WEEK 1 TERM 2:
MSIN0097
Predictive Analytics
A P MOORE
PREDICTIVE ANALYTICS
A bit about me
MY BACKGROUND CURRENT | PREVIOUS
Alastair Moore
Head of Analytics and Machine Learning
Senior Teaching Fellow Predictive Analytics
MSc Business Analytics
UCL School of Management
MBA Programme,
Innovation and Entrepreneurship in Europe Emerging Business Technologies
BiMBA Peking University
Co-Founder Satalia.com
Co-Founder WeArePopUp.com
Par tner AMAAMS LLP
SATALIA.COM
MISHCON DE REYA
OPEN BANKING
WORKING WITH NESTA ON CHALLENGE LEAD APPROACHES
Source: www.openupchallenge.io
DLT SYSTEMS
USE OF DISTRIBUTED LEDGERS IN REAL ESTATE
How will we transact property in the future?
Source: www.gov.uk/government/news/hm-land-registry-to-explore-the-benefits-of-blockchain
RESEARCH INTERESTS
PREDICTIVE ANALYTICS
Course overview
TODAY
LECTURE / WORKSHOP SCHEDULE
~90 mins
8:30 am
9:30 am
10:30 am
11:30 am
Video 1
Video 2
Video 3
Asynchronous: Video lectures
1hr
2hr
3hr
90 mins
Review
Problem 1
Discussion
Problem 2
Discussion
Synchronous: Lectures/Tutorials
HANDS ON MACHINE LEARNING GIT
Week 1 Week 2
Week 3
Week 6
Week 7 Week 8
Source: https://github.com/ageron/handson-ml
Week 4 Week 5
TEACHING SUPPORT
Kamil Tylinski Teaching Assistant
kamil.tylinski.16@ucl.ac.uk
Jiangbo Shangguan Teaching Assistant
j.shangguan.17@ucl.ac.uk
Bartos Kultys Teaching Assistant
bartosz.kultys.18@ucl.ac.uk
Editha Nemsic
Teaching Assistant
editha.nemsic.19@ucl.ac.uk
Dr Viviana Culmone Teaching Assistant
v.culmone@ucl.ac.uk
Walter Hernandez
Teaching Assistant
walter.hernandez.18@ucl.ac.uk
ASSESSMEN TS
— Individual Coursework – 60% – 2000words
– Due Date: Friday 26th February 2021
— Group Coursework – 40% – 4-5pergroup
– Deadline for Group formation: Friday 29th January 2021 (week 3) – 2000words
– Due Date:Thursday 18th March 2021
PREDICTIVE ANALYTICS
Review
DEEP LEARNING STATE OF THE ART
– Initial human generated text
– Part automated text generation
– Automated video generation
– Automated audio generation
– Automated translation
WAVES OF INVESTMENT THE ROLE OF INDUSTRIALIZATION
https://machinelearnings.co/winning-strategies-for-applied-ai-companies-f02cac0a6ad8
TOWER & MOAT
Source: https://blog.gardeviance.org/2014/07/tower-and-moat.html
MACHINE LEARNING JARGON
— Model
— Interpolating / Extrapolating — Data Bias
— Noise / Outliers
— Learning algorithm
— Inference algorithm
— Supervised learning
— Unsupervised learning
— Classification
— Regression
— Clustering
— Decomposition
— Parameters
— Optimisation
— Training data
— Testing data
— Error metric
— Linear model
— Parametric model
— Model variance
— Model bias
— Model generalization
— Overfitting
— Goodness-of-fit
— Hyper-parameters
— Failure modes
— Confusion matrix
— True Positive
— False Negative
— Data density
— Partition
— Hidden parameter
— High dimensional space
— Low dimensional space
— Separable data
— Manifold / Decision surface
— Hyper cube / volume / plane
机器学习 行话
— 模型
— 内插 / 外推 — 数据偏差
