• MySQL jdbc:mysql://server.my.domain:3306/MyDatabase
• ODBC
jdbc:odbc:DSN name (replace DSN name with the DSN that you want to use)
• Oracle (thin driver) jdbc:oracle:thin:@server.my.domain:1526:orcl (Note: @machineName:port:SID)
for the Express Edition you can use jdbc:oracle:thin:@server.my.domain:1521:XE
• PostgreSQL
jdbc:postgresql://server.my.domain:5432/MyDatabase
You can also specify user and password directly in the URL: jdbc:postgresql://server.my.domain:5432/MyDatabase?user=<...>&password=<...> where you have to replace the <...> with the correct values
• sqlite 3.x jdbc:sqlite:/path/to/database.db (you can access only local files)
13.3 Missing Datatypes
Sometimes (e.g. with MySQL) it can happen that a column type cannot be interpreted. In that case it is necessary to map the name of the column type to the Java type it should be interpreted as. E.g. the MySQL type TEXT is returned as BLOB from the JDBC driver and has to be mapped to String (0 represents String – the mappings can be found in the comments of the properties file):
182 13.5. TROUBLESHOOTING 183 13.5 Troubleshooting 184 CHAPTER 13. DATABASES Chapter 14 186 Step 3: Open the database 188 CHAPTER 14. WINDOWS DATABASES Part IV Appendix Chapter 15 Research 192 CHAPTER 15. Bash shell script get wekatechinfo.sh
CHAPTER 13.
DATABASES
Java type Java method Identifier Weka attribute type String getString() 0 nominal
boolean getBoolean() 1 nominal
double getDouble() 2 numeric
byte getByte() 3 numeric short getByte() 4 numeric int getInteger() 5 numeric long getLong() 6 numeric float getFloat() 7 numeric date getDate() 8 date text getString() 9 string time getTime() 10 date
In the props file one lists now the type names that the database returns and what Java type it represents (via the identifier), e.g.:
CHAR=0
VARCHAR=0
CHAR and VARCHAR are both String types, hence they are interpreted as String (identifier 0)
Note: in case database types have blanks, one needs to replace those blanks with an underscore, e.g., DOUBLE PRECISION must be listed like this:
DOUBLE_PRECISION=2
13.4 Stored Procedures
Let’s say you’re tired of typing the same query over and over again. A good way to shorten that, is to create a stored procedure.
PostgreSQL 7.4.x
The following example creates a procedure called emplyoee name that returns the names of all the employees in table employee. Even though it doesn’t make much sense to create a stored procedure for this query, nonetheless, it shows how to create and call stored procedures in PostgreSQL.
• Create
CREATE OR REPLACE FUNCTION public.employee_name() RETURNS SETOF text AS ’select name from employee’ LANGUAGE ’sql’ VOLATILE;
• SQL statement to call procedure SELECT * FROM employee_name()
• Retrieve data via InstanceQuery
java weka.experiment.InstanceQuery
-Q “SELECT * FROM employee_name()” -U
• In case you’re experiencing problems connecting to your database, check out the WEKA Mailing List (see Weka homepage for more information). It is possible that somebody else encountered the same problem as you and you’ll find a post containing the solution to your problem.
• Specific MS SQL Server 2000 Troubleshooting
– Error Establishing Socket with JDBC Driver
Add TCP/IP to the list of protocols as stated in the following article: http://support.microsoft.com/default.aspx?scid=kb;en-us;313178
– Login failed for user ’sa’. Reason: Not associated with a trusted SQL Server connection.
For changing the authentication to mixed mode see the following article:
http://support.microsoft.com/kb/319930/en-us
• MS SQL Server 2005: TCP/IP is not enabled for SQL Server, or the server or port number specified is incorrect.Verify that SQL Server is lis- tening with TCP/IP on the specified server and port. This might be re- ported with an exception similar to: ”The login has failed. The TCP/IP connection to the host has failed.” This indicates one of the following:
– SQL Server is installed but TCP/IP has not been installed as a net- work protocol for SQL Server by using the SQL Server Network Util- ity for SQL Server 2000, or the SQL Server Configuration Manager for SQL Server 2005
– TCP/IP is installed as a SQL Server protocol, but it is not listening on the port specified in the JDBC connection URL. The default port is 1433.
– The port that is used by the server has not been opened in the firewall
• The Added driver: … output on the commandline does not mean that the actual class was found, but only that Weka will attempt to load the class later on in order to establish a database connection.
• The error message No suitable driver can be caused by the following:
– The JDBC driver you are attempting to load is not in the CLASS- PATH (Note: using -jar in the java commandline overwrites the CLASSPATH environment variable!). Open the SimpleCLI, run the command java weka.core.SystemInfo and check whether the prop- erty java.class.path lists your database jar. If not correct your CLASSPATH or the Java call you start Weka with.
– The JDBC driver class is misspelled in the jdbcDriver property or you have multiple entries of jdbcDriver (properties files need unique keys !)
– The jdbcURL property has a spelling error and tries to use a non- existing protocol or you listed it multiple times, which doesn’t work either (remember, properties files need unique keys!)
Windows databases
A common query we get from our users is how to open a Windows database in the Weka Explorer. This page is intended as a guide to help you achieve this. It is a complicated process and we cannot guarantee that it will work for you. The process described makes use of the JDBC-ODBC bridge that is part of Sun’s JRE/JDK 1.3 (and higher).
The following instructions are for Windows 2000. Under other Windows versions there may be slight differences.
Step 1: Create a User DSN
1. Go to the Control Panel
2. Choose Adminstrative Tools
3. Choose Data Sources (ODBC)
4. At the User DSN tab, choose Add… 5. Choose database
• Microsoft Access
(a) Note: Make sure your database is not open in another application before following the steps below.
(b) Choose the Microsoft Access driver and click Finish
(c) Give the source a name by typing it into the Data Source
Name field
(d) In the Database section, choose Select…
(e) Browse to find your database file, select it and click OK
(f) Click OK to finalize your DSN
• Microsoft SQL Server 2000 (Desktop Engine)
(a) Choose the SQL Server driver and click Finish
(b) Give the source a name by typing it into the Name field
(c) Add a description for this source in the Description field
(d) Select the server you’re connecting to from the Server combobox
185
CHAPTER 14. WINDOWS DATABASES
(e) (f)
(g)
(h)
(b) Give the source a name by typing it into the Data Source Name field
(c) Add a description for this source in the Description field
(d) Specify the server you’re connecting to in Server
(e) Fill in the user to use for connecting to the database in the User field, the same for the password
(f) Choose the database for this DSN from the Database combobox
(g) Click on OK
6. Your DSN should now be listed in the User Data Sources list Step 2: Set up the DatabaseUtils.props file
You will need to configure a file called DatabaseUtils.props. This file already exists under the path weka/experiment/ in the weka.jar file (which is just a ZIP file) that is part of the Weka download. In this directory you will also find a sample file for ODBC connectivity, called DatabaseUtils.props.odbc, and one specifically for MS Access, called DatabaseUtils.props.msaccess, also using ODBC. You should use one of the sample files as basis for your setup, since they already contain default values specific to ODBC access.
This file needs to be recognized when the Explorer starts. You can achieve this by making sure it is in the working directory or the home directory (if you are unsure what the terms working directory and home directory mean, see the Notes section). The easiest is probably the second alternative, as the setup will apply to all the Weka instances on your machine.
Just make sure that the file contains the following lines at least:
jdbcDriver=sun.jdbc.odbc.JdbcOdbcDriver jdbcURL=jdbc:odbc:dbname
where dbname is the name you gave the user DSN. (This can also be changed once the Explorer is running.)
For the verification of the authenticity of the login ID choose
With SQL Server…
Check Connect to SQL Server to obtain default settings…
and supply the user ID and password with which you installed the Desktop Engine
Just click on Next until it changes into Finish and click this, too
For testing purposes, click on Test Data Source… – the result should be TESTS COMPLETED SUCCESSFULLY!
Click on OK
(a) Choose the MySQL ODBC driver and click Finish
(i)
• MySQL
1. Start up the Weka Explorer.
2. Choose Open DB…
3. The URL should read ”jdbc:odbc:dbname” where dbname is the name you gave the user DSN.
4. Click Connect
5. Enter a Query, e.g., ”select * from tablename” where tablename is the name of the database table you want to read. Or you could put a more complicated SQL query here instead.
6. Click Execute
7. When you’re satisfied with the returned data, click OK to load the data
into the Preprocess panel.
Notes
• Working directory
The directory a process is started from. When you start Weka from the Windows Start Menu, then this directory would be Weka’s installation directory (the java process is started from that directory).
• Home directory
The directory that contains all the user’s data. The exact location depends on the operating system and the version of the operating system. It is stored in the following environment variable:
– Unix/Linux $HOME
– Windows %USERPROFILE%
– Cygwin $USERPROFILE
You should be able output the value in a command prompt/terminal with the echo command. E.g., for Windows this would be:
echo %USERPROFILE%
187
189
15.1 Citing Weka
If you want to refer to Weka in a publication, please cite following SIGKDD Explorations1 paper. The full citation is:
Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Pe- ter Reutemann, Ian H. Witten (2009); The WEKA Data Mining Software: An Update; SIGKDD Explorations, Volume 11, Issue 1.
15.2 Paper references
Due to the introduction of the weka.core.TechnicalInformationHandler in- terface it is now easy to extract all the paper references via weka.core.ClassDiscovery and weka.core.TechnicalInformation.
The script listed at the end, extracts all the paper references from Weka based on a given jar file and dumps it to stdout. One can either generate simple plain text output (option -p) or BibTeX compliant one (option -b).
Typical use (after an ant exejar) for BibTeX:
get_wekatechinfo.sh -d ../ -w ../dist/weka.jar -b > ../tech.txt
(command is issued from the same directory the Weka build.xml is located in)
1http://www.kdd.org/explorations/issues/11-1-2009-07/p2V11n1.pdf 191
#!/bin/bash #
# This script prints the information stored in TechnicalInformationHandlers
# to stdout. #
# FracPete, $Revision: 4582 $
# the usage of this script function usage()
{
RESEARCH
echo
echo “${0##*/} -d
echo
echo “Prints the information stored in TechnicalInformationHandlers to stdout.” echo
echo ” -h
echo ” -d
echo ”
echo ”
echo ” -w
echo ”
echo ” -p
echo ” -b
echo
}
# generates a filename out of the classname TMP and returns it in TMP
# uses the directory in DIR function class_to_filename() {
TMP=$DIR”/”‘echo $TMP | sed s/”\.”/”\/”/g‘”.java” }
# variables
DIR=”.”
PLAINTEXT=”no”
BIBTEX=”no”
WEKA=”” TECHINFOHANDLER=”weka.core.TechnicalInformationHandler” TECHINFO=”weka.core.TechnicalInformation” CLASSDISCOVERY=”weka.core.ClassDiscovery”
# interprete parameters
while getopts “:hpbw:d:” flag do
case $flag in
this help”
the directory to look for packages, must be the one just above” the ’weka’ package, default: $DIR”
the weka jar to use, if not in CLASSPATH”
prints the information in plaintext format”
prints the information in BibTeX format”
p) b) d) w) h)
*)
esac done
# either
if [ “$PLAINTEXT” = “$BIBTEX” ] then
echo
echo “ERROR: either -p or -b has to be given!” echo
usage
exit 2
fi
PLAINTEXT=”yes” ;;
BIBTEX=”yes”
;;
DIR=$OPTARG ;; WEKA=$OPTARG ;;
usage exit 0 ;; usage exit 1 ;;
plaintext or bibtex
15.2. PAPER REFERENCES 193
# do we have everything?
if [ “$DIR” = “” ] || [ ! -d “$DIR” ] then
echo
echo “ERROR: no directory or non-existing one provided!” echo
usage
exit 3
fi
# generate Java call if [ “$WEKA” = “” ] then
JAVA=”java” else
JAVA=”java -classpath $WEKA” fi
if [ “$PLAINTEXT” = “yes” ] then
CMD=”$JAVA $TECHINFO -plaintext” elif [ “$BIBTEX” = “yes” ]
then
CMD=”$JAVA $TECHINFO -bibtex” fi
# find packages
TMP=‘find $DIR -mindepth 1 -type d | grep -v CVS | sed s/”.*weka”/”weka”/g | sed s/”\/”/./g‘ PACKAGES=‘echo $TMP | sed s/” “/,/g‘
# get technicalinformationhandlers
TECHINFOHANDLERS=‘$JAVA weka.core.ClassDiscovery $TECHINFOHANDLER $PACKAGES | grep “\. weka” | sed s/”.*weka”/weka/g‘
# output information
echo
for i in $TECHINFOHANDLERS do
TMP=$i;class_to_filename
# exclude internal classes if [ ! -f $TMP ]
then
continue fi
$CMD -W $i
echo done
194 CHAPTER 15. RESEARCH
Chapter 16 Using the API
Using the graphical tools, like the Explorer, or just the command-line is in most cases sufficient for the normal user. But WEKA’s clearly defined API (“ap- plication programming interface”) makes it very easy to “embed” it in another projects. This chapter covers the basics of how to achieve the following common tasks from source code:
• Setting options
• Creating datasets in memory • Loading and saving data
• Filtering
• Classifying
• Clustering
• Selecting attributes
• Visualization
• Serialization
Even though most of the code examples are for the Linux platform, using for- ward slashes in the paths and file names, they do work on the MS Windows plat- form as well. To make the examples work under MS Windows, one only needs to adapt the paths, changing the forward slashes to backslashes and adding a drive letter where necessary.
Note
WEKA is released under the GNU General Public License version 21 (GPLv2), i.e., that derived code or code that uses WEKA needs to be released under the GPLv2 as well. If one is just using WEKA for a personal project that does not get released publicly then one is not affected. But as soon as one makes the project publicly available (e.g., for download), then one needs to make the source code available under the GLPv2 as well, alongside the binaries.
1 http://www.gnu.org/licenses/gpl- 2.0.html 195
196 CHAPTER 16. USING THE API 16.1 Option handling
Configuring an object, e.g., a classifier, can either be done using the appro- priate get/set-methods for the property that one wishes to change, like the Explorer does. Or, if the class implements the weka.core.OptionHandler in- terface, one can just use the object’s ability to parse command-line options via the setOptions(String[]) method (the counterpart of this method is getOptions(), which returns a String[] array). The difference between the two approaches is, that the setOptions(String[]) method cannot be used to set the options incrementally. Default values are used for all options that haven’t been explicitly specified in the options array.
The most basic approach is to assemble the String array by hand. The following example creates an array with a single option (“-R”) that takes an argument (“1”) and initializes the Remove filter with this option:
import weka.filters.unsupervised.attribute.Remove; …
String[] options = new String[2];
options[0] = “-R”;
options[1] = “1”;
Remove rm = new Remove(); rm.setOptions(options);
Since the setOptions(String[]) method expects a fully parsed and correctly split up array (which is done by the console/command prompt), some common pitfalls with this approach are:
• Combination of option and argument – Using “-R 1” as an element of the String array will fail, prompting WEKA to output an error message stating that the option “R 1” is unknown.
• Trailing blanks – Using “-R ” will fail as well, since no trailing blanks are removed and therefore option “R ” will not be recognized.
The easiest way to avoid these problems, is to provide a String array that has been generated automatically from a single command-line string using the splitOptions(String) method of the weka.core.Utils class. Here is an ex- ample:
import weka.core.Utils;
…
String[] options = Utils.splitOptions(“-R 1”);
As this method ignores whitespaces, using “ -R 1” or “-R 1 ” will return the same result as “-R 1”.
Complicated command-lines with lots of nested options, e.g., options for the support-vector machine classifier SMO (package weka.classifiers.functions) including a kernel setup, are a bit tricky, since Java requires one to escape dou- ble quotes and backslashes inside a String. The Wiki[2] article “Use Weka in your Java code” references the Java class OptionsToCode, which turns any command-line into appropriate Java source code. This example class is also available from the Weka Examples collection[3]: weka.core.OptionsToCode.
16.1. OPTION HANDLING 197 Instead of using the Remove filter’s setOptions(String[]) method, the
following code snippet uses the actual set-method for this property:
import weka.filters.unsupervised.attribute.Remove; …
Remove rm = new Remove(); rm.setAttributeIndices(“1”);
In order to find out, which option belongs to which property, i.e., get/set- method, it is best to have a look at the setOptions(String[]) and getOptions() methods. In case these methods use the member variables directly, one just has to look for the methods making this particular member variable accessible to the outside.
Using the set-methods, one will most likely come across ones that re- quire a weka.core.SelectedTag as parameter. An example for this, is the setEvaluation method of the meta-classifier GridSearch (located in package weka.classifiers.meta). The SelectedTag class is used in the GUI for dis- playing drop-down lists, enabling the user to chose from a predefined list of values. GridSearch allows the user to chose the statistical measure to base the evaluation on (accuracy, correlation coefficient, etc.).
A SelectedTag gets constructed using the array of all possible weka.core.Tag elements that can be chosen and the integer or string ID of the Tag. For in- stance, GridSearch’s setOptions(String[]) method uses the supplied string ID to set the evaluation type (e.g., “ACC” for accuracy), or, if the evaluation option is missing, the default integer ID EVALUATION ACC. In both cases, the array TAGS EVALUATION is used, which defines all possible options:
import weka.core.SelectedTag;
…
String tmpStr = Utils.getOption(’E’, options); if (tmpStr.length() != 0)
setEvaluation(new SelectedTag(tmpStr, TAGS_EVALUATION)); else
setEvaluation(new SelectedTag(EVALUATION_CC, TAGS_EVALUATION));
198 CHAPTER 16. USING THE API 16.2 Loading data
Before any filter, classifier or clusterer can be applied, data needs to be present. WEKA enables one to load data from files (in various file formats) and also from databases. In the latter case, it is assumed in that the database connection is set up and working. See chapter 13 for more details on how to configure WEKA correctly and also more information on JDBC (Java Database Connectivity) URLs.
Example classes, making use of the functionality covered in this section, can be found in the wekaexamples.core.converters package of the Weka Exam- ples collection[3].
The following classes are used to store data in memory:
• weka.core.Instances – holds a complete dataset. This data structure is row-based; single rows can be accessed via the instance(int) method using a 0-based index. Information about the columns can be accessed via theattribute(int)method. Thismethodreturnsweka.core.Attribute objects (see below).
• weka.core.Instance – encapsulates a single row. It is basically a wrap- per around an array of double primitives. Since this class contains no information about the type of the columns, it always needs access to a weka.core.Instances object (see methods dataset and setDataset). The class weka.core.SparseInstance is used in case of sparse data.
• weka.core.Attribute – holds the type information about a single column in the dataset. It stores the type of the attribute, as well as the labels for nominal attributes, the possible values for string attributes or the datasets for relational attributes (these are just weka.core.Instances objects again).
16.2.1 Loading data from files
When loading data from files, one can either let WEKA choose the appropriate loader (the available loaders can be found in the weka.core.converters pack- age) based on the file’s extension or one can use the correct loader directly. The latter case is necessary if the files do not have the correct extension.
The DataSource class (inner class of the weka.core.converters.ConverterUtils class) can be used to read data from files that have the appropriate file extension.
Here are some examples:
import weka.core.converters.ConverterUtils.DataSource;
import weka.core.Instances;
…
Instances data1 = DataSource.read(“/some/where/dataset.arff”); Instances data2 = DataSource.read(“/some/where/dataset.csv”); Instances data3 = DataSource.read(“/some/where/dataset.xrff”);
In case the file does have a different file extension than is normally associated with the loader, one has to use a loader directly. The following example loads a CSV (“comma-separated values”) file:
import weka.core.converters.CSVLoader;
16.2. LOADING DATA 199
import weka.core.Instances;
import java.io.File;
…
CSVLoader loader = new CSVLoader(); loader.setSource(new File(“/some/where/some.data”)); Instances data = loader.getDataSet();
NB: Not all file formats allow to store information about the class attribute (e.g., ARFF stores no information about class attribute, but XRFF does). If a class attribute is required further down the road, e.g., when using a classifier, it can be set with the setClassIndex(int) method:
// uses the first attribute as class attribute if (data.classIndex() == -1)
data.setClassIndex(0); …
// uses the last attribute as class attribute if (data.classIndex() == -1)
data.setClassIndex(data.numAttributes() – 1);
16.2.2 Loading data from databases
For loading data from databases, one of the following two classes can be used:
• weka.experiment.InstanceQuery
• weka.core.converters.DatabaseLoader
The differences between them are, that the InstanceQuery class allows one to retrieve sparse data and the DatabaseLoader can retrieve the data incremen- tally.
