Assignment 3: Frequent Itemsets, Clustering, Advertising
Formative, Weight (15%), Learning objectives (1, 2, 3), Abstraction (4), Design (4), Communication (4), Data (5), Programming (5)
1 Overview
Read the following carefully as it di↵ers from the last assignment.
For students who are taking the course COMP SCI 3306 (i.e., undergraduate students), this assignment can be done in groups consisting of two students. If you have problems finding a group partner use the forum to search for group partners.
For other students who are taking the course COMP SCI 7306, this assign- ment should be done individually.
References to sections, examples, etc. refer to the book of “Leskovec, Ra- jaraman and Ullman: Mining Massive Datasets (Second Edition)”.
2 Assignment
Exercise 1 Frequent Itemsets (15+15+10+10 points)
For this exercise, you have to read Section 6.4 up to 6.4.3.
1. Implement the simple, randomized algorithm given in 6.4.1
2. Implement the algorithm of Savasere, Omiecinski, and Navathe (SON al- gorithm) in 6.4.3
3. Compare the two algorithms on the datasets T10I4D100K, T40I10D100K, chess, connect, mushroom, pumsb, pumsb star provided at
http://fimi.ua.ac.be/data/
and report the outcomes.
1
COMP SCI 3306, COMP SCI 7306 Mining Big Data Semester 1, 2020
4. Experiment with dierent sample sizes in the simple randomized algorithm such as 1, 2, 5, 10% and compare your results (including the result pro- duced by the SON algorithm).
Your approach should be as e cient as possible in terms of runtime and memory requirements.
Report on challenges that you might have observed in the implementation and by running experiments.
Exercise 2 Clustering (10+20 points)
1. Perform a hierarchical clustering on the one-dimensional set of points
1,4,9,16,25,36,49,64,81.
assuming the clusters are represented by their centroid (average), and at each step the clusters with the closest centroids are merged. (Exercise 7.2.1)
2. Implement the K-means algorithm and carry out experiments on the pro- vided Iris dataset.
a) You are asked to plot the K-means results by plotting the first 2 di- mensions of the input data as well as the converged centroids.
b) Provide some discussions about how you pick the value of K in K-means.
For the Iris data, only use the first 4 dimension for this exercise. In other words, discard the label information.
Exercise 3 Advertising (Exercise 8.4.1) (10+10 points)
Consider Example 8.7. Suppose that there are three advertisers A,B, and
C. There are three queries x,y, and z. Each advertiser has a budget of 2. Advertiser A only bids on x, B bids on x and y, and C bids on x,y, and z. Note that on the query sequence xxyyzz, the optimal o ne algorithm would yield a revenue of 6, since all queries can be assigned.
3
1. 2.
Show that the greedy algorithm will assign at least 4 of the 6 queries xxyyzz.
Find another sequence of queries such that the greedy algorithm can assign as few as half the queries that the optimal o✏ine algorithm would assign to that sequence.
Procedure for handing in the assignment
Work should be handed in using Canvas. The submission should include:
2
COMP SCI 3306, COMP SCI 7306 Mining Big Data Semester 1, 2020
• a PDF file of your solutions for theoretical assignments. The solutions should contain of a detailed description of how to obtain the result.
For Exercise 2.2, you should properly provide comments in your code to show your understanding.
• all source files, all the project files.
• a README.txt file containing instructions to run the code, the names,
student numbers, and email addresses of the group members.
3