— 噪声/离群值 — 学习算法
— 推断算法
— 监督学习
— 无监督学习 — 分类
— 回归
— 聚类
— 分解
— 参数
— 优化
— 训练数据 — 测试数据 — 误差指标 — 线性模型 — 参数模型 — 模型方差 — 模型偏差 — 模型泛化 — 过拟合 — 拟合优度 — 超参数
— 失败模式
— 混淆矩阵
— 真正例
— 假反例
— 数据密度
— 划分
— 隐藏参数
— 高维空间
— 低维空间
— 可分数据
— 流形/ 决策面
— 超立方体/超体积/超平 面
MODEL / 模型 MÓXÍNG
MACHINE LEARNING
Data + modelàprediction
MACHINE LEARNING DATA DRIVEN AI
Assume there is enough data to find statistical associations to solve specific tasks
Data + modelàprediction
Define how well the model solves the task and adapt the parameters to maximize performance
LEARNING A FUNCTION
𝑥→𝑦
𝑥 →𝑓(𝑥)→𝑦
LEARNING A FUNCTION
𝑥→𝑦
𝑥 →𝑓(𝑥)→𝑦
Measured data
Features Inferred/Predicted/Estimated value
Trueinitialvalue𝑥 →𝑥’→𝑓 𝑥 =𝑦’ →𝑦
(world state) True target value
Learned/Fitted function (world state) From n observations
LEARNING A FUNCTION
𝑥→𝑦
𝑥 →𝑓(𝑥)→𝑦
Measured data
Features Inferred/Predicted/Estimated value
Trueinitialvalue𝑥 →𝑥’→𝑓 𝑥 =𝑦’ →𝑦
(world state) True target value
Learned/Fitted function (world state) From n observations
input 𝑥→ 𝑓 𝑥 →𝑦 output
INTERPOLATING / 内插 NÈI CHĀ
EXTRAPOLATING / 外推 WAÌ TUĪ
NOISE, OUTLIERS / 噪声,离群值 ZÀOSHĒNG , LÍ QÚN ZHÍ
LEARNING ALGORITHM / 学习算法 XUÉXÍ SUÀNFǍ
INFERENCE ALGORITHM / 推断算法 TUĪDUÀN SUÀNFǍ
SUPERVISED LEARNING / 监督学习 JIĀNDŪ XUÉXÍ
UNSUPERVISED LEARNING / 无监督学习 WÚ JIĀNDŪ XUÉXÍ
CLASSIFICATION / 分类 FĒNLÈI
REGRESSION / 回归分析 HUÍGUĪ FĒNXĪ
CLUSTERING / 聚类 JÙ LÈI
DECOMPOSITION / 分解 FĒNJIĚ
PREDICTIVE ANALYTICS
Perception, Patterns and Gestalt
PREDICTIVE ANALYTICS
The increase in computing power
TRAINING TIMES
PREDICTIVE ANALYTICS
Problem 1
~15 mins group work ~15 mins discussion
PRACTICAL TOOLS ML CANVAS
The Machine Leaning Cana (0.4) Deiged f: Deiged b: Dae: Ieai: .
Learn
Deciin
H ae edici ed
ake decii ha ide
he ed ale he ed-e?
ML ak
I, edic, e f ble.
Vale Piin
Wha ae e ig d f he ed-e() f he edicie e? Wha bjecie ae e eig?
Daa Sce
Which a daa ce ca e e (ieal ad eeal)?
Cllecing Daa
H d e ge e daa lea f (i ad )?
Feae
I eeeai eaced f a daa ce.
Bilding Mdel
Whe d e ceae/dae
del ih e aiig
daa? H lg d e hae feaie aiig i ad ceae a del?
Making Pedicin
Whe d e ake edici e i? H lg d e hae feaie a e i ad ake a edici?
Offline Ealain
Mehd ad eic ealae he e befe dele.
Lie Ealain and Mniing
Mehd ad eic ealae he e afe dele, ad aif ale ceai.
machineleaningcana.cm b Li Dad, Ph.D. Liceed de a Ceaie C Aibi-ShaeAlike 4.0 Ieaial Licee.
The Machine Leaning Cana (0.4) Deiged f: Deiged b: Dae: Ieai: .
Deciin
H ae edici ed
ake decii ha ide
he ed ale he ed-e?
ML ak
I, edic, e f ble.
Making Pedicin
Whe d e ake edici e i? H lg d e hae feaie a e i ad ake a edici?
Offline Ealain
Mehd ad eic ealae he e befe dele.
Vale Piin
Wha ae e ig d f he ed-e() f he edicie e? Wha bjecie ae e eig?
Daa Sce
Which a daa ce ca e e (ieal ad eeal)?
Cllecing Daa
H d e ge e daa lea f (i ad )?
Feae
I eeeai eaced f a daa ce.
Bilding Mdel
Whe d e ceae/dae
del ih e aiig
daa? H lg d e hae feaie aiig i ad ceae a del?
Lie Ealain and Mniing
Mehd ad eic ealae he e afe dele, ad aif ale ceai.