Here is an example of using the InstanceQuery class:
import weka.core.Instances;
import weka.experiment.InstanceQuery;
…
InstanceQuery query = new InstanceQuery(); query.setDatabaseURL(“jdbc_url”); query.setUsername(“the_user”); query.setPassword(“the_password”); query.setQuery(“select * from whatsoever”);
// if your data is sparse, then you can say so, too: // query.setSparseData(true);
Instances data = query.retrieveInstances();
And an example using the DatabaseLoader class in “batch retrieval”:
import weka.core.Instances;
import weka.core.converters.DatabaseLoader;
…
DatabaseLoader loader = new DatabaseLoader(); loader.setSource(“jdbc_url”, “the_user”, “the_password”); loader.setQuery(“select * from whatsoever”);
Instances data = loader.getDataSet();
200 CHAPTER 16. USING THE API The DatabaseLoader is used in “incremental mode” as follows:
import weka.core.Instance;
import weka.core.Instances;
import weka.core.converters.DatabaseLoader;
…
DatabaseLoader loader = new DatabaseLoader(); loader.setSource(“jdbc_url”, “the_user”, “the_password”); loader.setQuery(“select * from whatsoever”);
Instances structure = loader.getStructure();
Instances data = new Instances(structure);
Instance inst;
while ((inst = loader.getNextInstance(structure)) != null)
data.add(inst);
Notes:
• Not all database systems allow incremental retrieval.
• Not all queries have a unique key to retrieve rows incrementally. In that
case, one can supply the necessary columns with the setKeys(String)
method (comma-separated list of columns).
• If the data cannot be retrieved in an incremental fashion, it is first fully
loaded into memory and then provided row-by-row (“pseudo-incremental”).
16.2. LOADING DATA 201
202 CHAPTER 16. USING THE API 16.3 Creating datasets in memory
Loading datasets from disk or database are not the only ways of obtaining data in WEKA: datasets can be created in memory or on-the-fly. Generating a dataset memory structure (i.e., a weka.core.Instances object) is a two-stage process:
1. Defining the format of the data by setting up the attributes. 2. Adding the actual data, row by row.
The class wekaexamples.core.CreateInstances of the Weka Examples collection[3] generates an Instances object containing all attribute types WEKA can handle
at the moment.
16.3.1 Defining the format
There are currently five different types of attributes available in WEKA:
• numeric – continuous variables
• date – date variables
• nominal – predefined labels
• string – textual data
• relational – contains other relations, e.g., the bags in case of multi- instance data
For all of the different attribute types, WEKA uses the same class, weka.core.Attribute, but with different constructors. In the following, these different constructors are explained.
• numeric – The easiest attribute type to create, as it requires only the name of the attribute:
Attribute numeric = new Attribute(“name_of_attr”);
• date – Date attributes are handled internally as numeric attributes, but in order to parse and present the date value correctly, the format of the date needs to be specified. The date and time patterns are explained in detail in the Javadoc of the java.text.SimpleDateFormat class. In the following, an example of how to create a date attribute using a date format of 4-digit year, 2-digit month and a 2-digit day, separated by hyphens:
Attribute date = new Attribute(“name_of_attr”, “yyyy-MM-dd”);
• nominal – Since nominal attributes contain predefined labels, one needs to supply these, stored in form of a weka.core.FastVector object:
FastVector labels = new FastVector(); labels.addElement(“label_a”); labels.addElement(“label_b”); labels.addElement(“label_c”); labels.addElement(“label_d”);
Attribute nominal = new Attribute(“name_of_attr”, labels);
• string – In contrast to nominal attributes, this type does not store a predefined list of labels. Normally used to store textual data, i.e., content of documents for text categorization. The same constructor as for the nominal attribute is used, but a null value is provided instead of an instance of FastVector:
Attribute string = new Attribute(“name_of_attr”, (FastVector) null);
16.3.
CREATING DATASETS IN MEMORY 203
•
relational – This attribute just takes another weka.core.Instances object for defining the relational structure in the constructor. The follow- ing code snippet generates a relational attribute that contains a relation with two attributes, a numeric and a nominal attribute:
FastVector atts = new FastVector();
atts.addElement(new Attribute(“rel.num”));
FastVector values = new FastVector(); values.addElement(“val_A”);
values.addElement(“val_B”);
values.addElement(“val_C”);
atts.addElement(new Attribute(“rel.nom”, values));
Instances rel_struct = new Instances(“rel”, atts, 0);
Attribute relational = new Attribute(“name_of_attr”, rel_struct);
A weka.core.Instances object is then created by supplying a FastVector object containing all the attribute objects. The following example creates a dataset with two numeric attributes and a nominal class attribute with two labels “no” and “yes”:
Attribute num1 = new Attribute(“num1”); Attribute num2 = new Attribute(“num2”); FastVector labels = new FastVector(); labels.addElement(“no”); labels.addElement(“yes”);
Attribute cls = new Attribute(“class”, labels); FastVector attributes = new FastVector(); attributes.addElement(num1); attributes.addElement(num2); attributes.addElement(cls);
Instances dataset = new Instances(“Test-dataset”, attributes, 0);
The final argument in the Instances constructor above tells WEKA how much memory to reserve for upcoming weka.core.Instance objects. If one knows how many rows will be added to the dataset, then it should be specified as it saves costly operations for expanding the internal storage. It doesn’t matter, if one aims to high with the amount of rows to be added, it is always possible to trim the dataset again, using the compactify() method.
16.3.2 Adding data
After the structure of the dataset has been defined, one can add the actual data to it, row by row. There are basically two constructors of the weka.core.Instance class that one can use for this purpose:
• Instance(double weight, double[] attValues) – generates an Instance object with the specified weight and the given double values. WEKA’s internal format is using doubles for all attribute types. For nominal, string and relational attributes this is just an index of the stored values.
• Instance(int numAttributes) – generates a new Instance object with weight 1.0 and all missing values.
The second constructor may be easier to use, but setting values via the Instance class’ methods is a bit costly, especially if one is adding a lot of rows. There- fore, the following code examples cover the first constructor. For simplicity, an Instances object “data” based on the code snippets for the different attribute introduced used above is used, as it contains all possible attribute types.
204 CHAPTER 16. USING THE API
For each instance, the first step is to create a new double array to hold the attribute values. It is important not to reuse this array, but always create a new one, since WEKA only references it and does not create a copy of it, when instantiating the Instance object. Reusing means changing the previously generated Instance object:
double[] values = new double[data.numAttributes()];
After that, the double array is filled with the actual values:
• numeric – just sets the numeric value:
values[0] = 1.23;
• date – turns the date string into a double value:
values[1] = data.attribute(1).parseDate(“2001-11-09”);
• nominal – determines the index of the label:
values[2] = data.attribute(2).indexOf(“label_b”);
• string – determines the index of the string, using the addStringValue method (internally, a hashtable holds all the string values):
values[3] = data.attribute(3).addStringValue(“This is a string”);
• relational – first, a new Instances object based on the attribute’s rela- tional definition has to be created, before the index of it can be determined, using the addRelation method:
Instances dataRel = new Instances(data.attribute(4).relation(),0); valuesRel = new double[dataRel.numAttributes()];
valuesRel[0] = 2.34;
valuesRel[1] = dataRel.attribute(1).indexOf(“val_C”); dataRel.add(new Instance(1.0, valuesRel));
values[4] = data.attribute(4).addRelation(dataRel);
Finally, an Instance object is generated with the initialized double array and
added to the dataset:
Instance inst = new Instance(1.0, values); data.add(inst);
16.4. GENERATING ARTIFICIAL DATA 205 16.4 Generating artificial data
Using WEKA’s data generators it is very easy to generate artificial datasets. There are two possible approaches to generating data, which get discussed in turn in the following sections.
16.4.1 Generate ARFF file
Simply generating an ARFF file is achieved using the static DataGenerator.makeData method. In order to write to a file, the generator needs to have a java.io.PrintWriter object for writing to, in this case a FileWriter.
The code below writes data generated by RDG1 to the file rdg1.arff :
import weka.datagenerators.DataGenerator;
import weka.datagenerators.classifiers.classification.RDG1; …
// configure generator
RDG1 generator = new RDG1();
generator.setMaxRuleSize(5);
// set where to write output to
java.io.PrintWriter output = new java.io.PrintWriter(
new java.io.BufferedWriter(new java.io.FileWriter(“rdg1.arff”))); generator.setOutput(output);
DataGenerator.makeData(generator, generator.getOptions()); output.flush();
output.close();
16.4.2 Generate Instances
Rather than writing the artificial data directly to a file, it is possible to ob- tain the data in the form of Instance/Instances directly. Depending on the generator, the data can be retrieved instance by instance (determined by getSingleModeFlag()), or as full dataset.
The example below generates data using the Agrawal generator:
import weka.datagenerators.classifiers.classification.Agrawal; …
// configure data generator
Agrawal generator = new Agrawal(); generator.setBalanceClass(true);
// initialize dataset and get header generator.setDatasetFormat(generator.defineDataFormat()); Instances header = generator.getDatasetFormat();
// generate data
if (generator.getSingleModeFlag()) {
for (int i = 0; i < generator.getNumExamplesAct(); i++) { Instance inst = generator.generateExample();
}
} else {
Instances data = generator.generateExamples(); }
206 CHAPTER 16. USING THE API 16.5 Randomizing data
Since learning algorithms can be prone to the order the data arrives in, random- izing (also called “shuffling”) the data is a common approach to alleviate this problem. Especially repeated randomizations, e.g., as during cross-validation, help to generate more realistic statistics.
WEKA offers two possibilities for randomizing a dataset:
• Using the randomize(Random) method of the weka.core.Instances ob- ject containing the data itself. This method requires an instance of the java.util.Random class. How to correctly instantiate such an object is explained below.
• UsingtheRandomizefilter(packageweka.filters.unsupervised.instance). For more information on how to use filters, see section 16.6.
A very important aspect of Machine Learning experiments is, that experiments have to be repeatable. Subsequent runs of the same experiment setup have to yield the exact same results. It may seem weird, but randomization is still possible in this scenario. Random number generates never return a completely random sequence of numbers anyway, only a pseudo-random one. In order to achieve repeatable pseudo-random sequences, seeded generators are used. Using the same seed value will always result in the same sequence then.
The default constructor of the java.util.Random random number generator class should never be used, as such created objects will generate most likely different sequences. The constructor Random(long), using a specified seed value, is the recommended one to use.
In order to get a more dataset-dependent randomization of the data, the getRandomNumberGenerator(int) method of the weka.core.Instances class can be used. This method returns a java.util.Random object that was seeded with the sum of the supplied seed and the hashcode of the string representation of a randomly chosen weka.core.Instance of the Instances object (using a random number generator seeded with the seed supplied to this method).
16.6. FILTERING 207 16.6 Filtering
In WEKA, filters are used to preprocess the data. They can be found below package weka.filters. Each filter falls into one of the following two categories:
• supervised – The filter requires a class attribute to be set.
• unsupervised – A class attribute is not required to be present.
And into one of the two sub-categories:
• attribute-based – Columns are processed, e.g., added or removed.
• instance-based – Rows are processed, e.g., added or deleted.
These categories should make it clear, what the difference between the two Discretize filters in WEKA is. The supervised one takes the class attribute and its distribution over the dataset into account, in order to determine the optimal number and size of bins, whereas the unsupervised one relies on a user- specified number of bins.
Apart from this classification, filters are either stream- or batch-based. Stream filters can process the data straight away and make it immediately available for collection again. Batch filters, on the other hand, need a batch of data to setup their internal data structures. The Add filter (this filter can be found in the weka.filters.unsupervised.attribute package) is an example of a stream filter. Adding a new attribute with only missing values does not require any sophisticated setup. However, the ReplaceMissingValues filter (same package as the Add filter) needs a batch of data in order to determine the means and modes for each of the attributes. Otherwise, the filter will not be able to replace the missing values with meaningful values. But as soon as a batch filter has been initialized with the first batch of data, it can also process data on a row-by-row basis, just like a stream filter.
Instance-based filters are a bit special in the way they handle data. As mentioned earlier, all filters can process data on a row-by-row basis after the first batch of data has been passed through. Of course, if a filter adds or removes rows from a batch of data, this no longer works when working in single-row processing mode. This makes sense, if one thinks of a scenario involving the FilteredClassifier meta-classifier: after the training phase (= first batch of data), the classifier will get evaluated against a test set, one instance at a time. If the filter now removes the only instance or adds instances, it can no longer be evaluated correctly, as the evaluation expects to get only a single result back. This is the reason why instance-based filters only pass through any subsequent batch of data without processing it. The Resample filters, for instance, act like this.
One can find example classes for filtering in the wekaexamples.filters package of the Weka Examples collection[3].
208 CHAPTER 16. USING THE API
The following example uses the Remove filter (the filter is located in package weka.filters.unsupervised.attribute) to remove the first attribute from a dataset. For setting the options, the setOptions(String[]) method is used.
import weka.core.Instances;
import weka.filters.Filter;
import weka.filters.unsupervised.attribute.Remove; ...
String[] options = new String[2];
options[0] = "-R";
options[1] = "1";
Remove remove = new Remove(); remove.setOptions(options); remove.setInputFormat(data);
Instances newData = Filter.useFilter(data, remove); // apply filter
A common trap to fall into is setting options after the setInputFormat(Instances) has been called. Since this method is (normally) used to determine the output format of the data, all the options have to be set before calling it. Otherwise,
all options set afterwards will be ignored.
16.6.1 Batch filtering
Batch filtering is necessary if two or more datasets need to be processed accord- ing to the same filter initialization. If batch filtering is not used, for instance when generating a training and a test set using the StringToWordVector fil- ter (package weka.filters.unsupervised.attribute), then these two filter runs are completely independent and will create two most likely incompatible datasets. Running the StringToWordVector on two different datasets, this will result in two different word dictionaries and therefore different attributes being generated.
The following code example shows how to standardize, i.e., transforming all numeric attributes to have zero mean and unit variance, a training and a test set with the Standardize filter (package weka.filters.unsupervised.attribute):
Instances train = ... // from somewhere Instances test = ... // from somewhere Standardize filter = new Standardize();
// initializing the filter once with training set filter.setInputFormat(train);
// configures the Filter based on train instances and returns // filtered instances
Instances newTrain = Filter.useFilter(train, filter);
// create new test set
Instances newTest = Filter.useFilter(test, filter);
// "range"
// first attribute
// new instance of filter
// set options
// inform filter about dataset // **AFTER** setting options
16.6. FILTERING 209 16.6.2 Filtering on-the-fly
Even though using the API gives one full control over the data and makes it eas-
ier to juggle several datasets at the same time, filtering data on-the-fly makes
life even easier. This handy feature is available through meta schemes in WEKA,
like FilteredClassifier (package weka.classifiers.meta), FilteredClusterer (package weka.clusterers), FilteredAssociator (package weka.associations) and FilteredAttributeEval/FilteredSubsetEval (in weka.attributeSelection). Instead of filtering the data beforehand, one just sets up a meta-scheme and lets
the meta-scheme do the filtering for one.
The following example uses the FilteredClassifier in conjunction with
the Remove filter to remove the first attribute (which happens to be an ID attribute) from the dataset and J48 (J48 is WEKA’s implementation of C4.5; package weka.classifiers.trees) as base-classifier. First the classifier is built with a training set and then evaluated with a separate test set. The actual and predicted class values are printed in the console. For more information on classification, see chapter 16.7.
import weka.classifiers.meta.FilteredClassifier; import weka.classifiers.trees.J48;
import weka.core.Instances;
import weka.filters.unsupervised.attribute.Remove; ...
Instances train = ... // from somewhere Instances test = ... // from somewhere
// filter
Remove rm = new Remove(); rm.setAttributeIndices("1"); // remove 1st attribute // classifier
J48 j48 = new J48();
j48.setUnpruned(true); // using an unpruned J48 // meta-classifier
FilteredClassifier fc = new FilteredClassifier(); fc.setFilter(rm);
fc.setClassifier(j48);
// train and output model
fc.buildClassifier(train);
System.out.println(fc);
for (int i = 0; i < test.numInstances(); i++) {
double pred = fc.classifyInstance(test.instance(i)); double actual = test.instance(i).classValue(); System.out.print("ID: "
+ test.instance(i).value(0)); System.out.print(", actual: "
+ test.classAttribute().value((int) actual)); System.out.println(", predicted: "
+ test.classAttribute().value((int) pred)); }
210 CHAPTER 16. USING THE API 16.7 Classification
Classification and regression algorithms in WEKA are called “classifiers” and are located below the weka.classifiers package. This section covers the following topics:
• Building a classifier – batch and incremental learning.
• Evaluating a classifier – various evaluation techniques and how to obtain
the generated statistics.
• Classifying instances – obtaining classifications for unknown data.
The Weka Examples collection[3] contains example classes covering classification in the wekaexamples.classifiers package.
16.7.1 Building a classifier
By design, all classifiers in WEKA are batch-trainable, i.e., they get trained on the whole dataset at once. This is fine, if the training data fits into memory. But there are also algorithms available that can update their internal model on-the-go. These classifiers are called incremental. The following two sections cover the batch and the incremental classifiers.
Batch classifiers
A batch classifier is really simple to build:
• set options – either using the setOptions(String[]) method or the ac- tual set-methods.
• train it – calling the buildClassifier(Instances) method with the training set. By definition, the buildClassifier(Instances) method resets the internal model completely, in order to ensure that subsequent calls of this method with the same data result in the same model (“re- peatable experiments”).
The following code snippet builds an unpruned J48 on a dataset:
import weka.core.Instances; import weka.classifiers.trees.J48; ...
Instances data = ...
String[] options = new String[1]; options[0] = "-U";
J48 tree = new J48(); tree.setOptions(options); tree.buildClassifier(data);
Incremental classifiers
All incremental classifiers in WEKA implement the interface UpdateableClassifier (located in package weka.classifiers). Bringing up the Javadoc for this par- ticular interface tells one what classifiers implement this interface. These classi-
fiers can be used to process large amounts of data with a small memory-footprint,
as the training data does not have to fit in memory. ARFF files, for instance, can be read incrementally (see chapter 16.2).
// from somewhere
// unpruned tree
// new instance of tree // set the options
// build classifier
16.7. CLASSIFICATION 211 Training an incremental classifier happens in two stages:
1. initializethemodelbycallingthebuildClassifier(Instances)method. One can either use a weka.core.Instances object with no actual data or one with an initial set of data.
2. updatethemodelrow-by-row,bycallingtheupdateClassifier(Instance) method.
The following example shows how to load an ARFF file incrementally using the ArffLoader class and train the NaiveBayesUpdateable classifier with one row at a time:
import weka.core.converters.ArffLoader;
import weka.classifiers.bayes.NaiveBayesUpdateable; import java.io.File;
...
// load data
ArffLoader loader = new ArffLoader(); loader.setFile(new File("/some/where/data.arff")); Instances structure = loader.getStructure(); structure.setClassIndex(structure.numAttributes() - 1);
// train NaiveBayes
NaiveBayesUpdateable nb = new NaiveBayesUpdateable(); nb.buildClassifier(structure);
Instance current;
while ((current = loader.getNextInstance(structure)) != null)
nb.updateClassifier(current);
212 CHAPTER 16. USING THE API 16.7.2 Evaluating a classifier
Building a classifier is only one part of the equation, evaluating how well it performs is another important part. WEKA supports two types of evaluation:
• Cross-validation – If one only has a single dataset and wants to get a reasonable realistic evaluation. Setting the number of folds equal to the number of rows in the dataset will give one leave-one-out cross-validation (LOOCV).
• Dedicated test set – The test set is solely used to evaluate the built clas- sifier. It is important to have a test set that incorporates the same (or similar) concepts as the training set, otherwise one will always end up with poor performance.
The evaluation step, including collection of statistics, is performed by the Evaluation class (package weka.classifiers).
Cross-validation
The crossValidateModel method of the Evaluation class is used to perform cross-validation with an untrained classifier and a single dataset. Supplying an untrained classifier ensures that no information leaks into the actual evaluation. Even though it is an implementation requirement, that the buildClassifier method resets the classifier, it cannot be guaranteed that this is indeed the case (“leaky” implementation). Using an untrained classifier avoids unwanted side- effects, as for each train/test set pair, a copy of the originally supplied classifier is used.
