Table of Contents

Copyright

Brief Table of Contents

Table of Contents

Foreword

Preface

Acknowledgments

About this Book

Chapter 1. Building applications for the intelligent web

1.1. An intelligent algorithm in action: Google Now

1.2. The intelligent-algorithm lifecycle

1.3. Further examples of intelligent algorithms

1.4. Things that intelligent applications are not

1.4.1. Intelligent algorithms are not all-purpose thinking machines

1.4.2. Intelligent algorithms are not a drop-in replacement for humans

1.4.3. Intelligent algorithms are not discovered by accident

1.5. Classes of intelligent algorithm

1.5.1. Artificial intelligence

1.5.2. Machine learning

1.5.3. Predictive analytics

1.6. Evaluating the performance of intelligent algorithms

1.6.1. Evaluating intelligence

1.6.2. Evaluating predictions

1.7. Important notes about intelligent algorithms

1.7.1. Your data is not reliable

1.7.2. Inference does not happen instantaneously

1.7.3. Size matters!

1.7.4. Different algorithms have different scaling characteristics

1.7.5. Everything is not a nail!

1.7.6. Data isn’t everything

1.7.7. Training time can be variable

1.7.8. Generalization is the goal

1.7.9. Human intuition is problematic

1.7.10. Think about engineering new features

1.7.11. Learn many different models

1.7.12. Correlation is not the same as causation

1.8. Summary

Chapter 2. Extracting structure from data: clustering and transforming your data

2.1. Data, structure, bias, and noise

2.2. The curse of dimensionality

2.3. K-means

2.3.1. K-means in action

2.4. The Gaussian mixture model

2.4.1. What is the Gaussian distribution?

2.4.2. Expectation maximization and the Gaussian distribution

2.4.3. The Gaussian mixture model

2.4.4. An example of learning using a Gaussian mixture model

2.5. The relationship between k-means and GMM

2.6. Transforming the data axis

2.6.1. Eigenvectors and eigenvalues

2.6.2. Principal component analysis

2.6.3. An example of principal component analysis

2.7. Summary

Chapter 3. Recommending relevant content

3.1. Setting the scene: an online movie store

3.2. Distance and similarity

3.2.1. A closer look at distance and similarity

3.2.2. Which is the best similarity formula?

3.3. How do recommendation engines work?

3.4. User-based collaborative filtering

3.5. Model-based recommendation using singular value decomposition

3.5.1. Singular value decomposition

3.5.2. Recommendation using SVD: choosing movies for a given user

3.5.3. Recommendation using SVD: choosing users for a given movie

3.6. The Netflix Prize

3.7. Evaluating your recommender

3.8. Summary

Chapter 4. Classification: placing things where they belong

4.1. The need for classification

4.2. An overview of classifiers

4.2.1. Structural classification algorithms

4.2.2. Statistical classification algorithms

4.2.3. The lifecycle of a classifier

4.3. Fraud detection with logistic regression

4.3.1. A linear-regression primer

4.3.2. From linear to logistic regression

4.3.3. Implementing fraud detection

4.4. Are your results credible?

4.5. Classification with very large datasets

4.6. Summary

Chapter 5. Case study: click prediction for online advertising

5.1. History and background

5.2. The exchange

5.2.1. Cookie matching

5.2.2. Bid

5.2.3. Bid win (or loss) notification

5.2.4. Ad placement

5.2.5. Ad monitoring

5.3. What is a bidder?

5.3.1. Requirements of a bidder

5.4. What is a decisioning engine?

5.4.1. Information about the user

5.4.2. Information about the placement

5.4.3. Contextual information

5.4.4. Data preparation

5.4.5. Decisioning engine model

5.4.6. Mapping predicted click-through rate to bid price

5.4.7. Feature engineering

5.4.8. Model training

5.5. Click prediction with Vowpal Wabbit

5.5.1. Vowpal Wabbit data format

5.5.2. Preparing the dataset

5.5.3. Testing the model

5.5.4. Model calibration

5.6. Complexities of building a decisioning engine

5.7. The future of real-time prediction

5.8. Summary

Chapter 6. Deep learning and neural networks

6.1. An intuitive approach to deep learning

6.2. Neural networks

6.3. The perceptron

6.3.1. Training

6.3.2. Training a perceptron in scikit-learn

6.3.3. A geometric interpretation of the perceptron for two inputs

6.4. Multilayer perceptrons

6.4.1. Training using backpropagation

6.4.2. Activation functions

6.4.3. Intuition behind backpropagation

6.4.4. Backpropagation theory

6.4.5. MLNN in scikit-learn

6.4.6. A learned MLP

6.5. Going deeper: from multilayer neural networks to deep learning

6.5.1. Restricted Boltzmann Machines

6.5.2. The Bernoulli Restricted Boltzmann Machine

6.5.3. RBMs in action

6.6. Summary

Chapter 7. Making the right choice

7.1. A/B testing

7.1.1. The theory

7.1.2. The code

7.1.3. Suitability of A/B

7.2. Multi-armed bandits

7.2.1. Multi-armed bandit strategies

7.3. Bayesian bandits in the wild

Factors impacting the Bayesian bandit

7.4. A/B vs. the Bayesian bandit

7.5. Extensions to multi-armed bandits

7.5.1. Contextual bandits

7.5.2. Adversarial bandits

7.6. Summary

Chapter 8. The future of the intelligent web

8.1. Future applications of the intelligent web

8.1.1. The internet of things

8.1.2. Home healthcare

8.1.3. The self-driving vehicle

8.1.4. Personalized physical advertising

8.1.5. The semantic web

8.2. Social implications of the intelligent web

Capturing data on the web

A motivating example: showing ads online

Data available for online advertising

Data collection: a naïve approach

Managing data collection at scale

Introducing Kafka

Replication in Kafka

Consumer groups, balancing, and ordering

Putting it all together

Evaluating Kafka: data collection at scale

Kafka design patterns

Kafka plus Storm

Kafka plus Hadoop

Index

List of Figures

List of Tables

List of Listings