Learn
machineleaningcana.cm b Li Dad, Ph.D. Liceed de a Ceaie C Aibi-ShaeAlike 4.0 Ieaial Licee.
The Machine Leaning Cana (0.4) Deiged f: Deiged b: Dae: Ieai: . Domain
Deciin
H ae edici ed
ake decii ha ide
he ed ale he ed-e?
ML ak
I, edic, e f ble.
Vale Piin
Wha ae e ig d f he ed-e() f he edicie e? Wha bjecie ae e eig?
Daa Sce
Which a daa ce ca e e (ieal ad eeal)?
Cllecing Daa
H d e ge e daa lea f (i ad )?
Making Pedicin
Whe d e ake edici e i? H lg d e hae feaie a e i ad ake a edici?
Offline Ealain
Mehd ad eic ealae he e befe dele.
Feae
I eeeai eaced f a daa ce.
Bilding Mdel
Whe d e ceae/dae
del ih e aiig
daa? H lg d e hae feaie aiig i ad ceae a del?
Lie Ealain and Mniing
Mehd ad eic ealae he e afe dele, ad aif ale ceai.
machineleaningcana.cm b Li Dad, Ph.D. Liceed de a Ceaie C Aibi-ShaeAlike 4.0 Ieaial Licee.
The Machine Leaning Cana (0.4) Deiged f: Deciin
H ae edici ed
ake decii ha ide
he ed ale he ed-e?
Deiged b:
Dae:
Ieai: .
Prediction Engine
ML ak
I, edic, e f ble.
Vale Piin
Wha ae e ig d f he ed-e() f he edicie e? Wha bjecie ae e eig?
Daa Sce
Which a daa ce ca e e (ieal ad eeal)?
Cllecing Daa
H d e ge e daa lea f (i ad )?
Making Pedicin
Whe d e ake edici e i? H lg d e hae feaie a e i ad ake a edici?
Offline Ealain
Mehd ad eic ealae he e befe dele.
Feae
I eeeai eaced f a daa ce.
Bilding Mdel
Whe d e ceae/dae
del ih e aiig
daa? H lg d e hae feaie aiig i ad ceae a del?
Lie Ealain and Mniing
Mehd ad eic ealae he e afe dele, ad aif ale ceai.
machineleaningcana.cm b Li Dad, Ph.D.
Liceed de a Ceaie C Aibi-ShaeAlike 4.0 Ieaial Licee.
HOMEWORK
Hands-on Machine Learning
Chapter 2: End-to-End Machine Learning Project
Try reading the Chapter from start to finish.We will work through the problem in class but please come prepared to discuss the case study.
It is easier to understand the different stages of a ML project if you follow one from start to finish.
PIPELINES
CALIFORNIA HOUSING DATASET HMLST – CHAPTER 2
DIS TRIBU TIONS
VISUALIZATION
CALIFORNIA HOUSING PRICES
END- TO-END
— Discover — Explore — Visualize
— Clean
— Sample — Impute — Encode — Transform
– Scale — Features — Pipelines
— Documentation — Presentation
— Launch — Monitor — Maintain
— Training/Validation splits — Modeling
— Tuning
— Error Analysis
ML PIPELINES
Source: https://epistasislab.github.io/tpot/
The Machine Leaning Cana (0.4) Deiged f: Deiged b: Dae: Ieai: .
Deciin
H ae edici ed
ake decii ha ide
he ed ale he ed-e?
ML ak
I, edic, e f ble.
Vale Piin
Wha ae e ig d f he ed-e() f he edicie e? Wha bjecie ae e eig?
Daa Sce
Which a daa ce ca e e (ieal ad eeal)?
Cllecing Daa
H d e ge e daa lea f (i ad )?
Making Pedicin
Whe d e ake edici e i? H lg d e hae feaie a e i ad ake a edici?
Offline Ealain
Mehd ad eic ealae he e befe dele.
Feae
I eeeai eaced f a daa ce.
Bilding Mdel
Whe d e ceae/dae
del ih e aiig
daa? H lg d e hae feaie aiig i ad ceae a del?
Lie Ealain and Mniing
Mehd ad eic ealae he e afe dele, ad aif ale ceai.
machineleaningcana.cm b Li Dad, Ph.D. Liceed de a Ceaie C Aibi-ShaeAlike 4.0 Ieaial Licee.
LECTURE 1 TERM 2:
MSIN0097
Predictive Analytics
A P MOORE