Before cross-validation is performed, the data gets randomized using the supplied random number generator (java.util.Random). It is recommended that this number generator is “seeded” with a specified seed value. Otherwise, subsequent runs of cross-validation on the same dataset will not yield the same results, due to different randomization of the data (see section 16.5 for more information on randomization).
The code snippet below performs 10-fold cross-validation with a J48 decision tree algorithm on a dataset newData, with random number generator that is seeded with “1”. The summary of the collected statistics is output to stdout.
16.7. CLASSIFICATION 213
import weka.classifiers.Evaluation; import weka.classifiers.trees.J48; import weka.core.Instances;
import java.util.Random;
...
Instances newData = ... // from somewhere
Evaluation eval = new Evaluation(newData);
J48 tree = new J48();
eval.crossValidateModel(tree, newData, 10, new Random(1)); System.out.println(eval.toSummaryString("\nResults\n\n", false));
The Evaluation object in this example is initialized with the dataset used in the evaluation process. This is done in order to inform the evaluation about the type of data that is being evaluated, ensuring that all internal data structures are setup correctly.
Train/test set
Using a dedicated test set to evaluate a classifier is just as easy as cross- validation. But instead of providing an untrained classifier, a trained classifier has to be provided now. Once again, the weka.classifiers.Evaluation class is used to perform the evaluation, this time using the evaluateModel method.
The code snippet below trains a J48 with default options on a training set and evaluates it on a test set before outputting the summary of the collected statistics:
import weka.core.Instances;
import weka.classifiers.Evaluation;
import weka.classifiers.trees.J48;
...
Instances train = ... // from somewhere
Instances test = ... // from somewhere
// train classifier
Classifier cls = new J48();
cls.buildClassifier(train);
// evaluate classifier and print some statistics
Evaluation eval = new Evaluation(train);
eval.evaluateModel(cls, test); System.out.println(eval.toSummaryString("\nResults\n\n", false));
214 CHAPTER 16. USING THE API Statistics
In the previous sections, the toSummaryString of the Evaluation class was already used in the code examples. But there are other summary methods for nominal class attributes available as well:
• toMatrixString – outputs the confusion matrix.
• toClassDetailsString – outputs TP/FP rates, precision, recall, F-measure,
AUC (per class).
• toCumulativeMarginDistributionString– outputs the cumulative mar-
gins distribution.
If one does not want to use these summary methods, it is possible to access the individual statistical measures directly. Below, a few common measures are listed:
• nominal class attribute
– correct() – The number of correctly classified instances. The in- correctly classified ones are available through incorrect().
– pctCorrect() – The percentage of correctly classified instances (ac- curacy). pctIncorrect() returns the number of misclassified ones.
– areaUnderROC(int) – The AUC for the specified class label index (0-based index).
• numeric class attribute
– correlationCoefficient() – The correlation coefficient.
• general
– meanAbsoluteError() – The mean absolute error.
– rootMeanSquaredError() – The root mean squared error.
– numInstances() – The number of instances with a class value. – unclassified() - The number of unclassified instances.
– pctUnclassified() - The percentage of unclassified instances.
For a complete overview, see the Javadoc page of the Evaluation class. By looking up the source code of the summary methods mentioned above, one can easily determine what methods are used for which particular output.
16.7. CLASSIFICATION 215 16.7.3 Classifying instances
After a classifier setup has been evaluated and proven to be useful, a built classi- fier can be used to make predictions and label previously unlabeled data. Section 16.6.2 already provided a glimpse of how to use a classifier’s classifyInstance method. This section here elaborates a bit more on this.
The following example uses a trained classifier tree to label all the instances in an unlabeled dataset that gets loaded from disk. After all the instances have been labeled, the newly labeled dataset gets written back to disk to a new file.
// load unlabeled data and set class attribute
Instances unlabeled = DataSource.read("/some/where/unlabeled.arff"); unlabeled.setClassIndex(unlabeled.numAttributes() - 1);
// create copy
Instances labeled = new Instances(unlabeled);
// label instances
for (int i = 0; i < unlabeled.numInstances(); i++) {
double clsLabel = tree.classifyInstance(unlabeled.instance(i));
labeled.instance(i).setClassValue(clsLabel); }
// save newly labeled data DataSink.write("/some/where/labeled.arff", labeled);
The above example works for classification and regression problems alike, as long as the classifier can handle numeric classes, of course. Why is that? The classifyInstance(Instance) method returns for numeric classes the regres- sion value and for nominal classes the 0-based index in the list of available class labels.
If one is interested in the class distribution instead, then one can use the distributionForInstance(Instance) method (this array sums up to 1). Of course, using this method makes only sense for classification problems. The code snippet below outputs the class distribution, the actual and predicted label side-by-side in the console:
// load data
Instances train = DataSource.read(args[0]); train.setClassIndex(train.numAttributes() - 1);
Instances test = DataSource.read(args[1]); test.setClassIndex(test.numAttributes() - 1);
// train classifier
J48 cls = new J48();
cls.buildClassifier(train);
// output predictions
System.out.println("# - actual - predicted - distribution"); for (int i = 0; i < test.numInstances(); i++) {
double pred = cls.classifyInstance(test.instance(i));
double[] dist = cls.distributionForInstance(test.instance(i)); System.out.print((i+1) + " - "); System.out.print(test.instance(i).toString(test.classIndex()) + " - "); System.out.print(test.classAttribute().value((int) pred) + " - "); System.out.println(Utils.arrayToString(dist));
}
216 CHAPTER 16. USING THE API
16.8. CLUSTERING 217 16.8 Clustering
Clustering is an unsupervised Machine Learning technique of finding patterns in the data, i.e., these algorithms work without class attributes. Classifiers, on the other hand, are supervised and need a class attribute. This section, similar to the one about classifiers, covers the following topics:
• Building a clusterer – batch and incremental learning.
• Evaluating a clusterer – how to evaluate a built clusterer.
• Clustering instances – determining what clusters unknown instances be-
long to.
Fully functional example classes are located in the wekaexamples.clusterers
package of the Weka Examples collection[3].
16.8.1 Building a clusterer
Clusterers, just like classifiers, are by design batch-trainable as well. They all can be built on data that is completely stored in memory. But a small subset of the cluster algorithms can also update the internal representation incrementally. The following two sections cover both types of clusterers.
Batch clusterers
Building a batch clusterer, just like a classifier, happens in two stages:
• set options – either calling the setOptions(String[]) method or the appropriate set-methods of the properties.
• build the model with training data – calling the buildClusterer(Instances) method. By definition, subsequent calls of this method must result in the same model (“repeatable experiments”). In other words, calling this method must completely reset the model.
Below is an example of building the EM clusterer with a maximum of 100 itera- tions. The options are set using the setOptions(String[]) method:
import weka.clusterers.EM;
import weka.core.Instances;
...
Instances data = ... // from somewhere
String[] options = new String[2];
options[0] = "-I"; // max. iterations options[1] = "100";
EM clusterer = new EM(); // new instance of clusterer clusterer.setOptions(options); // set the options clusterer.buildClusterer(data); // build the clusterer
Incremental clusterers
Incremental clusterers in WEKA implement the interface UpdateableClusterer (package weka.clusterers). Training an incremental clusterer happens in three stages, similar to incremental classifiers:
1. initialize the model by calling the buildClusterer(Instances) method. Once again, one can either use an empty weka.core.Instances object or one with an initial set of data.
2. updatethemodelrow-by-rowbycallingthetheupdateClusterer(Instance) method.
3. finish the training by calling updateFinished() method. In case cluster algorithms need to perform computational expensive post-processing or clean up operations.
218 CHAPTER 16. USING THE API An ArffLoader is used in the following example to build the Cobweb clusterer
incrementally:
import weka.clusterers.Cobweb;
import weka.core.Instance;
import weka.core.Instances;
import weka.core.converters.ArffLoader; ...
// load data
ArffLoader loader = new ArffLoader();
loader.setFile(new File("/some/where/data.arff"));
Instances structure = loader.getStructure();
// train Cobweb
Cobweb cw = new Cobweb();
cw.buildClusterer(structure);
Instance current;
while ((current = loader.getNextInstance(structure)) != null)
cw.updateClusterer(current); cw.updateFinished();
16.8. CLUSTERING 219 16.8.2 Evaluating a clusterer
Evaluation of clusterers is not as comprehensive as the evaluation of classi- fiers. Since clustering is unsupervised, it is also a lot harder determining how good a model is. The class used for evaluating cluster algorithms, is ClusterEvaluation (package weka.clusterers).
In order to generate the same output as the Explorer or the command-line, one can use the evaluateClusterer method, as shown below:
import weka.clusterers.EM;
import weka.clusterers.ClusterEvaluation;
...
String[] options = new String[2];
options[0] = "-t";
options[1] = "/some/where/somefile.arff"; System.out.println(ClusterEvaluation.evaluateClusterer(new EM(), options));
Or, if the dataset is already present in memory, one can use the following ap- proach:
import weka.clusterers.ClusterEvaluation; import weka.clusterers.EM;
import weka.core.Instances;
...
Instances data = ... // from somewhere
EM cl = new EM();
cl.buildClusterer(data);
ClusterEvaluation eval = new ClusterEvaluation(); eval.setClusterer(cl);
eval.evaluateClusterer(new Instances(data)); System.out.println(eval.clusterResultsToString());
Density based clusterers, i.e., algorithms that implement the interface named DensityBasedClusterer (package weka.clusterers) can be cross-validated and the log-likelyhood obtained. Using the MakeDensityBasedClusterer meta- clusterer, any non-density based clusterer can be turned into such. Here is an
example of likelyhood:
import
import
import
import
...
Instances data = ... DensityBasedClusterer clusterer = new ... double logLikelyhood = ClusterEvaluation.crossValidateModel(
clusterer, data, 10, new Random(1));
cross-validating a density based clusterer and obtaining the log-
weka.clusterers.ClusterEvaluation; weka.clusterers.DensityBasedClusterer; weka.core.Instances; java.util.Random;
// from somewhere
// the clusterer to evaluate
// cross-validate
// with 10 folds
// and random number generator // with seed 1
220 CHAPTER 16. USING THE API Classes to clusters
Datasets for supervised algorithms, like classifiers, can be used to evaluate a clusterer as well. This evaluation is called classes-to-clusters, as the clusters are mapped back onto the classes.
This type of evaluation is performed as follows:
1. create a copy of the dataset containing the class attribute and remove the class attribute, using the Remove filter (this filter is located in package weka.filters.unsupervised.attribute).
2. build the clusterer with this new data.
3. evaluate the clusterer now with the original data.
And here are the steps translated into code, using EM as the clusterer being evaluated:
1. create a copy of data without class attribute
Instances data = ... // from somewhere
Remove filter = new Remove(); filter.setAttributeIndices("" + (data.classIndex() + 1)); filter.setInputFormat(data);
Instances dataClusterer = Filter.useFilter(data, filter);
2. build the clusterer
EM clusterer = new EM();
// set further options for EM, if necessary... clusterer.buildClusterer(dataClusterer);
3. evaluate the clusterer
ClusterEvaluation eval = new ClusterEvaluation(); eval.setClusterer(clusterer); eval.evaluateClusterer(data);
// print results System.out.println(eval.clusterResultsToString());
16.8. CLUSTERING 221 16.8.3 Clustering instances
Clustering of instances is very similar to classifying unknown instances when using classifiers. The following methods are involved:
• clusterInstance(Instance) – determines the cluster the Instance would belong to.
• distributionForInstance(Instance)– predicts the cluster membership for this Instance. The sum of this array adds up to 1.
The code fragment outlined below trains an EM clusterer on one dataset and outputs for a second dataset the predicted clusters and cluster memberships of the individual instances:
import weka.clusterers.EM;
import weka.core.Instances;
...
Instances dataset1 = ... // from somewhere Instances dataset2 = ... // from somewhere // build clusterer
EM clusterer = new EM(); clusterer.buildClusterer(dataset1);
// output predictions
System.out.println("# - cluster - distribution"); for (int i = 0; i < dataset2.numInstances(); i++) {
int cluster = clusterer.clusterInstance(dataset2.instance(i));
double[] dist = clusterer.distributionForInstance(dataset2.instance(i)); System.out.print((i+1));
System.out.print(" - ");
System.out.print(cluster);
System.out.print(" - ");
System.out.print(Utils.arrayToString(dist));
System.out.println();
}
222 CHAPTER 16. USING THE API 16.9 Selecting attributes
Preparing one’s data properly is a very important step for getting the best re- sults. Reducing the number of attributes can not only help speeding up runtime with algorithms (some algorithms’ runtime are quadratic in regards to number of attributes), but also help avoid “burying” the algorithm in a mass of at- tributes, when only a few are essential for building a good model.
There are three different types of evaluators in WEKA at the moment:
• single attribute evaluators – perform evaluations on single attributes. These classes implement the weka.attributeSelection.AttributeEvaluator interface. The Ranker search algorithm is usually used in conjunction with these algorithms.
• attribute subset evaluators – work on subsets of all the attributes in the dataset. The weka.attributeSelection.SubsetEvaluator interface is implemented by these evaluators.
• attribute set evaluators – evaluate sets of attributes. Not to be con- fused with the subset evaluators, as these classes are derived from the weka.attributeSelection.AttributeSetEvaluator superclass.
Most of the attribute selection schemes currently implemented are supervised, i.e., they require a dataset with a class attribute. Unsupervised evaluation algorithms are derived from one of the following superclasses:
• weka.attributeSelection.UnsupervisedAttributeEvaluator e.g., LatentSemanticAnalysis, PrincipalComponents
• weka.attributeSelection.UnsupervisedSubsetEvaluator none at the moment
Attribute selection offers filtering on-the-fly, like classifiers and clusterers, as well:
• weka.attributeSelection.FilteredAttributeEval – filter for evalua- tors that evaluate attributes individually.
• weka.attributeSelection.FilteredSubsetEval – for filtering evalua- tors that evaluate subsets of attributes.
So much about the differences among the various attribute selection algorithms and back to how to actually perform attribute selection. WEKA offers three different approaches:
• Using a meta-classifier – for performing attribute selection on-the-fly (sim- ilar to FilteredClassifier’s filtering on-the-fly).
• Using a filter - for preprocessing the data.
• Low-level API usage - instead of using the meta-schemes (classifier or
filter), one can use the attribute selection API directly as well.
The following sections cover each of the topics, accompanied with a code exam- ple. For clarity, the same evaluator and search algorithm is used in all of these examples.
Feel free to check out the example classes of the Weka Examples collection[3], located in the wekaexamples.attributeSelection package.
16.9. SELECTING ATTRIBUTES 223 16.9.1 Using the meta-classifier
The meta-classifier AttributeSelectedClassifier (this classifier is located in package weka.classifiers.meta), is similar to the FilteredClassifier. But instead of taking a base-classifier and a filter as parameters to perform the filtering, the AttributeSelectedClassifier uses a search algorithm (derived from weka.attributeSelection.ASEvaluation), an evaluator (superclass is weka.attributeSelection.ASSearch) to perform the attribute selection and a base-classifier to train on the reduced data.
This example here uses J48 as base-classifier, CfsSubsetEval as evaluator and a backwards operating GreedyStepwise as search method:
import weka.attributeSelection.CfsSubsetEval;
import weka.attributeSelection.GreedyStepwise;
import weka.classifiers.Evaluation;
import weka.classifiers.meta.AttributeSelectedClassifier; import weka.classifiers.trees.J48;
import weka.core.Instances;
...
Instances data = ... // from somewhere
// setup meta-classifier
AttributeSelectedClassifier classifier = new AttributeSelectedClassifier(); CfsSubsetEval eval = new CfsSubsetEval();
GreedyStepwise search = new GreedyStepwise(); search.setSearchBackwards(true);
J48 base = new J48();
classifier.setClassifier(base);
classifier.setEvaluator(eval);
classifier.setSearch(search);
// cross-validate classifier
Evaluation evaluation = new Evaluation(data); evaluation.crossValidateModel(classifier, data, 10, new Random(1)); System.out.println(evaluation.toSummaryString());
224 CHAPTER 16. USING THE API 16.9.2 Using the filter
In case the data only needs to be reduced in dimensionality, but not used for training a classifier, then the filter approach is the right one. The AttributeSelection filter (package weka.filters.supervised.attribute) takes an evaluator and
a search algorithm as parameter.
The code snippet below uses once again CfsSubsetEval as evaluator and a backwards operating GreedyStepwise as search algorithm. It just outputs the reduced data to stdout after the filtering step:
import weka.attributeSelection.CfsSubsetEval; import weka.attributeSelection.GreedyStepwise; import weka.core.Instances;
import weka.filters.Filter;
import weka.filters.supervised.attribute.AttributeSelection; ...
Instances data = ... // from somewhere
// setup filter
AttributeSelection filter = new AttributeSelection(); CfsSubsetEval eval = new CfsSubsetEval(); GreedyStepwise search = new GreedyStepwise(); search.setSearchBackwards(true); filter.setEvaluator(eval);
filter.setSearch(search); filter.setInputFormat(data);
// filter data
Instances newData = Filter.useFilter(data, filter); System.out.println(newData);
16.9. SELECTING ATTRIBUTES 225 16.9.3 Using the API directly
Using the meta-classifier or the filter approach makes attribute selection fairly easy. But it might not satify everybody’s needs. For instance, if one wants to obtain the ordering of the attributes (using Ranker) or retrieve the indices of the selected attributes instead of the reduced data.
Just like the other examples, the one shown here uses the CfsSubsetEval evaluator and the GreedyStepwise search algorithm (in backwards mode). But instead of outputting the reduced data, only the selected indices are printed in the console:
import weka.attributeSelection.AttributeSelection; import weka.attributeSelection.CfsSubsetEval; import weka.attributeSelection.GreedyStepwise; import weka.core.Instances;
...
Instances data = ... // from somewhere
// setup attribute selection
AttributeSelection attsel = new AttributeSelection(); CfsSubsetEval eval = new CfsSubsetEval(); GreedyStepwise search = new GreedyStepwise(); search.setSearchBackwards(true); attsel.setEvaluator(eval);
attsel.setSearch(search);
// perform attribute selection attsel.SelectAttributes(data);
int[] indices = attsel.selectedAttributes(); System.out.println(
"selected attribute indices (starting with 0):\n" + Utils.arrayToString(indices));
226 CHAPTER 16. USING THE API 16.10 Saving data
Saving weka.core.Instances objects is as easy as reading the data in the first place, though the process of storing the data again is far less common than of reading the data into memory. The following two sections cover how to save the data in files and in databases.
Just like with loading the data in chapter 16.2, examples classes for saving data can be found in the wekaexamples.core.converters package of the Weka Examples collection[3];
16.10.1 Saving data to files
Once again, one can either let WEKA choose the appropriate converter for sav-
ing the data or use an explicit converter (all savers are located in the weka.core.converters package). The latter approach is necessary, if the file name under which the data
will be stored does not have an extension that WEKA recognizes.
Use the DataSink class (inner class of weka.core.converters.ConverterUtils), if the extensions are not a problem. Here are a few examples:
import weka.core.Instances;
import weka.core.converters.ConverterUtils.DataSink; ...
// data structure to save
Instances data = ...
// save as ARFF DataSink.write("/some/where/data.arff", data);
// save as CSV DataSink.write("/some/where/data.csv", data);
And here is an example of using the CSVSaver converter explicitly:
import weka.core.Instances;
import weka.core.converters.CSVSaver;
import java.io.File;
...
// data structure to save
Instances data = ...
// save as CSV
CSVSaver saver = new CSVSaver(); saver.setInstances(data);
saver.setFile(new File("/some/where/data.csv")); saver.writeBatch();
16.10.2 Saving data to databases
Apart from the KnowledgeFlow, saving to databases is not very obvious in WEKA, unless one knows about the DatabaseSaver converter. Just like the DatabaseLoader, the saver counterpart can store the data either in batch mode or incrementally as well.
16.10. SAVING DATA 227 The first example shows how to save the data in batch mode, which is the
easier way of doing it:
import weka.core.Instances;
import weka.core.converters.DatabaseSaver;
...
// data structure to save
Instances data = ...
// store data in database
DatabaseSaver saver = new DatabaseSaver(); saver.setDestination("jdbc_url", "the_user", "the_password"); // we explicitly specify the table name here: saver.setTableName("whatsoever2"); saver.setRelationForTableName(false);
// or we could just update the name of the dataset:
// saver.setRelationForTableName(true);
// data.setRelationName("whatsoever2"); saver.setInstances(data);
saver.writeBatch();
Saving the data incrementally, requires a bit more work, as one has to specify that writing the data is done incrementally (using the setRetrieval method), as well as notifying the saver when all the data has been saved:
import weka.core.Instances;
import weka.core.converters.DatabaseSaver;
...
// data structure to save
Instances data = ...
// store data in database
DatabaseSaver saver = new DatabaseSaver(); saver.setDestination("jdbc_url", "the_user", "the_password"); // we explicitly specify the table name here: saver.setTableName("whatsoever2"); saver.setRelationForTableName(false);
// or we could just update the name of the dataset:
// saver.setRelationForTableName(true);
// data.setRelationName("whatsoever2"); saver.setRetrieval(DatabaseSaver.INCREMENTAL); saver.setStructure(data);
count = 0;
for (int i = 0; i < data.numInstances(); i++) {
saver.writeIncremental(data.instance(i)); }
// notify saver that we’re finished saver.writeIncremental(null);
228 CHAPTER 16. USING THE API 16.11 Visualization
The concepts covered in this chapter are also available through the example classes of the Weka Examples collection[3]. See the following packages:
• wekaexamples.gui.graphvisualizer • wekaexamples.gui.treevisualizer • wekaexamples.gui.visualize
16.11.1 ROC curves
WEKA can generate “Receiver operating characteristic” (ROC) curves, based on the collected predictions during an evaluation of a classifier. In order to display a ROC curve, one needs to perform the following steps:
1. Generate the plotable data based on the Evaluation’s collected predic-
tions, using the ThresholdCurve class (package weka.classifiers.evaluation).
2. Put the plotable data into a plot container, an instance of the PlotData2D class (package weka.gui.visualize).
3. Add the plot container to a visualization panel for displaying the data, an instance of the ThresholdVisualizePanel class (package weka.gui.visualize).
4. Add the visualization panel to a JFrame (package javax.swing) and dis- play it.
And now, the four steps translated into actual code:
1. Generate the plotable data
Evaluation eval = ... // from somewhere
ThresholdCurve tc = new ThresholdCurve();
int classIndex = 0; // ROC for the 1st class label
Instances curve = tc.getCurve(eval.predictions(), classIndex);
2. Put the plotable into a plot container
PlotData2D plotdata = new PlotData2D(curve); plotdata.setPlotName(curve.relationName()); plotdata.addInstanceNumberAttribute();
3. Add the plot container to a visualization panel
ThresholdVisualizePanel tvp = new ThresholdVisualizePanel(); tvp.setROCString("(Area under ROC = " +
Utils.doubleToString(ThresholdCurve.getROCArea(curve),4)+")"); tvp.setName(curve.relationName());
tvp.addPlot(plotdata);
4. Add the visualization panel to a JFrame
final JFrame jf = new JFrame("WEKA ROC: " + tvp.getName()); jf.setSize(500,400);
jf.getContentPane().setLayout(new BorderLayout()); jf.getContentPane().add(tvp, BorderLayout.CENTER); jf.setDefaultCloseOperation(JFrame.DISPOSE_ON_CLOSE); jf.setVisible(true);
16.11. VISUALIZATION 229 16.11.2 Graphs
Classes implementing the weka.core.Drawable interface can generate graphs of their internal models which can be displayed. There are two different types of graphs available at the moment, which are explained in the subsequent sections:
• Tree – decision trees.
• BayesNet – bayesian net graph structures.
16.11.2.1 Tree
It is quite easy to display the internal tree structure of classifiers like J48 or M5P (package weka.classifiers.trees). The following example builds a J48 classifier on a dataset and displays the generated tree visually using the TreeVisualizer class (package weka.gui.treevisualizer). This visu- alization class can be used to view trees (or digraphs) in GraphViz’s DOT language[26].
import weka.classifiers.trees.J48;
import weka.core.Instances;
import weka.gui.treevisualizer.PlaceNode2; import weka.gui.treevisualizer.TreeVisualizer; import java.awt.BorderLayout;
import javax.swing.JFrame;
...
Instances data = ... // from somewhere
// train classifier
J48 cls = new J48(); cls.buildClassifier(data);
// display tree
TreeVisualizer tv = new TreeVisualizer(
null, cls.graph(), new PlaceNode2());
JFrame jf = new JFrame("Weka Classifier Tree Visualizer: J48"); jf.setDefaultCloseOperation(JFrame.DISPOSE_ON_CLOSE); jf.setSize(800, 600);
jf.getContentPane().setLayout(new BorderLayout()); jf.getContentPane().add(tv, BorderLayout.CENTER); jf.setVisible(true);
// adjust tree
tv.fitToScreen();
230 CHAPTER 16. USING THE API 16.11.2.2 BayesNet
The graphs that the BayesNet classifier (package weka.classifiers.bayes) generates can be displayed using the GraphVisualizer class (located in package weka.gui.graphvisualizer). The GraphVisualizer can display graphs that are either in GraphViz’s DOT language[26] or in XML BIF[20] format. For displaying DOT format, one needs to use the method readDOT, and for the BIF format the method readBIF.
The following code snippet trains a BayesNet classifier on some data and then displays the graph generated from this data in a frame:
import weka.classifiers.bayes.BayesNet;
import weka.core.Instances;
import weka.gui.graphvisualizer.GraphVisualizer; import java.awt.BorderLayout;
import javax.swing.JFrame;
...
Instances data = ... // from somewhere
// train classifier
BayesNet cls = new BayesNet(); cls.buildClassifier(data);
// display graph
GraphVisualizer gv = new GraphVisualizer(); gv.readBIF(cls.graph());
JFrame jf = new JFrame("BayesNet graph"); jf.setDefaultCloseOperation(JFrame.DISPOSE_ON_CLOSE); jf.setSize(800, 600); jf.getContentPane().setLayout(new BorderLayout()); jf.getContentPane().add(gv, BorderLayout.CENTER); jf.setVisible(true);
// layout graph
gv.layoutGraph();
16.12. SERIALIZATION 231 16.12 Serialization
Serialization2 is the process of saving an object in a persistent form, e.g., on the harddisk as a bytestream. Deserialization is the process in the opposite direction, creating an object from a persistently saved data structure. In Java, an object can be serialized if it imports the java.io.Serializable interface. Members of an object that are not supposed to be serialized, need to be declared with the keyword transient.
In the following are some Java code snippets for serializing and deserializing a J48 classifier. Of course, serialization is not limited to classifiers. Most schemes in WEKA, like clusterers and filters, are also serializable.
Serializing a classifier
The weka.core.SerializationHelper class makes it easy to serialize an ob- ject. For saving, one can use one of the write methods:
import weka.classifiers.Classifier;
import weka.classifiers.trees.J48;
import weka.core.converters.ConverterUtils.DataSource; import weka.core.SerializationHelper;
...
// load data
Instances inst = DataSource.read("/some/where/data.arff"); inst.setClassIndex(inst.numAttributes() - 1);
// train J48
Classifier cls = new J48();
cls.buildClassifier(inst);
// serialize model SerializationHelper.write("/some/where/j48.model", cls);
Deserializing a classifier
Deserializing an object can be achieved by using one of the read methods:
import weka.classifiers.Classifier; import weka.core.SerializationHelper; ...
// deserialize model
Classifier cls = (Classifier) SerializationHelper.read( "/some/where/j48.model");
2 http://en.wikipedia.org/wiki/Serialization
232 CHAPTER 16. USING THE API Deserializing a classifier saved from the Explorer
The Explorer does not only save the built classifier in the model file, but also the header information of the dataset the classifier was built with. By storing the dataset information as well, one can easily check whether a serialized classifier can be applied on the current dataset. The readAll methods returns an array with all objects that are contained in the model file.
import weka.classifiers.Classifier; import weka.core.Instances;
import weka.core.SerializationHelper; ...
// the current data to use with classifier
Instances current = ... // from somewhere
// deserialize model
Object o[] = SerializationHelper.readAll("/some/where/j48.model"); Classifier cls = (Classifier) o[0];
Instances data = (Instances) o[1]; // is the data compatible?
if (!data.equalHeaders(current))
throw new Exception("Incompatible data!");
Serializing a classifier for the Explorer
If one wants to serialize the dataset header information alongside the classifier, just like the Explorer does, then one can use one of the writeAll methods:
import weka.classifiers.Classifier;
import weka.classifiers.trees.J48;
import weka.core.converters.ConverterUtils.DataSource; import weka.core.SerializationHelper;
...
// load data
Instances inst = DataSource.read("/some/where/data.arff"); inst.setClassIndex(inst.numAttributes() - 1);
// train J48
Classifier cls = new J48();
cls.buildClassifier(inst);
// serialize classifier and header information
Instances header = new Instances(inst, 0); SerializationHelper.writeAll(
"/some/where/j48.model", new Object[]{cls, header});
Chapter 17 Extending WEKA
For most users, the existing WEKA framework will be sufficient to perform the task at hand, offering a wide range of filters, classifiers, clusterers, etc. Researchers, on the other hand, might want to add new algorithms and compare them against existing ones. The framework with its existing algorithms is not set in stone, but basically one big plugin framework. With WEKA’s automatic discovery of classes on the classpath, adding new classifiers, filters, etc. to the existing framework is very easy.
Though algorithms like clusterers, associators, data generators and attribute selection are not covered in this chapter, their implemention is very similar to the one of implementing a classifier. You basically choose a superclass to derive your new algorithm from and then implement additional interfaces, if necessary. Just check out the other algorithms that are already implemented.
The section covering the GenericObjectEditor (see chapter 18.4) shows you how to tell WEKA where to find your class(es) and therefore making it/them available in the GUI (Explorer/Experimenter) via the GenericObjectEditor.
233
234 CHAPTER 17. EXTENDING WEKA
17.1. WRITING A NEW CLASSIFIER 235 17.1 Writing a new Classifier
17.1.1 Choosing the base class
The ancestor of all classifiers in WEKA is weka.classifiers.Classifier, an abstract class. Your new classifier must be derived from this class at least to be visible through the GenericObjectEditor. But in order to make implemen- tations of new classifiers even easier, WEKA comes already with a range of other abstract classes derived from weka.classifiers.Classifier. In the fol- lowing you will find an overview that will help you decide what base class to use for your classifier. For better readability, the weka.classifiers prefix was dropped from the class names:
• simple classifier
– Classifier – not randomizable
– RandomizableClassifier – randomizable
• meta classifier
– single base classifier
∗ SingleClassifierEnhancer – not randomizable, not iterated
∗ RandomizableSingleClassifierEnhancer – randomizable, not
iterated
∗ IteratedSingleClassifierEnhancer – not randomizable, iter-
ated
∗ RandomizableIteratedSingleClassifierEnhancer – random-
izable, iterated
– multiple base classifiers
∗ MultipleClassifiersCombiner – not randomizable
∗ RandomizableMultipleClassifiersCombiner – randomizable
If you are still unsure about what superclass to choose, then check out the Javadoc of those superclasses. In the Javadoc you will find all the classifiers that are derived from it, which should give you a better idea whether this particular superclass is suited for your needs.
17.1.2 Additional interfaces
The abstract classes listed above basically just implement various combinations of the following two interfaces:
• weka.core.Randomizable – to allow (seeded) randomization taking place
• weka.classifiers.IterativeClassifier – to make the classifier an it-
erated one
But these interfaces are not the only ones that can be implemented by a classifier. Here is a list for further interfaces:
• weka.core.AdditionalMeasureProducer – the classifier returns addi- tional information, e.g., J48 returns the tree size with this method.
• weka.core.WeightedInstancesHandler – denotes that the classifier can make use of weighted Instance objects (the default weight of an Instance is 1.0).
• weka.core.TechnicalInformationHandler – for returning paper refer- ences and publications this classifier is based on.
• weka.classifiers.Sourcable – classifiers implementing this interface can return Java code of a built model, which can be used elsewhere.
• weka.classifiers.UpdateableClassifier – for classifiers that can be trained incrementally, i.e., row by row like NaiveBayesUpdateable.
236 CHAPTER 17. EXTENDING WEKA 17.1.3 Packages
A few comments about the different sub-packages in the weka.classifiers package:
• bayes – contains bayesian classifiers, e.g., NaiveBayes
• evaluation – classes related to evaluation, e.g., confusion matrix, thresh-
old curve (= ROC)
• functions – e.g., Support Vector Machines, regression algorithms, neural
nets
• lazy – “learning” is performed at prediction time, e.g., k-nearest neighbor
(k-NN)
• meta – meta-classifiers that use a base one or more classifiers as input,
e.g., boosting, bagging or stacking
• mi – classifiers that handle multi-instance data
• misc – various classifiers that don’t fit in any another category
• rules – rule-based classifiers, e.g., ZeroR
• trees – tree classifiers, like decision trees with J48 a very common one
17.1. WRITING A NEW CLASSIFIER 237 17.1.4 Implementation
In the following you will find information on what methods need to be imple- mented and other coding guidelines for methods, option handling and documen- tation of the source code.
17.1.4.1 Methods
This section explains what methods need to be implemented in general and more specialized ones in case of meta-classifiers (either with single or multiple base-classifiers).
General
Here is an overview of methods that your new classifier needs to implement in order to integrate nicely into the WEKA framework:
globalInfo()
returns a short description that is displayed in the GUI, like the Explorer or Experimenter. How long this description will be is really up to you, but it should be sufficient to understand the classifier’s underlying algorithm. If the classifier implements the weka.core.TechnicalInformationHandler interface then you could refer to the publication(s) by extending the returned string by getTechnicalInformation().toString().
listOptions()
returns a java.util.Enumeration of weka.core.Option objects. This enu- meration is used to display the help on the command-line, hence it needs to return the Option objects of the superclass as well.
setOptions(String[])
parses the options that the classifier would receive from a command-line invoca- tion. A parameter and argument are always two elements in the string array. A common mistake is to use a single cell in the string array for both of them, e.g., "-S 1" instead of "-S","1". You can use the methods getOption and getFlag of the weka.core.Utils class to retrieve the values of an option or to ascertain whether a flag is present. But note that these calls remove the option and, if applicable, the argument from the string array (“destructive”). The last call in the setOptions methods should always be the super.setOptions(String[]) one, in order to pass on any other arguments still present in the array to the superclass. The following code snippet just parses the only option “alpha” that an imaginary classifier defines:
import weka.core.Utils;
...
public void setOptions(String[] options) throws Exception {
String tmpStr = Utils.getOption("alpha", options); if (tmpStr.length() == 0) {
setAlpha(0.75);
}
else { setAlpha(Double.parseDouble(tmpStr));
}
super.setOptions(options); }
238 CHAPTER 17. EXTENDING WEKA
getOptions()
returns a string array of command-line options that resemble the current clas- sifier setup. Supplying this array to the setOptions(String[]) method must result in the same configuration. This method will get called in the GUI when copying a classifier setup to the clipboard. Since handling of arrays is a bit cum- bersome in Java (due to fixed length), using an instance of java.util.Vector is a lot easier for creating the array that needs to be returned. The following code snippet just adds the only option “alpha” that the classifier defines to the array that is being returned, including the options of the superclass:
import java.util.Arrays; import java.util.Vector;
...
public String[] getOptions() {
Vector
result.add(“” + getAlpha()); result.addAll(Arrays.asList(super.getOptions())); // superclass return result.toArray(new String[result.size()]);
}
Note, that the getOptions() method requires you to add the preceding dash for
an option, opposed to the getOption/getFlag calls in the setOptions method.
getCapabilities()
returns meta-information on what type of data the classifier can handle, in regards to attributes and class attributes. See section “Capabilities” on page 236 for more information.
buildClassifier(Instances)
builds the model from scratch with the provided dataset. Each subsequent call of this method must result in the same model being built. The buildClassifier method also tests whether the supplied data can be handled at all by the clas- sifier, utilizing the capabilities returned by the getCapabilities() method:
public void buildClassifier(Instances data) throws Exception { // test data against capabilities getCapabilities().testWithFail(data);
// remove instances with missing class value,
// but don’t modify original data data = new Instances(data); data.deleteWithMissingClass(); // actual model generation
… }
toString()
is used for outputting the built model. This is not required, but it is useful for the user to see properties of the model. Decision trees normally ouput the tree, support vector machines the support vectors and rule-based classifiers the generated rules.
17.1. WRITING A NEW CLASSIFIER 239
distributionForInstance(Instance)
returns the class probabilities array of the prediction for the given weka.core.Instance object. If your classifier handles nominal class attributes, then you need to over-
ride this method.
classifyInstance(Instance)
returnstheclassificationorregressionforthegivenweka.core.Instanceobject.
In case of a nominal class attribute, this method returns the index of the class
label that got predicted. You do not need to override this method in this case as
the weka.classifiers.Classifier superclass already determines the class la-
bel index based on the probabilities array that the distributionForInstance(Instance) method returns (it returns the index in the array with the highest probability;
in case of ties the first one). For numeric class attributes, you need to override this method, as it has to return the regression value predicted by the model.
main(String[])
executes the classifier from command-line. If your new algorithm is called FunkyClassifier, then use the following code as your main method:
/**
* Main method for executing this classifier.
*
* @param args the options, use “-h” to display options */
public static void main(String[] args) { runClassifier(new FunkyClassifier(), args);
}
Note: the static method runClassifier (defined in the abstract superclass weka.classifiers.Classifier) handles all the appropriate calls and catches and processes any exceptions as well.
240 CHAPTER 17. EXTENDING WEKA Meta-classifiers
Meta-classifiers define a range of other methods that you might want to override. Normally, this should not be the case. But if your classifier requires the base- classifier(s) to be of a certain type, you can override the specific set-method and add additional checks.
SingleClassifierEnhancer
The following methods are used for handling the single base-classifier of this meta-classifier.
defaultClassifierString()
returns the class name of the classifier that is used as the default one for this meta-classifier.
setClassifier(Classifier)
sets the classifier object. Override this method if you require further checks, like that the classifiers needs to be of a certain class. This is necessary, if you still want to allow the user to parametrize the base-classifier, but not choose another classifier with the GenericObjectEditor. Be aware that this method does not create a copy of the provided classifier.
getClassifier()
returns the currently set classifier object. Note, this method returns the internal object and not a copy.
MultipleClassifiersCombiner
This meta-classifier handles its multiple base-classifiers with the following meth- ods:
setClassifiers(Classifier[])
sets the array of classifiers to use as base-classifiers. If you require the base- classifiers to implement a certain interface or be of a certain class, then override this method and add the necessary checks. Note, this method does not create a copy of the array, but just uses this reference internally.
getClassifiers()
returns the array of classifiers that is in use. Careful, this method returns the internal array and not a copy of it.
getClassifier(int)
returns the classifier from the internal classifier array specified by the given index. Once again, this method does not return a copy of the classifier, but the actual object used by this classifier.
17.1. WRITING A NEW CLASSIFIER 241 17.1.4.2 Guidelines
WEKA’s code base requires you to follow a few rules. The following sections can be used as guidelines in writing your code.
Parameters
There are two different ways of setting/obtaining parameters of an algorithm. Both of them are unfortunately completely independent, which makes option handling so prone to errors. Here are the two:
1. command-line options, using the setOptions/getOptions methods 2. using the properties through the GenericObjectEditor in the GUI
Each command-line option must have a corresponding GUI property and vice versa. In case of GUI properties, the get- and set-method for a property must comply with Java Beans style in order to show up in the GUI. You need to supply three methods for each property:
• public void set
• public
• public String
With a property called “alpha” of type “double”, we get the following method signatures:
• public void setAlpha(double) • public double getAlpha()
• public String alphaTipText()
These get- and set-methods should be used in the getOptions and setOptions methods as well, to impose the same checks when getting/setting parameters.
Randomization
In order to get repeatable experiments, one is not allowed to use unseeded random number generators like Math.random(). Instead, one has to instantiate a java.util.Random object in the buildClassifier(Instances) method with a specific seed value. The seed value can be user supplied, of course, which all the Randomizable… abstract classifiers already implement.
242 CHAPTER 17. EXTENDING WEKA
Capabilities
By default, the weka.classifiers.Classifier superclass returns an object that denotes that the classifier can handle any type of data. This is useful for rapid prototyping of new algorithms, but also very dangerous. If you do not specifically define what type of data can be handled by your classifier, you can end up with meaningless models or errors. This can happen if you devise a new classifier which is supposed to handle only numeric attributes. By using the value(int/Attribute) method of a weka.core.Instance to obtain the numeric value of an attribute, you also obtain the internal format of nominal, string and relational attributes. Of course, treating these attribute types as numeric ones does not make any sense. Hence it is highly recommended (and required for contributions) to override this method in your own classifier.
There are three different types of capabilities that you can define:
1. attribute related – e.g., nominal, numeric, date, missing values, …
2. class attribute related – e.g., no-class, nominal, numeric, missing class
values, …
3. miscellaneous – e.g., only multi-instance data, minimum number of in-
stances in the training data
There are some special cases:
• incremental classifiers – need to set the minimum number of instances in the training data to 0, since the default is 1: setMinimumNumberInstances(0)
• multi-instance classifiers – in order to signal that the special multi-instance format (bag-id, bag-data, class) is used, they need to enable the following capability:
enable(Capability.ONLY MULTIINSTANCE)
These classifiers also need to implement the interface specific to multi- instance, weka.core.MultiInstanceCapabilitiesHandler, which returns the capabilities for the bag-data.
• cluster algorithms – since clusterers are unsupervised algorithms, they cannot process data with the class attribute set. The capability that denotes that an algorithm can handle data without a class attribute is Capability.NO CLASS
And a note on enabling/disabling nominal attributes or nominal class attributes. These operations automatically enable/disable the binary, unary and empty nominal capabilities as well. The following sections list a few examples of how to configure the capabilities.
17.1. WRITING A NEW CLASSIFIER 243
Simple classifier
A classifier that handles only numeric classes and numeric and nominal at- tributes, but no missing values at all, would configure the Capabilities object like this:
public Capabilities getCapabilities() { Capabilities result = new Capabilities(this); // attributes result.enable(Capability.NOMINAL_ATTRIBUTES); result.enable(Capability.NUMERIC_ATTRIBUTES); // class result.enable(Capability.NUMERIC_CLASS); return result;
}
Another classifier, that only handles binary classes and only nominal attributes and missing values, would implement the getCapabilities() method as fol- lows:
public Capabilities getCapabilities() { Capabilities result = new Capabilities(this); // attributes result.enable(Capability.NOMINAL_ATTRIBUTES); result.enable(Capability.MISSING_VALUES);
// class result.enable(Capability.BINARY_CLASS); result.disable(Capability.UNNARY_CLASS); result.enable(Capability.MISSING_CLASS_VALUES); return result;
}
Meta-classifier
Meta-classifiers, by default, just return the capabilities of their base classifiers – in case of descendants of the weka.classifier.MultipleClassifiersCombiner, an AND over all the Capabilities of the base classifiers is returned.
Due to this behavior, the capabilities depend – normally – only on the cur- rently configured base classifier(s). To soften filtering for certain behavior, meta- classifiers also define so-called Dependencies on a per-Capability basis. These dependencies tell the filter that even though a certain capability is not sup- ported right now, it is possible that it will be supported with a different base classifier. By default, all capabilities are initialized as Dependencies.
weka.classifiers.meta.LogitBoost, e.g., is restricted to nominal classes. For that reason it disables the Dependencies for the class:
result.disableAllClasses(); // disable all class types result.disableAllClassDependencies(); // no dependencies! result.enable(Capability.NOMINAL_CLASS); // only nominal classes allowed
244 CHAPTER 17. EXTENDING WEKA
Javadoc
In order to keep code-quality high and maintenance low, source code needs to be well documented. This includes the following Javadoc requirements:
• class
– description of the classifier
– listing of command-line parameters – publication(s), if applicable
– @author and @version tag
• methods (all, not just public)
– each parameter is documented
– return value, if applicable, is documented
– exception(s) are documented
– the setOptions(String[]) method also lists the command-line pa-
rameters
Most of the class-related and the setOptions Javadoc is already available through the source code:
• description of the classifier – globalInfo()
• listing of command-line parameters – listOptions()
• publication(s), if applicable – getTechnicalInformation()
In order to avoid manual syncing between Javadoc and source code, WEKA comes with some tools for updating the Javadoc automatically. The following tools take a concrete class and update its source code (the source code directory needs to be supplied as well, of course):
• weka.core.AllJavadoc – executes all Javadoc-producing classes (this is the tool, you would normally use)
• weka.core.GlobalInfoJavadoc – updates the globalinfo tags
• weka.core.OptionHandlerJavadoc – updates the option tags
• weka.core.TechnicalInformationHandlerJavadoc – updates the tech-
nical tags (plain text and BibTeX)
These tools look for specific comment tags in the source code and replace every- thing in between the start and end tag with the documentation obtained from the actual class.
• description of the classifier
will be automatically replaced
• listing of command-line parameters
will be automatically replaced
• publication(s), if applicable
will be automatically replaced
for a shortened, plain-text version use the following:
will be automatically replaced
17.1. WRITING A NEW CLASSIFIER 245 Here is a template of a Javadoc class block for an imaginary classifier that also
implements the weka.core.TechnicalInformationHandler interface:
/**
*
*
*
* @author John Doe (john dot doe at no dot where dot com)
* @version $Revision: 6192 $ */
The template for any classifier’s setOptions(String[]) method is as follows:
/**
* Parses a given list of options. *
*
* @param options the list of options as an array of strings
* @throws Exception if an option is not supported */
Running the weka.core.AllJavadoc tool over this code will output code with the comments filled out accordingly.
Revisions
Classifiers implement the weka.core.RevisionHandler interface. This pro- vides the functionality of obtaining the Subversion revision from within Java. Classifiers that are not part of the official WEKA distribution do not have to implement the method getRevision() as the weka.classifiers.Classifier class already implements this method. Contributions, on the other hand, need to implement it as follows, in order to obtain the revision of this particular source file:
/**
* Returns the revision string. *
* @return the revision */
public String getRevision() {
return RevisionUtils.extract(“$Revision: 6192 $”);
}
Note, a commit into Subversion will replace the revision number above with the actual revision number.
246 CHAPTER 17. EXTENDING WEKA
Testing
WEKA provides already a test framework to ensure correct basic functionality of a classifier. It is essential for the classifier to pass these tests.
Option handling
You can check the option handling of your classifier with the following tool from command-line:
weka.core.CheckOptionHandler -W classname [– additional parameters]
All tests need to return yes.
GenericObjectEditor
The CheckGOE class checks whether all the properties available in the GUI have a tooltip accompanying them and whether the globalInfo() method is declared:
weka.core.CheckGOE -W classname [– additional parameters]
All tests, once again, need to return yes.
Source code
Classifiers that implement the weka.classifiers.Sourcable interface can out- put Java code of the built model. In order to check the generated code, one should not only compile the code, but also test it with the following test class:
weka.classifiers.CheckSource
This class takes the original WEKA classifier, the generated code and the dataset used for generating the model (and an optional class index) as parameters. It builds the WEKA classifier on the dataset and compares the output, the one from the WEKA classifier and the one from the generated source code, whether they are the same.
Here is an example call for weka.filters.trees.J48 and the generated class weka.filters.WEKAWrapper (it wraps the actual generated code in a pseudo-classifier):
java weka.classifiers.CheckSource \ -W weka.classifiers.trees.J48 \ -S weka.classifiers.WEKAWrapper \ -t data.arff
It needs to return Tests OK!.
Unit tests
In order to make sure that your classifier applies to the WEKA criteria, you
should add your classifier to the junit unit test framework, i.e., by creating a Test
class. The superclass for classifier unit tests is weka.classifiers.AbstractClassifierTest.
17.2. WRITING A NEW FILTER 247 17.2 Writing a new Filter
The “work horses” of preprocessing in WEKA are filters. They perform many tasks, from resampling data, to deleting and standardizing attributes. In the following are two different approaches covered that explain in detail how to implement a new filter:
• default – this is how filters had to be implemented in the past.
• simple – since there are mainly two types of filters, batch or stream, ad- ditional abstract classes were introduced to speed up the implementation
process.
17.2.1 Default approach
The default approach is the most flexible, but also the most complicated one for writing a new filter. This approach has to be used, if the filter cannot be written using the simple approach described further below.
17.2.1.1 Implementation
The following methods are of importance for the implementation of a filter and explained in detail further down. It is also a good idea studying the Javadoc of these methods as declared in the weka.filters.Filter class:
• getCapabilities()
• setInputFormat(Instances) • getInputFormat()
• setOutputFormat(Instances) • getOutputFormat()
• input(Instance)
• bufferInput(Instance)
• push(Instance)
• output()
• batchFinished()
• flushInput()
• getRevision()
But only the following ones normally need to be modified:
• getCapabilities()
• setInputFormat(Instances) • input(Instance)
• batchFinished()
• getRevision()
For more information on “Capabilities” see section 17.2.3. Please note, that the weka.filters.Filter superclass does not implement the weka.core.OptionHandler interface. See section “Option handling” on page 243.
248 CHAPTER 17. EXTENDING WEKA
setInputFormat(Instances)
With this call, the user tells the filter what structure, i.e., attributes, the input data has. This method also tests, whether the filter can actually process this data, according to the capabilities specified in the getCapabilities() method.
If the output format of the filter, i.e., the new Instances header, can be determined based alone on this information, then the method should set the output format via setOutputFormat(Instances) and return true, otherwise it has to return false.
getInputFormat()
This method returns an Instances object containing all currently buffered Instance objects from the input queue.
setOutputFormat(Instances)
setOutputFormat(Instances) defines the new Instances header for the out- put data. For filters that work on a row-basis, there should not be any changes between the input and output format. But filters that work on attributes, e.g., removing, adding, modifying, will affect this format. This method must be called with the appropriate Instances object as parameter, since all Instance objects being processed will rely on the output format (they use it as dataset that they belong to).
getOutputFormat()
This method returns the currently set Instances object that defines the output format. In case setOutputFormat(Instances) has not been called yet, this method will return null.
input(Instance)
returns true if the given Instance can be processed straight away and can be collected immediately via the output() method (after adding it to the output queue via push(Instance), of course). This is also the case if the first batch of data has been processed and the Instance belongs to the second batch. Via isFirstBatchDone() one can query whether this Instance is still part of the first batch or of the second.
If the Instance cannot be processed immediately, e.g., the filter needs to collect all the data first before doing some calculations, then it needs to be buffered with bufferInput(Instance) until batchFinished() is called. In this case, the method needs to return false.
bufferInput(Instance)
In case an Instance cannot be processed immediately, one can use this method to buffer them in the input queue. All buffered Instance objects are available via the getInputFormat() method.
push(Instance)
adds the given Instance to the output queue.
output()
Returns the next Instance object from the output queue and removes it from there. In case there is no Instance available this method returns null.
17.2. WRITING A NEW FILTER 249
batchFinished()
signals the end of a dataset being pushed through the filter. In case of a filter
that could not process the data of the first batch immediately, this is the place to determine what the output format will be (and set if via setOutputFormat(Instances)) and finally process the input data. The currently available data can be retrieved
with the getInputFormat() method. After processing the data, one needs to
call flushInput() to remove all the pending input data.
flushInput()
flushInput() removes all buffered Instance objects from the input queue. This method must be called after all the Instance objects have been processed in the batchFinished() method.
Option handling
If the filter should be able to handle command-line options, then the inter- face weka.core.OptionHandler needs to be implemented. In addition to that, the following code should be added at the end of the setOptions(String[]) method:
if (getInputFormat() != null) { setInputFormat(getInputFormat());
}
This will inform the filter about changes in the options and therefore reset it.
250 CHAPTER 17. EXTENDING WEKA 17.2.1.2 Examples
The following examples, covering batch and stream filters, illustrate the filter framework and how to use it.
Unseeded random number generators like Math.random() should never be used since they will produce different results in each run and repeatable exper- iments are essential in machine learning.
BatchFilter
This simple batch filter adds a new attribute called blah at the end of the dataset. The rows of this attribute contain only the row’s index in the data. Since the batch-filter does not have to see all the data before creating the output format, the setInputFormat(Instances) sets the output format and returns true (indicating that the output format can be queried immediately). The batchFinished() method performs the processing of all the data.
import weka.core.*;
import weka.core.Capabilities.*;
public class BatchFilter extends Filter {
public String globalInfo() {
return “A batch filter that adds an additional attribute ’blah’ at the end ”
+ “containing the index of the processed instance. The output format ” + “can be collected immediately.”;
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities(); result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); // filter doesn’t need class to be set return result;
}
public boolean setInputFormat(Instances instanceInfo) throws Exception { super.setInputFormat(instanceInfo);
Instances outFormat = new Instances(instanceInfo, 0); outFormat.insertAttributeAt(new Attribute(“blah”),
outFormat.numAttributes());
setOutputFormat(outFormat);
return true; // output format is immediately available
}
public boolean batchFinished() throws Exception { if (getInputFormat() = null)
throw new NullPointerException(“No input instance format defined”); Instances inst = getInputFormat();
Instances outFormat = getOutputFormat();
for (int i = 0; i < inst.numInstances(); i++) {
double[] newValues = new double[outFormat.numAttributes()]; double[] oldValues = inst.instance(i).toDoubleArray(); System.arraycopy(oldValues, 0, newValues, 0, oldValues.length); newValues[newValues.length - 1] = i;
push(new Instance(1.0, newValues)); }
flushInput();
m_NewBatch = true; m_FirstBatchDone = true;
return (numPendingOutput() != 0);
}
public static void main(String[] args) { runFilter(new BatchFilter(), args);
} }
17.2. WRITING A NEW FILTER 251 BatchFilter2
In contrast to the first batch filter, this one here cannot determine the output format immediately (the number of instances in the first batch is part of the attribute name now). This is done in the batchFinished() method.
import weka.core.*;
import weka.core.Capabilities.*;
public class BatchFilter2 extends Filter {
public String globalInfo() {
return "A batch filter that adds an additional attribute ’blah’ at the end "
+ "containing the index of the processed instance. The output format " + "cannot be collected immediately.";
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities(); result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); // filter doesn’t need class to be set return result;
}
public boolean batchFinished() throws Exception { if (getInputFormat() = null)
throw new NullPointerException("No input instance format defined"); // output format still needs to be set (depends on first batch of data) if (!isFirstBatchDone()) {
Instances outFormat = new Instances(getInputFormat(), 0); outFormat.insertAttributeAt(new Attribute(
"blah-" + getInputFormat().numInstances()), outFormat.numAttributes()); setOutputFormat(outFormat);
}
Instances inst = getInputFormat();
Instances outFormat = getOutputFormat();
for (int i = 0; i < inst.numInstances(); i++) {
double[] newValues = new double[outFormat.numAttributes()]; double[] oldValues = inst.instance(i).toDoubleArray(); System.arraycopy(oldValues, 0, newValues, 0, oldValues.length); newValues[newValues.length - 1] = i;
push(new Instance(1.0, newValues)); }
flushInput();
m_NewBatch = true; m_FirstBatchDone = true;
return (numPendingOutput() != 0);
}
public static void main(String[] args) { runFilter(new BatchFilter2(), args);
} }
252 CHAPTER 17. EXTENDING WEKA BatchFilter3
As soon as this batch filter’s first batch is done, it can process Instance objects immediately in the input(Instance) method. It adds a new attribute which contains just a random number, but the random number generator being used is seeded with the number of instances from the first batch.
import weka.core.*;
import weka.core.Capabilities.*; import java.util.Random;
public class BatchFilter3 extends Filter {
protected int m_Seed; protected Random m_Random;
public String globalInfo() {
return "A batch filter that adds an attribute ’blah’ at the end "
+ "containing a random number. The output format cannot be collected " + "immediately.";
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities(); result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); // filter doesn’t need class to be set return result;
}
public boolean input(Instance instance) throws Exception { if (getInputFormat() = null)
throw new NullPointerException("No input instance format defined"); if (isNewBatch()) {
resetQueue();
m_NewBatch = false; }
if (isFirstBatchDone()) convertInstance(instance);
else bufferInput(instance);
return isFirstBatchDone(); }
public boolean batchFinished() throws Exception { if (getInputFormat() = null)
throw new NullPointerException("No input instance format defined");
// output format still needs to be set (random number generator is seeded // with number of instances of first batch)
if (!isFirstBatchDone()) {
m_Seed = getInputFormat().numInstances();
Instances outFormat = new Instances(getInputFormat(), 0); outFormat.insertAttributeAt(new Attribute(
"blah-" + getInputFormat().numInstances()), outFormat.numAttributes()); setOutputFormat(outFormat);
}
Instances inst = getInputFormat();
for (int i = 0; i < inst.numInstances(); i++) {
convertInstance(inst.instance(i)); }
flushInput();
m_NewBatch = true; m_FirstBatchDone = true;
m_Random = null;
return (numPendingOutput() != 0);
}
protected void convertInstance(Instance instance) { if (m_Random = null)
m_Random = new Random(m_Seed);
double[] newValues = new double[instance.numAttributes() + 1]; double[] oldValues = instance.toDoubleArray(); newValues[newValues.length - 1] = m_Random.nextInt(); System.arraycopy(oldValues, 0, newValues, 0, oldValues.length); push(new Instance(1.0, newValues));
}
public static void main(String[] args) { runFilter(new BatchFilter3(), args);
} }
17.2. WRITING A NEW FILTER 253 StreamFilter
This stream filter adds a random number (the seed value is hard-coded) at the end of each Instance of the input data. Since this does not rely on having access to the full data of the first batch, the output format is accessible immediately after using setInputFormat(Instances). All the Instance objects are imme- diately processed in input(Instance) via the convertInstance(Instance) method, which pushes them immediately to the output queue.
import weka.core.*;
import weka.core.Capabilities.*; import java.util.Random;
public class StreamFilter extends Filter {
protected Random m_Random;
public String globalInfo() {
return "A stream filter that adds an attribute ’blah’ at the end "
+ "containing a random number. The output format can be collected " + "immediately.";
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities(); result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); // filter doesn’t need class to be set return result;
}
public boolean setInputFormat(Instances instanceInfo) throws Exception { super.setInputFormat(instanceInfo);
Instances outFormat = new Instances(instanceInfo, 0); outFormat.insertAttributeAt(new Attribute("blah"),
outFormat.numAttributes());
setOutputFormat(outFormat);
m_Random = new Random(1);
return true; // output format is immediately available
}
public boolean input(Instance instance) throws Exception { if (getInputFormat() = null)
throw new NullPointerException("No input instance format defined"); if (isNewBatch()) {
resetQueue();
m_NewBatch = false; }
convertInstance(instance);
return true; // can be immediately collected via output() }
protected void convertInstance(Instance instance) {
double[] newValues = new double[instance.numAttributes() + 1]; double[] oldValues = instance.toDoubleArray(); newValues[newValues.length - 1] = m_Random.nextInt(); System.arraycopy(oldValues, 0, newValues, 0, oldValues.length); push(new Instance(1.0, newValues));
}
public static void main(String[] args) { runFilter(new StreamFilter(), args);
} }
254 CHAPTER 17. EXTENDING WEKA
17.2.2 Simple approach
The base filters and interfaces are all located in the following package:
weka.filters
One can basically divide filters roughly into two different kinds of filters:
• batch filters – they need to see the whole dataset before they can start processing it, which they do in one go
• stream filters – they can start producing output right away and the data just passes through while being modified
You can subclass one of the following abstract filters, depending on the kind of classifier you want to implement:
• weka.filters.SimpleBatchFilter • weka.filters.SimpleStreamFilter
These filters simplify the rather general and complex framework introduced by the abstract superclass weka.filters.Filter. One only needs to implement a couple of abstract methods that will process the actual data and override, if necessary, a few existing methods for option handling.
17.2.2.1 SimpleBatchFilter
Only the following abstract methods need to be implemented:
• globalInfo() – returns a short description of what the filter does; will be displayed in the GUI
• determineOutputFormat(Instances) – generates the new format, based on the input data
• process(Instances) – processes the whole dataset in one go
• getRevision() – returns the Subversion revision information, see section
“Revisions” on page 252
If more options are necessary, then the following methods need to be overridden:
• listOptions() – returns an enumeration of the available options; these are printed if one calls the filter with the -h option
• setOptions(String[]) – parses the given option array, that were passed from command-line
• getOptions() – returns an array of options, resembling the current setup of the filter
See section “Methods” on page 231 and section “Parameters” on page 235 for more information.
17.2. WRITING A NEW FILTER 255 In the following an example implementation that adds an additional at-
tribute at the end, containing the index of the processed instance:
import weka.core.*;
import weka.core.Capabilities.*; import weka.filters.*;
public class SimpleBatch extends SimpleBatchFilter {
public String globalInfo() {
return "A simple batch filter that adds an additional attribute ’blah’ at the end "
+ "containing the index of the processed instance.";
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities();
result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); //// filter doesn’t need class to be set// return result;
}
protected Instances determineOutputFormat(Instances inputFormat) { Instances result = new Instances(inputFormat, 0); result.insertAttributeAt(new Attribute("blah"), result.numAttributes()); return result;
}
protected Instances process(Instances inst) {
Instances result = new Instances(determineOutputFormat(inst), 0); for (int i = 0; i < inst.numInstances(); i++) {
double[] values = new double[result.numAttributes()]; for (int n = 0; n < inst.numAttributes(); n++)
values[n] = inst.instance(i).value(n); values[values.length - 1] = i; result.add(new Instance(1, values));
}
return result;
}
public static void main(String[] args) { runFilter(new SimpleBatch(), args);
} }
256 CHAPTER 17. EXTENDING WEKA
17.2.2.2 SimpleStreamFilter
Only the following abstract methods need to be implemented for a stream filter:
• globalInfo() – returns a short description of what the filter does; will be displayed in the GUI
• determineOutputFormat(Instances) – generates the new format, based on the input data
• process(Instance) – processes a single instance and turns it from the old format into the new one
• getRevision() – returns the Subversion revision information, see section “Revisions” on page 252
If more options are necessary, then the following methods need to be overridden:
• listOptions() – returns an enumeration of the available options; these are printed if one calls the filter with the -h option
• setOptions(String[]) – parses the given option array, that were passed from command-line
• getOptions() – returns an array of options, resembling the current setup of the filter
See also section 17.1.4.1, covering “Methods” for classifiers.
17.2. WRITING A NEW FILTER 257
In the following an example implementation of a stream filter that adds an extra attribute at the end, which is filled with random numbers. The reset() method is only used in this example, since the random number generator needs to be re-initialized in order to obtain repeatable results.
import weka.core.*;
import weka.core.Capabilities.*; import weka.filters.*;
import java.util.Random;
public class SimpleStream extends SimpleStreamFilter {
protected Random m_Random;
public String globalInfo() {
return "A simple stream filter that adds an attribute ’blah’ at the end "
+ "containing a random number.";
}
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities();
result.enableAllAttributes();
result.enableAllClasses();
result.enable(Capability.NO_CLASS); //// filter doesn’t need class to be set// return result;
}
protected void reset() { super.reset();
m_Random = new Random(1);
}
protected Instances determineOutputFormat(Instances inputFormat) { Instances result = new Instances(inputFormat, 0); result.insertAttributeAt(new Attribute("blah"), result.numAttributes()); return result;
}
protected Instance process(Instance inst) {
double[] values = new double[inst.numAttributes() + 1]; for (int n = 0; n < inst.numAttributes(); n++)
values[n] = inst.value(n); values[values.length - 1] = m_Random.nextInt(); Instance result = new Instance(1, values); return result;
}
public static void main(String[] args) { runFilter(new SimpleStream(), args);
} }
A real-world implementation of a stream filter is the MultiFilter class (pack- age weka.filters), which passes the data through all the filters it contains. Depending on whether all the used filters are streamable or not, it acts either as stream filter or as batch filter.
258 CHAPTER 17. EXTENDING WEKA
17.2.2.3 Internals
Some useful methods of the filter classes:
• isNewBatch() – returns true if an instance of the filter was just instan- tiated or a new batch was started via the batchFinished() method.
• isFirstBatchDone() – returns true as soon as the first batch was finished via the batchFinished() method. Useful for supervised filters, which should not be altered after being trained with the first batch of instances.
17.2.3 Capabilities
Filters implement the weka.core.CapabilitiesHandler interface like the clas- sifiers. This method returns what kind of data the filter is able to process. Needs to be adapted for each individual filter, since the default implementation allows the processing of all kinds of attributes and classes. Otherwise correct function- ing of the filter cannot be guaranteed. See section “Capabilities” on page 236 for more information.
17.2.4 Packages
A few comments about the different filter sub-packages:
• supervised – contains supervised filters, i.e., filters that take class distri- butions into account. Must implement the weka.filters.SupervisedFilter interface.
– attribute – filters that work column-wise. – instance – filters that work row-wise.
• unsupervised – contains unsupervised filters, i.e., they work without taking any class distributions into account. The filter must implement the weka.filters.UnsupervisedFilter interface.
– attribute – filters that work column-wise. – instance – filters that work row-wise.
Javadoc
The Javadoc generation works the same as with classifiers. See section “Javadoc” on page 238 for more information.
17.2.5 Revisions
Filters, like classifiers, implement the weka.core.RevisionHandler interface. This provides the functionality of obtaining the Subversion revision from within Java. Filters that are not part of the official WEKA distribution do not have to implement the method getRevision() as the weka.filters.Filter class already implements this method. Contributions, on the other hand, need to implement it, in order to obtain the revision of this particular source file. See section “Revisions” on page 239.
17.2. WRITING A NEW FILTER 259 17.2.6 Testing
WEKA provides already a test framework to ensure correct basic functionality of a filter. It is essential for the filter to pass these tests.
17.2.6.1 Option handling
You can check the option handling of your filter with the following tool from command-line:
weka.core.CheckOptionHandler -W classname [-- additional parameters]
All tests need to return yes. 17.2.6.2 GenericObjectEditor
The CheckGOE class checks whether all the properties available in the GUI have a tooltip accompanying them and whether the globalInfo() method is declared:
weka.core.CheckGOE -W classname [-- additional parameters]
All tests, once again, need to return yes. 17.2.6.3 Source code
Filters that implement the weka.filters.Sourcable interface can output Java code of their internal representation. In order to check the generated code, one should not only compile the code, but also test it with the following test class:
weka.filters.CheckSource
This class takes the original WEKA filter, the generated code and the dataset used for generating the source code (and an optional class index) as parameters. It builds the WEKA filter on the dataset and compares the output, the one from the WEKA filter and the one from the generated source code, whether they are the same.
Here is an example call for weka.filters.unsupervised.attribute.ReplaceMissingValues and the generated class weka.filters.WEKAWrapper (it wraps the actual gen-
erated code in a pseudo-filter):
java weka.filters.CheckSource \
-W weka.filters.unsupervised.attribute.ReplaceMissingValues \ -S weka.filters.WEKAWrapper \
-t data.arff
It needs to return Tests OK!. 17.2.6.4 Unit tests
In order to make sure that your filter applies to the WEKA criteria, you should add your filter to the junit unit test framework, i.e., by creating a Test class. The superclass for filter unit tests is weka.filters.AbstractFilterTest.
260 CHAPTER 17. EXTENDING WEKA 17.3 Writing other algorithms
The previous sections covered how to implement classifiers and filters. In the following you will find some information on how to implement clusterers, as- sociators and attribute selection algorithms. The various algorithms are only covered briefly, since other important components (capabilities, option handling, revisions) have already been discussed in the other chapters.
17.3.1 Clusterers
Superclasses and interfaces
All clusterers implement the interface weka.clusterers.Clusterer, but most algorithms will be most likely derived (directly or further up in the class hier- archy) from the abstract superclass weka.clusterers.AbstractClusterer.
weka.clusterers.SingleClustererEnhancer is used for meta-clusterers, like the FilteredClusterer that filters the data on-the-fly for the base-clusterer.
Here are some common interfaces that can be implemented:
• weka.clusterers.DensityBasedClusterer – for clusterers that can esti- mate the density for a given instance. AbstractDensityBasedClusterer already implements this interface.
• weka.clusterers.UpdateableClusterer – clusterers that can generate their model incrementally implement this interface, like CobWeb.
• NumberOfClustersRequestable – is for clusterers that allow to specify the number of clusters to generate, like SimpleKMeans.
• weka.core.Randomizable – for clusterers that support randomization in
one way or another. RandomizableClusterer, RandomizableDensityBasedClusterer and RandomizableSingleClustererEnhancer all implement this inter-
face already.
Methods
In the following a short description of methods that are common to all cluster algorithms, see also the Javadoc for the Clusterer interface.
buildClusterer(Instances)
Like the buildClassifier(Instances) method, this method completely re- builds the model. Subsequent calls of this method with the same dataset must result in exactly the same model being built. This method also tests the training data against the capabilities of this this clusterer:
public void buildClusterer(Instances data) throws Exception { // test data against capabilities getCapabilities().testWithFail(data);
// actual model generation
... }
clusterInstance(Instance)
returns the index of the cluster the provided Instance belongs to.
17.3. WRITING OTHER ALGORITHMS 261
distributionForInstance(Instance)
returns the cluster membership for this Instance object. The membership is a double array containing the probabilities for each cluster.
numberOfClusters()
returns the number of clusters that the model contains, after the model has been generated with the buildClusterer(Instances) method.
getCapabilities()
see section “Capabilities” on page 236 for more information.
toString()
should output some information on the generated model. Even though this is not required, it is rather useful for the user to get some feedback on the built model.
main(String[])
executes the clusterer from command-line. If your new algorithm is called FunkyClusterer, then use the following code as your main method:
/**
* Main method for executing this clusterer.
*
* @param args the options, use "-h" to display options */
public static void main(String[] args) { AbstractClusterer.runClusterer(new FunkyClusterer(), args);
}
Testing
For some basic tests from the command-line, you can use the following test class:
weka.clusterers.CheckClusterer -W classname [further options]
For junit tests, you can subclass the weka.clusterers.AbstractClustererTest class and add additional tests.
262 CHAPTER 17. EXTENDING WEKA
17.3.2 Attribute selection
Attribute selection consists basically of two different types of classes:
• evaluator – determines the merit of single attributes or subsets of at- tributes
• search algorithm – the search heuristic
Each of the them will be discussed separately in the following sections.
Evaluator
The evaluator algorithm is responsible for determining merit of the current attribute selection.
Superclasses and interfaces
The ancestor for all evaluators is the weka.attributeSelection.ASEvaluation class.
Here are some interfaces that are commonly implemented by evaluators:
• AttributeEvaluator – evaluates only single attributes
• SubsetEvaluator – evaluates subsets of attributes
• AttributeTransformer – evaluators that transform the input data
Methods
In the following a brief description of the main methods of an evaluator.
buildEvaluator(Instances)
Generates the attribute evaluator. Subsequent calls of this method with the same data (and the same search algorithm) must result in the same attributes being selected. This method also checks the data against the capabilities:
public void buildEvaluator (Instances data) throws Exception { // can evaluator handle data? getCapabilities().testWithFail(data);
// actual initialization of evaluator
... }
postProcess(int[])
can be used for optional post-processing of the selected attributes, e.g., for ranking purposes.
17.3. WRITING OTHER ALGORITHMS 263
main(String[])
executes the evaluator from command-line. If your new algorithm is called FunkyEvaluator, then use the following code as your main method:
/**
* Main method for executing this evaluator.
*
* @param args the options, use "-h" to display options */
public static void main(String[] args) { ASEvaluation.runEvaluator(new FunkyEvaluator(), args);
}
Search
The search algorithm defines the heuristic of searching, e.g, exhaustive search, greedy or genetic.
Superclasses and interfaces
The ancestor for all search algorithms is the weka.attributeSelection.ASSearch class.
Interfaces that can be implemented, if applicable, by a search algorithm:
• RankedOutputSearch – for search algorithms that produce ranked lists of attributes
• StartSetHandler – search algorithms that can make use of a start set of attributes implement this interface
Methods
Search algorithms are rather basic classes in regards to methods that need to be implemented. Only the following method needs to be implemented:
search(ASEvaluation,Instances)
uses the provided evaluator to guide the search.
Testing
For some basic tests from the command-line, you can use the following test class:
weka.attributeSelection.CheckAttributeSelection -eval classname -search classname [further options]
For junit tests, you can subclass the weka.attributeSelection.AbstractEvaluatorTest or weka.attributeSelection.AbstractSearchTest class and add additional
tests.
264 CHAPTER 17. EXTENDING WEKA
17.3.3 Associators
Superclasses and interfaces
The interface weka.associations.Associator is common to all associator al- gorithms. But most algorithms will be derived from AbstractAssociator, an abstract class implementing this interface. As with classifiers and clusterers, you can also implement a meta-associator, derived from SingleAssociatorEnhancer. An example for this is the FilteredAssociator, which filters the training data on-the-fly for the base-associator.
The only other interface that is used by some other association algorithms, is the weka.clusterers.CARuleMiner one. Associators that learn class associ- ation rules implement this interface, like Apriori.
Methods
The associators are very basic algorithms and only support building of the model.
buildAssociations(Instances)
Like the buildClassifier(Instances) method, this method completely re- builds the model. Subsequent calls of this method with the same dataset must result in exactly the same model being built. This method also tests the training data against the capabilities:
public void buildAssociations(Instances data) throws Exception { // other necessary setups
...
// test data against capabilities getCapabilities().testWithFail(data);
// actual model generation
... }
getCapabilities()
see section “Capabilities” on page 236 for more information.
toString()
should output some information on the generated model. Even though this is not required, it is rather useful for the user to get some feedback on the built model.
17.3. WRITING OTHER ALGORITHMS 265
main(String[])
executes the associator from command-line. If your new algorithm is called FunkyAssociator, then use the following code as your main method:
/**
* Main method for executing this associator.
*
* @param args the options, use "-h" to display options */
public static void main(String[] args) { AbstractAssociator.runAssociator(new FunkyAssociator(), args);
}
Testing
For some basic tests from the command-line, you can use the following test class:
weka.associations.CheckAssociator -W classname [further options]
For junit tests, you can subclass the weka.associations.AbstractAssociatorTest class and add additional tests.
266 CHAPTER 17. EXTENDING WEKA 17.4 Extending the Explorer
The plugin architecture of the Explorer allows you to add new functionality easily without having to dig into the code of the Explorer itself. In the following you will find information on how to add new tabs, like the “Classify” tab, and new visualization plugins for the “Classify” tab.
17.4.1 Adding tabs
The Explorer is a handy tool for initial exploration of your data – for proper statistical evaluation, the Experimenter should be used instead. But if the available functionality is not enough, you can always add your own custom- made tabs to the Explorer.
17.4.1.1 Requirements
Here is roughly what is required in order to add a new tab (the examples below go into more detail):
• your class must be derived from javax.swing.JPanel
• theinterfaceweka.gui.explorer.Explorer.ExplorerPanelmustbeim-
plemented by your class
• optional interfaces
– weka.gui.explorer.Explorer.LogHandler – in case you want to take advantage of the logging in the Explorer
– weka.gui.explorer.Explorer.CapabilitiesFilterChangeListener – in case your class needs to be notified of changes in the Capabilities, e.g., if new data is loaded into the Explorer
• addingtheclassnameofyourclasstotheTabspropertyintheExplorer.props file
17.4.1.2 Examples
The following examples demonstrate the plugin architecture. Only the neces- sary details are discussed, as the full source code is available from the WEKA Examples [3] (package wekaexamples.gui.explorer).
SQL worksheet
Purpose
Displaying the SqlViewer as a tab in the Explorer instead of using it either via the Open DB... button or as standalone application. Uses the existing components already available in WEKA and just assembles them in a JPanel. Since this tab does not rely on a dataset being loaded into the Explorer, it will be used as a standalone one.
17.4. EXTENDING THE EXPLORER 267
Useful for people who are working a lot with databases and would like to have an SQL worksheet available all the time instead of clicking on a button every time to open up a database dialog.
Implementation
• class is derived from javax.swing.JPanel and implements the interface weka.gui.Explorer.ExplorerPanel (the full source code also imports the weka.gui.Explorer.LogHandler interface, but that is only additional functionality):
public class SqlPanel extends JPanel
implements ExplorerPanel {
• some basic members that we need to have /** the parent frame */
protected Explorer m_Explorer = null;
/** sends notifications when the set of working instances gets changed*/ protected PropertyChangeSupport m_Support = new PropertyChangeSupport(this);
• methods we need to implement due to the used interfaces /** Sets the Explorer to use as parent frame */
public void setExplorer(Explorer parent) m_Explorer = parent;
}
/** returns the parent Explorer frame */ public Explorer getExplorer() {
return m_Explorer; }
/** Returns the title for the tab in the public String getTabTitle() {
{
Explorer */
return "SQL"; // what’s displayed as tab-title, e.g., Classify }
/** Returns the tooltip for the tab in the Explorer */ public String getTabTitleToolTip() {
return "Retrieving data from databases"; // the tooltip of the tab }
/** ignored, since we "generate" data and not receive it */ public void setInstances(Instances inst) {
}
/** PropertyChangeListener which will be notified of value changes. */ public void addPropertyChangeListener(PropertyChangeListener l) {
m_Support.addPropertyChangeListener(l); }
/** Removes a PropertyChangeListener. */
public void removePropertyChangeListener(PropertyChangeListener l) {
m_Support.removePropertyChangeListener(l); }
268 CHAPTER 17. • additional GUI elements
/** the actual SQL worksheet */ protected SqlViewer m_Viewer;
/** the panel for the buttons */ protected JPanel m_PanelButtons;
EXTENDING WEKA
/** the Load button - makes the data available in the Explorer */ protected JButton m_ButtonLoad = new JButton("Load data");
/** displays the current query */
protected JLabel m_LabelQuery = new JLabel("");
• loading the data into the Explorer by clicking on the Load button will fire a propertyChange event:
m_ButtonLoad.addActionListener(new ActionListener() { public void actionPerformed(ActionEvent evt){
m_Support.firePropertyChange("", null, null); }
});
• the propertyChange event will perform the actual loading of the data, hence we add an anonymous property change listener to our panel:
addPropertyChangeListener(new PropertyChangeListener() { public void propertyChange(PropertyChangeEvent e) {
try {
// load data
InstanceQuery query = new InstanceQuery(); query.setDatabaseURL(m_Viewer.getURL()); query.setUsername(m_Viewer.getUser()); query.setPassword(m_Viewer.getPassword());
Instances data = query.retrieveInstances(m_Viewer.getQuery());
// set data in preprocess panel (also notifies of capabilties changes)
getExplorer().getPreprocessPanel().setInstances(data); }
catch (Exception ex) { ex.printStackTrace();
} }
});
• In order to add our SqlPanel to the list of tabs displayed in the Ex- plorer, we need to modify the Explorer.props file (just extract it from the weka.jar and place it in your home directory). The Tabs property must look like this:
Tabs=weka.gui.explorer.SqlPanel,\ weka.gui.explorer.ClassifierPanel,\ weka.gui.explorer.ClustererPanel,\ weka.gui.explorer.AssociationsPanel,\ weka.gui.explorer.AttributeSelectionPanel,\ weka.gui.explorer.VisualizePanel
17.4. EXTENDING THE EXPLORER 269 Screenshot
270 CHAPTER 17. EXTENDING WEKA
Artificial data generation
Purpose
Instead of only having a Generate... button in the PreprocessPanel or using it from command-line, this example creates a new panel to be displayed as extra tab in the Explorer. This tab will be available regardless whether a dataset is already loaded or not (= standalone).
Implementation
• class is derived from javax.swing.JPanel and implements the interface weka.gui.Explorer.ExplorerPanel (the full source code also imports the weka.gui.Explorer.LogHandler interface, but that is only additional functionality):
public class GeneratorPanel extends JPanel
implements ExplorerPanel {
• some basic members that we need to have (the same as for the SqlPanel class):
/** the parent frame */
protected Explorer m_Explorer = null;
/** sends notifications when the set of working instances gets changed*/ protected PropertyChangeSupport m_Support = new PropertyChangeSupport(this);
• methods we need to implement due to the used interfaces (almost identical to SqlPanel):
/** Sets the Explorer to use as parent frame */
public void setExplorer(Explorer parent) m_Explorer = parent;
}
/** returns the parent Explorer frame */ public Explorer getExplorer() {
return m_Explorer; }
/** Returns the title for the tab in the public String getTabTitle() {
{
Explorer */
return "DataGeneration"; // what’s displayed as tab-title, e.g., Classify }
/** Returns the tooltip for the tab in the Explorer */ public String getTabTitleToolTip() {
return "Generating artificial datasets"; // the tooltip of the tab }
/** ignored, since we "generate" data and not receive it */
public void setInstances(Instances inst) {
}
/** PropertyChangeListener which will be notified of value changes. */ public void addPropertyChangeListener(PropertyChangeListener l) {
m_Support.addPropertyChangeListener(l); }
/** Removes a PropertyChangeListener. */
public void removePropertyChangeListener(PropertyChangeListener l) {
m_Support.removePropertyChangeListener(l); }
17.4. EXTENDING THE EXPLORER 271
• additional GUI elements:
/** the GOE for the generators */
protected GenericObjectEditor m_GeneratorEditor = new GenericObjectEditor();
/** the text area for the output of the generated data */ protected JTextArea m_Output = new JTextArea();
/** the Generate button */
protected JButton m_ButtonGenerate = new JButton("Generate");
/** the Use button */
protected JButton m_ButtonUse = new JButton("Use");
• the Generate button does not load the generated data directly into the Explorer, but only outputs it in the JTextArea (the Use button loads the data - see further down):
m_ButtonGenerate.addActionListener(new ActionListener(){ public void actionPerformed(ActionEvent evt){
DataGenerator generator = (DataGenerator) m_GeneratorEditor.getValue(); String relName = generator.getRelationName();
String cname = generator.getClass().getName().replaceAll(".*\\.", ""); String cmd = generator.getClass().getName();
if (generator instanceof OptionHandler)
cmd += " "+Utils.joinOptions(((OptionHandler)generator).getOptions());
try {
// generate data
StringWriter output = new StringWriter(); generator.setOutput(new PrintWriter(output)); DataGenerator.makeData(generator, generator.getOptions()); m_Output.setText(output.toString());
}
catch (Exception ex) {
ex.printStackTrace(); JOptionPane.showMessageDialog(
getExplorer(), "Error generating data:\n" + ex.getMessage(),
"Error", JOptionPane.ERROR_MESSAGE); }
generator.setRelationName(relName); }
});
• the Use button finally fires a propertyChange event that will load the data into the Explorer:
m_ButtonUse.addActionListener(new ActionListener(){ public void actionPerformed(ActionEvent evt){
m_Support.firePropertyChange("", null, null); }
});
272 CHAPTER 17. EXTENDING WEKA • the propertyChange event will perform the actual loading of the data,
hence we add an anonymous property change listener to our panel:
addPropertyChangeListener(new PropertyChangeListener() { public void propertyChange(PropertyChangeEvent e) {
try {
Instances data = new Instances(new StringReader(m_Output.getText())); // set data in preprocess panel (also notifies of capabilties changes) getExplorer().getPreprocessPanel().setInstances(data);
}
catch (Exception ex) {
ex.printStackTrace(); JOptionPane.showMessageDialog(
getExplorer(), "Error generating data:\n" + ex.getMessage(),
"Error", JOptionPane.ERROR_MESSAGE); }
} });
• In order to add our GeneratorPanel to the list of tabs displayed in the Explorer, we need to modify the Explorer.props file (just extract it from the weka.jar and place it in your home directory). The Tabs property must look like this:
Tabs=weka.gui.explorer.GeneratorPanel:standalone,\ weka.gui.explorer.ClassifierPanel,\ weka.gui.explorer.ClustererPanel,\ weka.gui.explorer.AssociationsPanel,\ weka.gui.explorer.AttributeSelectionPanel,\ weka.gui.explorer.VisualizePanel
• Note: the standalone option is used to make the tab available without requiring the preprocess panel to load a dataset first.
Screenshot
17.4. EXTENDING THE EXPLORER 273
Experimenter ”light”
Purpose
By default the Classify panel only performs 1 run of 10-fold cross-validation. Since most classifiers are rather sensitive to the order of the data being pre- sented to them, those results can be too optimistic or pessimistic. Averaging the results over 10 runs with differently randomized train/test pairs returns more reliable results. And this is where this plugin comes in: it can be used to obtain statistical sound results for a specific classifier/dataset combination, without having to setup a whole experiment in the Experimenter.
Implementation
• Since this plugin is rather bulky, we omit the implementation details, but the following can be said:
– based on the weka.gui.explorer.ClassifierPanel
– the actual code doing the work follows the example in the Using the
Experiment API wiki article [2]
• In order to add our ExperimentPanel to the list of tabs displayed in the Explorer, we need to modify the Explorer.props file (just extract it from the weka.jar and place it in your home directory). The Tabs property must look like this:
Tabs=weka.gui.explorer.ClassifierPanel,\ weka.gui.explorer.ExperimentPanel,\ weka.gui.explorer.ClustererPanel,\ weka.gui.explorer.AssociationsPanel,\ weka.gui.explorer.AttributeSelectionPanel,\ weka.gui.explorer.VisualizePanel
Screenshot
274 CHAPTER 17. EXTENDING WEKA
17.4.2 Adding visualization plugins
Introduction
As of WEKA version 3.5.3 you can easily add visualization plugins in the Ex- plorer (Classify panel). This makes it easy to implement custom visualizations, if the ones WEKA offers are not sufficient. The following examples can be found in the Examples collection [3] (package wekaexamples.gui.visualize.plugins).
Requirements
• custom visualization class must implement the following interface weka.gui.visualize.plugins.VisualizePlugin
• the class must either reside in the following package (visualization classes are automatically discovered during run-time)
weka.gui.visualize.plugins
• or you must list the package this class belongs to in the properties file weka/gui/GenericPropertiesCreator.props (or the equivalent in your
home directory) under the key weka.gui.visualize.plugins.VisualizePlugin.
Implementation
The visualization interface contains the following four methods
• getMinVersion – This method returns the minimum version (inclusive) of WEKA that is necessary to execute the plugin, e.g., 3.5.0.
• getMaxVersion – This method returns the maximum version (exclusive) of WEKA that is necessary to execute the plugin, e.g., 3.6.0.
• getDesignVersion – Returns the actual version of WEKA this plugin was designed for, e.g., 3.5.1
• getVisualizeMenuItem – The JMenuItem that is returned via this method will be added to the plugins menu in the popup in the Explorer. The ActionListener for clicking the menu item will most likely open a new frame containing the visualized data.
17.4. EXTENDING THE EXPLORER 275
Examples
Table with predictions
The PredictionTable.java example simply displays the actual class label and the one predicted by the classifier. In addition to that, it lists whether it was an incorrect prediction and the class probability for the correct class label.
276 CHAPTER 17. EXTENDING WEKA Bar plot with probabilities
The PredictionError.java example uses the JMathTools library (needs the jmathplot.jar [27] in the CLASSPATH) to display a simple bar plot of the predictions. The correct predictions are displayed in blue, the incorrect ones in red. In both cases the class probability that the classifier returned for the correct class label is displayed on the y axis. The x axis is simply the index of the prediction starting with 0.
Chapter 18
Technical documentation
18.1 ANT
What is ANT? This is how the ANT homepage (http://ant.apache.org/) defines its tool:
Apache Ant is a Java-based build tool. In theory, it is kind of like Make, but without Make’s wrinkles.
18.1.1 Basics
• the ANT build file is based on XML (http://www.w3.org/XML/)
• the usual name for the build file is:
build.xml
• invocation—the usual build file needs not be specified explicitly, if it’s in the current directory; if not target is specified, the default one is used
ant [-f
• displaying all the available targets of a build file ant [-f
18.1.2 Weka and ANT
• a build file for Weka is available from subversion • some targets of interest
– clean–Removes the build, dist and reports directories; also any class files in the source tree
– compile–Compile weka and deposit class files in ${path_modifier}/build/classes
– docs–Make javadocs into ${path_modifier}/doc
– exejar–Create an executable jar file in ${path_modifier}/dist
277
278 CHAPTER 18. TECHNICAL DOCUMENTATION 18.2 CLASSPATH
The CLASSPATH environment variable tells Java where to look for classes. Since Java does the search in a first-come-first-serve kind of manner, you’ll have to take care where and what to put in your CLASSPATH. I, personally, never use the environment variable, since I’m working often on a project in different versions in parallel. The CLASSPATH would just mess up things, if you’re not careful (or just forget to remove an entry). ANT (http://ant.apache.org/) offers a nice way for building (and separating source code and class files) Java projects. But still, if you’re only working on totally separate projects, it might be easiest for you to use the environment variable.
18.2.1 Setting the CLASSPATH
In the following we add the mysql-connector-java-5.1.7-bin.jar to our CLASSPATH variable (this works for any other jar archive) to make it possible to access MySQL databases via JDBC.
Win32 (2k and XP)
We assume that the mysql-connector-java-5.1.7-bin.jar archive is located in the following directory:
C:\Program Files\Weka-3-7
In the Control Panel click on System (or right click on My Computer and select Properties) and then go to the Avanced tab. There you’ll find a button called Environment Variables, click it. Depending on, whether you’re the only person using this computer or it’s a lab computer shared by many, you can either create a new system-wide (you’re the only user) environment variable or a user depen- dent one (recommended for multi-user machines). Enter the following name for the variable.
CLASSPATH
and add this value
C:\Program Files\Weka-3-7\mysql-connector-java-5.1.7-bin.jar
If you want to add additional jars, you will have to separate them with the path separator, the semicolon “;” (no spaces!).
Unix/Linux
We make the assumption that the mysql jar is located in the following directory:
/home/johndoe/jars/
18.2. CLASSPATH 279 Open a shell and execute the following command, depending on the shell you’re
using:
• bash
export CLASSPATH=$CLASSPATH:/home/johndoe/jars/mysql-connector-java-5.1.7-bin.jar
• c shell
setenv CLASSPATH $CLASSPATH:/home/johndoe/jars/mysql-connector-java-5.1.7-bin.jar
Cygwin
The process is like with Unix/Linux systems, but since the host system is Win32 and therefore the Java installation also a Win32 application, you’ll have to use the semicolon ; as separator for several jars.
18.2.2 RunWeka.bat
From version 3.5.4, Weka is launched differently under Win32. The simple batch file got replaced by a central launcher class (= RunWeka.class) in combination with an INI-file (= RunWeka.ini). The RunWeka.bat only calls this launcher class now with the appropriate parameters. With this launcher approach it is possible to define different launch scenarios, but with the advantage of hav- ing placeholders, e.g., for the max heap size, which enables one to change the memory for all setups easily.
The key of a command in the INI-file is prefixed with cmd_, all other keys are considered placeholders:
cmd_blah=java … command blah bloerk= … placeholder bloerk
A placeholder is surrounded in a command with #:
cmd_blah=java #bloerk#
Note: The key wekajar is determined by the -w parameter with which the
launcher class is called.
By default, the following commands are predefined:
• default
The default Weka start, without a terminal window.
• console
For debugging purposes. Useful as Weka gets started from a terminal window.
• explorer
The command that’s executed if one double-clicks on an ARFF or XRFF file.
In order to change the maximum heap size for all those commands, one only has to modify the maxheap placeholder.
280 CHAPTER 18. TECHNICAL DOCUMENTATION
For more information check out the comments in the INI-file.
18.2.3 java -jar
When you’re using the Java interpreter with the -jar option, be aware of the fact that it overwrites your CLASSPATH and not augments it. Out of conve- nience, people often only use the -jar option to skip the declaration of the main class to start. But as soon as you need more jars, e.g., for database access, you need to use the -classpath option and specify the main class.
Here’s once again how you start the Weka Main-GUI with your current CLASSPATH variable (and 128MB for the JVM):
• Linux
java -Xmx128m -classpath $CLASSPATH:weka.jar weka.gui.Main
• Win32
java -Xmx128m -classpath “%CLASSPATH%;weka.jar” weka.gui.Main
18.3 Subversion 18.3.1 General
The Weka Subversion repository is accessible and browseable via the following URL:
https://svn.scms.waikato.ac.nz/svn/weka/
A Subversion repository has usually the following layout:
root |
+- trunk
|
+- tags
|
+- branches
Where trunk contains the main trunk of the development, tags snapshots in time of the repository (e.g., when a new version got released) and branches development branches that forked off the main trunk at some stage (e.g., legacy versions that still get bugfixed).
18.3.2 Source code
The latest version of the Weka source code can be obtained with this URL:
https://svn.scms.waikato.ac.nz/svn/weka/trunk/weka
If you want to obtain the source code of the book version, use this URL:
18.3. SUBVERSION 281 https://svn.scms.waikato.ac.nz/svn/weka/branches/book2ndEd-branch/weka
18.3.3 JUnit
The latest version of Weka’s JUnit tests can be obtained with this URL:
https://svn.scms.waikato.ac.nz/svn/weka/trunk/tests
And if you want to obtain the JUnit tests of the book version, use this URL:
https://svn.scms.waikato.ac.nz/svn/weka/branches/book2ndEd-branch/tests
18.3.4 Specific version
Whenever a release of Weka is generated, the repository gets tagged
• dev-X-Y-Z
the tag for a release of the developer version, e.g., dev-3.7.0 for Weka 3.7.0 https://svn.scms.waikato.ac.nz/svn/weka/tags/dev-3-7-0
• stable-X-Y-Z
the tag for a release of the book version, e.g., stable-3-4-15 for Weka 3.4.15 https://svn.scms.waikato.ac.nz/svn/weka/tags/stable-3-4-15
18.3.5 Clients
Commandline
Modern Linux distributions already come with Subversion either pre-installed or easily installed via the package manager of the distribution. If that shouldn’t be case, or if you are using Windows, you have to download the appropriate client from the Subversion homepage (http://subversion.tigris.org/).
A checkout of the current developer version of Weka looks like this:
svn co https://svn.scms.waikato.ac.nz/svn/weka/trunk/weka
SmartSVN
SmartSVN (http://smartsvn.com/) is a Java-based, graphical, cross-platform client for Subversion. Though it is not open-source/free software, the foundation version is for free.
TortoiseSVN
Under Windows, TortoiseCVS was a CVS client, neatly integrated into the Windows Explorer. TortoiseSVN (http://tortoisesvn.tigris.org/) is the equivalent for Subversion.
282 CHAPTER 18. TECHNICAL DOCUMENTATION 18.4 GenericOb jectEditor
18.4.1 Introduction
As of version 3.4.4 it is possible for WEKA to dynamically discover classes at runtime (rather than using only those specified in the GenericObjectEditor.props (GOE) file). In some versions (3.5.8, 3.6.0) this facility was not enabled by de- fault as it is a bit slower than the GOE file approach, and, furthermore, does not function in environments that do not have a CLASSPATH (e.g., application servers). Later versions (3.6.1, 3.7.0) enabled the dynamic discovery again, as WEKA can now distinguish between being a standalone Java application or being run in a non-CLASSPATH environment.
If you wish to enable or disable dynamic class discovery, the relevant file to edit is GenericPropertiesCreator.props (GPC). You can obtain this file either from the weka.jar or weka-src.jar archive. Open one of these files with an archive manager that can handle ZIP files (for Windows users, you can use 7-Zip (http://7-zip.org/) for this) and navigate to the weka/gui directory, where the GPC file is located. All that is required, is to change the UseDynamic property in this file from false to true (for enabling it) or the other way round (for disabling it). After changing the file, you just place it in your home directory. In order to find out the location of your home directory, do the following:
• Linux/Unix
– Open a terminal
– run the following command:
echo $HOME
• Windows
– Open a command-primpt
– run the following command:
echo %USERPROFILE%
If dynamic class discovery is too slow, e.g., due to an enormous CLASSPATH, you can generate a new GenericObjectEditor.props file and then turn dy- namic class discovery off again. It is assumed that you already placed the GPC file in your home directory (see steps above) and that the weka.jar jar archive with the WEKA classes is in your CLASSPATH (otherwise you have to add it to the java call using the -classpath option).
18.4. GENERICOBJECTEDITOR 283 For generating the GOE file, execute the following steps:
• generateanewGenericObjectEditor.propsfileusingthefollowingcom- mand:
– Linux/Unix
java weka.gui.GenericPropertiesCreator \ $HOME/GenericPropertiesCreator.props \ $HOME/GenericObjectEditor.props
– Windows (command must be in one line)
java weka.gui.GenericPropertiesCreator %USERPROFILE%\GenericPropertiesCreator.props %USERPROFILE%\GenericObjectEditor.props
• edit the GenericPropertiesCreator.props file in your home directory and set UseDynamic to false.
A limitation of the GOE prior to 3.4.4 was, that additional classifiers, filters, etc., had to fit into the same package structure as the already existing ones, i.e., all had to be located below weka. WEKA can now display multiple class hierarchies in the GUI, which makes adding new functionality quite easy as we will see later in an example (it is not restricted to classifiers only, but also works with all the other entries in the GPC file).
18.4.2 File Structure
The structure of the GOE is a key-value-pair, separated by an equals-sign. The value is a comma separated list of classes that are all derived from the su- perclass/superinterface key. The GPC is slightly different, instead of declar- ing all the classes/interfaces one need only to specify all the packages de- scendants are located in (only non-abstract ones are then listed). E.g., the weka.classifiers.Classifier entry in the GOE file looks like this:
weka.classifiers.Classifier=\ weka.classifiers.bayes.AODE,\ weka.classifiers.bayes.BayesNet,\ weka.classifiers.bayes.ComplementNaiveBayes,\ weka.classifiers.bayes.NaiveBayes,\ weka.classifiers.bayes.NaiveBayesMultinomial,\ weka.classifiers.bayes.NaiveBayesSimple,\ weka.classifiers.bayes.NaiveBayesUpdateable,\ weka.classifiers.functions.LeastMedSq,\ weka.classifiers.functions.LinearRegression,\ weka.classifiers.functions.Logistic,\
…
The entry producing the same output for the classifiers in the GPC looks like this (7 lines instead of over 70 for WEKA 3.4.4):
weka.classifiers.Classifier=\ weka.classifiers.bayes,\ weka.classifiers.functions,\ weka.classifiers.lazy,\ weka.classifiers.meta,\ weka.classifiers.trees,\ weka.classifiers.rules
284 CHAPTER 18. TECHNICAL DOCUMENTATION 18.4.3 Exclusion
It may not always be desired to list all the classes that can be found along the CLASSPATH. Sometimes, classes cannot be declared abstract but still shouldn’t be listed in the GOE. For that reason one can list classes, interfaces, superclasses for certain packages to be excluded from display. This exclusion is done with the following file:
weka/gui/GenericPropertiesCreator.excludes
The format of this properties file is fairly simple:
Where the
file and the
any class derived from this will be excluded
• I – Interface
any class implementing this interface will be excluded
• C – Class
exactly this class will be excluded
Here are a few examples:
# exclude all ResultListeners that also implement the ResultProducer interface # (all ResultProducers do that!)
weka.experiment.ResultListener=\
I:weka.experiment.ResultProducer
# exclude J48 and all SingleClassifierEnhancers weka.classifiers.Classifier=\
C:weka.classifiers.trees.J48,\ S:weka.classifiers.SingleClassifierEnhancer
18.4. GENERICOBJECTEDITOR 285 18.4.4 Class Discovery
Unlike the Class.forName(String) method that grabs the first class it can find in the CLASSPATH, and therefore fixes the location of the package it found the class in, the dynamic discovery examines the complete CLASSPATH you are starting the Java Virtual Machine (= JVM) with. This means that you can have several parallel directories with the same WEKA package structure, e.g., the standard release of WEKA in one directory (/distribution/weka.jar) and another one with your own classes (/development/weka/…), and display all of the classifiers in the GUI. In case of a name conflict, i.e., two directories contain the same class, the first one that can be found is used. In a nutshell, your java call of the GUIChooser can look like this:
java -classpath “/development:/distribution/weka.jar” weka.gui.GUIChooser
Note: Windows users have to replace the “:” with “;” and the forward slashes with backslashes.
18.4.5 Multiple Class Hierarchies
In case you are developing your own framework, but still want to use your clas- sifiers within WEKA that was not possible with WEKA prior to 3.4.4. Starting with the release 3.4.4 it is possible to have multiple class hierarchies being dis- played in the GUI. If you have developed a modified version of NaiveBayes, let us call it DummyBayes and it is located in the package dummy.classifiers then you will have to add this package to the classifiers list in the GPC file like this:
weka.classifiers.Classifier=\ weka.classifiers.bayes,\ weka.classifiers.functions,\ weka.classifiers.lazy,\ weka.classifiers.meta,\ weka.classifiers.trees,\ weka.classifiers.rules,\ dummy.classifiers
286 CHAPTER 18. TECHNICAL DOCUMENTATION Your java call for the GUIChooser might look like this:
java -classpath “weka.jar:dummy.jar” weka.gui.GUIChooser
Starting up the GUI you will now have another root node in the tree view of the classifiers, called root, and below it the weka and the dummy package hierarchy as you can see here:
18.4. GENERICOBJECTEDITOR 287 18.4.6 Capabilities
Version 3.5.3 of Weka introduced the notion of Capabilities. Capabilities basi- cally list what kind of data a certain object can handle, e.g., one classifier can handle numeric classes, but another cannot. In case a class supports capabili- ties the additional buttons Filter… and Remove filter will be available in the GOE. The Filter… button pops up a dialog which lists all available Capabilities:
One can then choose those capabilities an object, e.g., a classifier, should have. If one is looking for classification problem, then the Nominal class Capability can be selected. On the other hand, if one needs a regression scheme, then the Capability Numeric class can be selected. This filtering mechanism makes the search for an appropriate learning scheme easier. After applying that filter, the tree with the objects will be displayed again and lists all objects that can handle all the selected Capabilities black, the ones that cannot grey and the ones that might be able to handle them blue (e.g., meta classifiers which depend on their base classifier(s)).
288 CHAPTER 18. TECHNICAL DOCUMENTATION 18.5 Properties
A properties file is a simple text file with this structure:
Comments start with the hash sign #.
To make a rather long property line more readable, one can use a backslash to continue on the next line. The Filter property, e.g., looks like this:
weka.filters.Filter= \ weka.filters.supervised.attribute, \ weka.filters.supervised.instance, \ weka.filters.unsupervised.attribute, \ weka.filters.unsupervised.instance
18.5.1 Precedence
The Weka property files (extension .props) are searched for in the following order:
• current directory
• the user’s home directory (*nix $HOME, Windows %USERPROFILE%) • the class path (normally the weka.jar file)
If Weka encounters those files it only supplements the properties, never overrides them. In other words, a property in the property file of the current directory has a higher precedence than the one in the user’s home directory.
Note: Under Cywgin (http://cygwin.com/), the home directory is still the Windows one, since the java installation will be still one for Windows.
18.5.2 Examples
• weka/gui/LookAndFeel.props
• weka/gui/GenericPropertiesCreator.props • weka/gui/beans/Beans.props
18.6. XML 289 18.6 XML
Weka now supports XML (http://www.w3c.org/XML/) (eXtensible Markup Language) in several places.
18.6.1 Command Line
WEKA now allows Classifiers and Experiments to be started using an -xml option followed by a filename to retrieve the command line options from the XML file instead of the command line.
For such simple classifiers like e.g. J48 this looks like overkill, but as soon as one uses Meta-Classifiers or Meta-Meta-Classifiers the handling gets tricky and one spends a lot of time looking for missing quotes. With the hierarchical structure of XML files it is simple to plug in other classifiers by just exchanging tags.
The DTD for the XML options is quite simple:
] >
The type attribute of the option tag needs some explanations. There are cur- rently four different types of options in WEKA:
• flag
The simplest option that takes no arguments, like e.g. the -V flag for inversing an selection.
• single
The option takes exactly one parameter, directly following after the op- tion, e.g., for specifying the trainings file with -t somefile.arff. Here the parameter value is just put between the opening and closing tag. Since single is the default value for the type tag we don’t need to specify it ex- plicitly.290 CHAPTER 18. TECHNICAL DOCUMENTATION
• hyphens
Meta-Classifiers like AdaBoostM1 take another classifier as option with the -W option, where the options for the base classifier follow after the –. And here it is where the fun starts: where to put parameters for the base classifier if the Meta-Classifier itself is a base classifier for another Meta-Classifier?
E.g., does -W weka.classifiers.trees.J48 — -C 0.001 become this:Internally, all the options enclosed by the options tag are pushed to the
end after the — if one transforms the XML into a command line string.
• quotes
A Meta-Classifier like Stacking can take several -B options, where each single one encloses other options in quotes (this itself can contain a Meta- Classifier!). From -B ‘‘weka.classifiers.trees.J48’’ we then get this XML:With the XML representation one doesn’t have to worry anymore about the level of quotes one is using and therefore doesn’t have to care about the correct escaping (i.e. ‘‘ … \” … \” …’’) since this is done automatically.
18.6. XML 291 And if we now put all together we can transform this more complicated com-
mand line (java and the CLASSPATH omitted):
Note: The type and value attribute of the outermost options tag is not used while reading the parameters. It is merely for documentation purposes, so that one knows which class was actually started from the command line.
Responsible Class(es): weka.core.xml.XMLOptions
292 CHAPTER 18. TECHNICAL DOCUMENTATION 18.6.2 Serialization of Experiments
It is now possible to serialize the Experiments from the WEKA Experimenter not only in the proprietary binary format Java offers with serialization (with this you run into problems trying to read old experiments with a newer WEKA version, due to different SerialUIDs), but also in XML. There are currently two different ways to do this:
• built-in
The built-in serialization captures only the necessary informations of an experiment and doesn’t serialize anything else. It’s sole purpose is to save the setup of a specific experiment and can therefore not store any built models. Thanks to this limitation we’ll never run into problems with mismatching SerialUIDs.
This kind of serialization is always available and can be selected via a Filter (*.xml) in the Save/Open-Dialog of the Experimenter.
The DTD is very simple and looks like this (for version 3.4.5):
]>
CDATA #REQUIRED>
CDATA “no”>
CDATA “no”>
CDATA “no”>
CDATA “3.4.5”>
Prior to versions 3.4.5 and 3.5.0 it looked like this:
] >
Responsible Class(es): weka.experiment.xml.XMLExperiment
for general Serialization:
weka.core.xml.XMLSerialization weka.core.xml.XMLBasicSerialization
18.6. XML 293
• KOML (http://old.koalateam.com/xml/serialization/)
The Koala Object Markup Language (KOML) is published under the LGPL (http://www.gnu.org/copyleft/lgpl.html) and is an alterna- tive way of serializing and derserializing Java Objects in an XML file. Like the normal serialization it serializes everything into XML via an Ob- jectOutputStream, including the SerialUID of each class. Even though we have the same problems with mismatching SerialUIDs it is at least pos- sible edit the XML files by hand and replace the offending IDs with the new ones.
In order to use KOML one only has to assure that the KOML classes are in the CLASSPATH with which the Experimenter is launched. As soon as KOML is present another Filter (*.koml) will show up in the Save/Open-Dialog.
The DTD for KOML can be found at http://old.koalateam.com/xml/koml12.dtd Responsible Class(es):
weka.core.xml.KOML
The experiment class can of course read those XML files if passed as input or out-
put file (see options of weka.experiment.Experiment and weka.experiment.RemoteExperiment
18.6.3 Serialization of Classifiers
The options for models of a classifier, -l for the input model and -d for the out- put model, now also supports XML serialized files. Here we have to differentiate between two different formats:
• built-in
The built-in serialization captures only the options of a classifier but not the built model. With the -l one still has to provide a training file, since we only retrieve the options from the XML file. It is possible to add more options on the command line, but it is no check performed whether they collide with the ones stored in the XML file.
The file is expected to end with .xml.
• KOML
Since the KOML serialization captures everything of a Java Object we can use it just like the normal Java serialization.
The file is expected to end with .koml.
The built-in serialization can be used in the Experimenter for loading/saving options from algorithms that have been added to a Simple Experiment. Unfor- tunately it is not possible to create such a hierarchical structure like mentioned in Section 18.6.1. This is because of the loss of information caused by the getOptions() method of classifiers: it returns only a flat String-Array and not a tree structure.
294 CHAPTER 18. TECHNICAL DOCUMENTATION Responsible Class(es):
weka.core.xml.KOML weka.classifiers.xml.XMLClassifier
18.6.4 Bayesian Networks
The GraphVisualizer (weka.gui.graphvisualizer.GraphVisualizer) can save graphs into the Interchange Format (http://www-2.cs.cmu.edu/∼fgcozman/Research/InterchangeFormat/) for Bayesian Networks (BIF). If started from command line with an XML filename as first parameter and not from the Explorer it can display the given file directly.
The DTD for BIF is this:
]>
Responsible Class(es):
weka.classifiers.bayes.BayesNet#toXMLBIF03() weka.classifiers.bayes.net.BIFReader weka.gui.graphvisualizer.BIFParser
18.6.5 XRFF files
With Weka 3.5.4 a new, more feature-rich, XML-based data format got intro- duced: XRFF. For more information, please see Chapter 10.
Chapter 19
Other resources
19.1 Mailing list
The WEKA Mailing list can be found here:
• http://list.scms.waikato.ac.nz/mailman/listinfo/wekalist for subscribing/unsubscribing the list
• https://list.scms.waikato.ac.nz/pipermail/wekalist/ (Mirrors: http://news.gmane.org/gmane.comp.ai.weka, http://www.nabble.com/WEKA-f435.html)
for searching previous posted messages
Before posting, please read the Mailing List Etiquette:
http://www.cs.waikato.ac.nz/∼ml/weka/mailinglist etiquette.html 19.2 Troubleshooting
Here are a few of things that are useful to know when you are having trouble installing or running Weka successfullyon your machine.
NB these java commands refer to ones executed in a shell (bash, command prompt, etc.) and NOT to commands executed in the SimpleCLI.
19.2.1 Weka download problems
When you download Weka, make sure that the resulting file size is the same as on our webpage. Otherwise things won’t work properly. Apparently some web browsers have trouble downloading Weka.
19.2.2 OutOfMemoryException
Most Java virtual machines only allocate a certain maximum amount of memory to run Java programs. Usually this is much less than the amount of RAM in your computer. However, you can extend the memory available for the virtual machine by setting appropriate options. With Sun’s JDK, for example, you can go
295
296 CHAPTER 19. OTHER RESOURCES java -Xmx100m …
to set the maximum Java heap size to 100MB. For more information about these options see http://java.sun.com/docs/hotspot/VMOptions.html.
19.2.2.1 Windows
Book version
You have to modify the JVM invocation in the RunWeka.bat batch file in your installation directory.
Developer version
• up to Weka 3.5.2
just like the book version.
• Weka 3.5.3
You have to modify the link in the Windows Start menu, if you’re starting the console-less Weka (only the link with console in its name executes the RunWeka.bat batch file)
• Weka 3.5.4 and higher Due to the new launching scheme, you no longer modify the batch file, but the RunWeka.ini file. In that particular file, you’ll have to change the maxheap placeholder. See section 18.2.2.
19.2.3 Mac OSX
In your Weka installation directory (weka-3-x-y.app) locate the Contents sub- directory and edit the Info.plist file. Near the bottom of the file you should see some text like:
Alter the 256M to something higher.
19.2.4 StackOverflowError
Try increasing the stack of your virtual machine. With Sun’s JDK you can use this command to increase the stacksize:
java -Xss512k …
19.2. TROUBLESHOOTING 297 to set the maximum Java stack size to 512KB. If still not sufficient, slowly
increase it.
19.2.5 just-in-time (JIT) compiler
For maximum enjoyment, use a virtual machine that incorporates a just-in- time compiler. This can speed things up quite significantly. Note also that there can be large differences in execution time between different virtual ma- chines.
19.2.6 CSV file conversion
Either load the CSV file in the Explorer or use the CVS converter on the com- mandline as follows:
java weka.core.converters.CSVLoader filename.csv > filename.arff
19.2.7 ARFF file doesn’t load
One way to figure out why ARFF files are failing to load is to give them to the Instances class. At the command line type the following:
java weka.core.Instances filename.arff
where you substitute ’filename’ for the actual name of your file. This should return an error if there is a problem reading the file, or show some statistics if the file is ok. The error message you get should give some indication of what is wrong.
19.2.8 Spaces in labels of ARFF files
A common problem people have with ARFF files is that labels can only have spaces if they are enclosed in single quotes, i.e. a label such as:
some value
should be written either ’some value’ or some value in the file.
19.2.9 CLASSPATH problems
Having problems getting Weka to run from a DOS/UNIX command prompt? Gettingjava.lang.NoClassDefFoundErrorexceptions? MostlikelyyourCLASS- PATH environment variable is not set correctly – it needs to point to the Weka.jar file that you downloaded with Weka (or the parent of the Weka direc- tory if you have extracted the jar). Under DOS this can be achieved with:
set CLASSPATH=c:\weka-3-4\weka.jar;%CLASSPATH%
298 CHAPTER 19. OTHER RESOURCES
Under UNIX/Linux something like:
export CLASSPATH=/home/weka/weka.jar:$CLASSPATH
An easy way to get avoid setting the variable this is to specify the CLASSPATH when calling Java. For example, if the jar file is located at c:\weka-3-4\weka.jar you can use:
java -cp c:\weka-3-4\weka.jar weka.classifiers… etc.
See also Section 18.2.
19.2.10 Instance ID
People often want to tag their instances with identifiers, so they can keep track of them and the predictions made on them.
19.2.10.1 Adding the ID
A new ID attribute is added real easy: one only needs to run the AddID filter over the dataset and it’s done. Here’s an example (at a DOS/Unix command prompt):
java weka.filters.unsupervised.attribute.AddID -i data_without_id.arff
-o data_with_id.arff
(all on a single line).
Note: the AddID filter adds a numeric attribute, not a String attribute
to the dataset. If you want to remove this ID attribute for the classifier in a FilteredClassifier environment again, use the Remove filter instead of the RemoveType filter (same package).
19.2.10.2 Removing the ID
If you run from the command line you can use the -p option to output predic- tions plus any other attributes you are interested in. So it is possible to have a string attribute in your data that acts as an identifier. A problem is that most classifiers don’t like String attributes, but you can get around this by using the RemoveType (this removes String attributes by default).
Here’s an example. Lets say you have a training file named train.arff, a testing file named test.arff, and they have an identifier String attribute as their 5th attribute. You can get the predictions from J48 along with the identifier strings by issuing the following command (at a DOS/Unix command prompt):
java weka.classifiers.meta.FilteredClassifier
-F weka.filters.unsupervised.attribute.RemoveType -W weka.classifiers.trees.J48
-t train.arff -T test.arff -p 5
(all on a single line).
19.2. TROUBLESHOOTING 299
If you want, you can redirect the output to a file by adding “> output.txt” to the end of the line.
In the Explorer GUI you could try a similar trick of using the String attribute identifiers here as well. Choose the FilteredClassifier, with RemoveType as the filter, and whatever classifier you prefer. When you visualize the results you will need click through each instance to see the identifier listed for each.
19.2.11 Visualization
Access to visualization from the ClassifierPanel, ClusterPanel and Attribute- Selection panel is available from a popup menu. Click the right mouse button over an entry in the Result list to bring up the menu. You will be presented with options for viewing or saving the text output and—depending on the scheme— further options for visualizing errors, clusters, trees etc.
19.2.12 Memory consumption and Garbage collector
There is the ability to print how much memory is available in the Explorer and Experimenter and to run the garbage collector. Just right click over the Status area in the Explorer/Experimenter.
19.2.13 GUIChooser starts but not Experimenter or Ex- plorer
The GUIChooser starts, but Explorer and Experimenter don’t start and output an Exception like this in the terminal:
/usr/share/themes/Mist/gtk-2.0/gtkrc:48: Engine “mist” is unsupported, ignoring —Registering Weka Editors—
java.lang.NullPointerException
at weka.gui.explorer.PreprocessPanel.addPropertyChangeListener(PreprocessPanel.java:519) at javax.swing.plaf.synth.SynthPanelUI.installListeners(SynthPanelUI.java:49)
at javax.swing.plaf.synth.SynthPanelUI.installUI(SynthPanelUI.java:38)
at javax.swing.JComponent.setUI(JComponent.java:652)
at javax.swing.JPanel.setUI(JPanel.java:131) …
This behavior happens only under Java 1.5 and Gnome/Linux, KDE doesn’t produce this error. The reason for this is, that Weka tries to look more “native” and therefore sets a platform-specific Swing theme. Unfortunately, this doesn’t seem to be working correctly in Java 1.5 together with Gnome. A workaround for this is to set the cross-platform Metal theme.
In order to use another theme one only has to create the following properties file in ones home directory:
LookAndFeel.props
With this content:
300 CHAPTER 19. OTHER RESOURCES Theme=javax.swing.plaf.metal.MetalLookAndFeel
19.2.14 KnowledgeFlow toolbars are empty
In the terminal, you will most likely see this output as well:
Failed to instantiate: weka.gui.beans.Loader
This behavior can happen under Gnome with Java 1.5, see Section 19.2.13 for a solution.
19.2.15 Links
• JavaVMoptions(http://java.sun.com/docs/hotspot/VMOptions.html)
301
302 BIBLIOGRAPHY
Bibliography
[1] Witten, I.H. and Frank, E. (2005) Data Mining: Practical machine learn- ing tools and techniques. 2nd edition Morgan Kaufmann, San Francisco.
[2] WekaWiki – http://weka.wikispaces.com/
[3] Weka Examples – A collection of example classes, as part of
an ANT project, included in the WEKA snapshots (available
for download on the homepage) or directly from subversion https://svn.scms.waikato.ac.nz/svn/weka/branches/stable-3-6/wekaexamples/
[4] J. Platt (1998): Machines using Sequential Minimal Optimization. In B. Schoelkopf and C. Burges and A. Smola, editors, Advances in Kernel Methods – Support Vector Learning.
[5] Drummond, C. and Holte, R. (2000) Explicitly representing expected cost: An alternative to ROC representation. Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Publishers, San Mateo, CA.
[6] Extensions for Weka’s main GUI on WekaWiki – http://weka.wikispaces.com/Extensions+for+Weka%27s+main+GUI
[7] Adding tabs in the Explorer on WekaWiki – http://weka.wikispaces.com/Adding+tabs+in+the+Explorer
[8] Explorer visualization plugins on WekaWiki – http://weka.wikispaces.com/Explorer+visualization+plugins
[9] Bengio, Y. and Nadeau, C. (1999) Inference for the Generalization Error.
[10] Ross Quinlan (1993). C4.5: Programs for Machine Learning, Morgan Kaufmann
Publishers, San Mateo, CA.
[11] Subversion – http://weka.wikispaces.com/Subversion
[12] HSQLDB – http://hsqldb.sourceforge.net/
[13] MySQL – http://www.mysql.com/
[14] Plotting multiple ROC curves on WekaWiki – http://weka.wikispaces.com/Plotting+multiple+ROC+curves
[15] R.R. Bouckaert. Bayesian Belief Networks: from Construction to Inference. Ph.D. thesis, University of Utrecht, 1995.
[16] W.L. Buntine. A guide to the literature on learning probabilistic networks from data. IEEE Transactions on Knowledge and Data Engineering, 8:195–210, 1996.
[17] J. Cheng, R. Greiner. Comparing bayesian network classifiers. Proceedings UAI, 101–107, 1999.
[18] C.K. Chow, C.N.Liu. Approximating discrete probability distributions with de- pendence trees. IEEE Trans. on Info. Theory, IT-14: 426–467, 1968.
BIBLIOGRAPHY 303
[19] G. Cooper, E. Herskovits. A Bayesian method for the induction of probabilistic
networks from data. Machine Learning, 9: 309–347, 1992.
[20] Cozman.Seehttp://www-2.cs.cmu.edu/∼fgcozman/Research/InterchangeFormat/ for details on XML BIF.
[21] N. Friedman, D. Geiger, M. Goldszmidt. Bayesian Network Classifiers. Machine Learning, 29: 131–163, 1997.
[22] D. Heckerman, D. Geiger, D. M. Chickering. Learning Bayesian networks: the combination of knowledge and statistical data. Machine Learining, 20(3): 197– 243, 1995.
[23] S.L. Lauritzen and D.J. Spiegelhalter. Local Computations with Probabilities on graphical structures and their applications to expert systems (with discussion). Journal of the Royal Statistical Society B. 1988, 50, 157-224
[24] Moore, A. and Lee, M.S. Cached Sufficient Statistics for Efficient Machine Learning with Large Datasets, JAIR, Volume 8, pages 67-91, 1998.
[25] Verma, T. and Pearl, J.: An algorithm for deciding if a set of observed indepen- dencies has a causal explanation. Proc. of the Eighth Conference on Uncertainty in Artificial Intelligence, 323-330, 1992.
[26] GraphViz.Seehttp://www.graphviz.org/doc/info/lang.htmlformoreinfor- mation on the DOT language.
[27] JMathPlot. See http://code.google.com/p/jmathplot/ for more information on the project.