Hands-on Machine Learning with Scikit-Learn and TensorFlow

Cover......Page 1
Copyright......Page 4
Table of Contents......Page 5
 Machine Learning in Your Projects......Page 15
 Objective and Approach......Page 16
 Roadmap......Page 17
 Other Resources......Page 18
 Conventions Used in This Book......Page 19
 O’Reilly Safari......Page 20
 Acknowledgments......Page 21
Part I. The Fundamentals of Machine Learning......Page 23
 Chapter 1. The Machine Learning Landscape......Page 25
  Why Use Machine Learning?......Page 26
  Types of Machine Learning Systems......Page 29
   Supervised/Unsupervised Learning......Page 30
   Batch and Online Learning......Page 36
   Instance-Based Versus Model-Based Learning......Page 39
   Insufficient Quantity of Training Data......Page 44
   Nonrepresentative Training Data......Page 46
   Irrelevant Features......Page 47
   Overfitting the Training Data......Page 48
   Stepping Back......Page 50
  Testing and Validating......Page 51
  Exercises......Page 53
  Working with Real Data......Page 55
   Frame the Problem......Page 57
   Select a Performance Measure......Page 59
   Create the Workspace......Page 62
   Download the Data......Page 65
   Take a Quick Look at the Data Structure......Page 67
   Create a Test Set......Page 71
   Visualizing Geographical Data......Page 75
   Looking for Correlations......Page 78
   Experimenting with Attribute Combinations......Page 81
  Prepare the Data for Machine Learning Algorithms......Page 82
   Data Cleaning......Page 83
   Handling Text and Categorical Attributes......Page 85
   Custom Transformers......Page 87
   Feature Scaling......Page 88
   Transformation Pipelines......Page 89
   Training and Evaluating on the Training Set......Page 91
   Better Evaluation Using Cross-Validation......Page 93
   Grid Search......Page 95
   Randomized Search......Page 97
   Analyze the Best Models and Their Errors......Page 98
   Evaluate Your System on the Test Set......Page 99
  Try It Out!......Page 100
  Exercises......Page 101
  MNIST......Page 103
  Performance Measures......Page 106
   Measuring Accuracy Using Cross-Validation......Page 107
   Confusion Matrix......Page 108
   Precision and Recall......Page 110
   Precision/Recall Tradeoff......Page 111
   The ROC Curve......Page 115
  Multiclass Classification......Page 117
  Error Analysis......Page 120
  Multilabel Classification......Page 124
  Multioutput Classification......Page 125
  Exercises......Page 126
 Chapter 4. Training Models......Page 129
  Linear Regression......Page 130
   The Normal Equation......Page 132
   Computational Complexity......Page 134
  Gradient Descent......Page 135
   Batch Gradient Descent......Page 138
   Stochastic Gradient Descent......Page 141
   Mini-batch Gradient Descent......Page 143
  Polynomial Regression......Page 145
  Learning Curves......Page 147
   Ridge Regression......Page 151
   Lasso Regression......Page 154
   Elastic Net......Page 156
   Early Stopping......Page 157
   Estimating Probabilities......Page 158
   Training and Cost Function......Page 159
   Decision Boundaries......Page 161
   Softmax Regression......Page 163
  Exercises......Page 167
  Linear SVM Classification......Page 169
   Soft Margin Classification......Page 170
  Nonlinear SVM Classification......Page 173
   Polynomial Kernel......Page 174
   Adding Similarity Features......Page 175
   Gaussian RBF Kernel......Page 176
  SVM Regression......Page 178
   Decision Function and Predictions......Page 180
   Training Objective......Page 181
   Quadratic Programming......Page 183
   The Dual Problem......Page 184
   Kernelized SVM......Page 185
   Online SVMs......Page 188
  Exercises......Page 189
  Training and Visualizing a Decision Tree......Page 191
  Making Predictions......Page 193
  The CART Training Algorithm......Page 195
  Gini Impurity or Entropy?......Page 196
  Regularization Hyperparameters......Page 197
  Regression......Page 199
  Instability......Page 201
  Exercises......Page 202
  Voting Classifiers......Page 205
  Bagging and Pasting......Page 209
   Bagging and Pasting in Scikit-Learn......Page 210
   Out-of-Bag Evaluation......Page 211
  Random Patches and Random Subspaces......Page 212
  Random Forests......Page 213
   Feature Importance......Page 214
  Boosting......Page 215
   AdaBoost......Page 216
   Gradient Boosting......Page 219
  Stacking......Page 224
  Exercises......Page 226
 Chapter 8. Dimensionality Reduction......Page 229
  The Curse of Dimensionality......Page 230
   Projection......Page 231
   Manifold Learning......Page 234
   Preserving the Variance......Page 235
   Principal Components......Page 236
   Projecting Down to d Dimensions......Page 237
   Explained Variance Ratio......Page 238
   Choosing the Right Number of Dimensions......Page 239
   PCA for Compression......Page 240
   Incremental PCA......Page 241
  Kernel PCA......Page 242
   Selecting a Kernel and Tuning Hyperparameters......Page 243
  LLE......Page 245
  Other Dimensionality Reduction Techniques......Page 247
  Exercises......Page 248
Part II. Neural Networks and Deep Learning......Page 251
 Chapter 9. Up and Running with TensorFlow......Page 253
  Creating Your First Graph and Running It in a Session......Page 256
  Managing Graphs......Page 258
  Linear Regression with TensorFlow......Page 259
   Manually Computing the Gradients......Page 261
   Using autodiff......Page 262
  Feeding Data to the Training Algorithm......Page 263
  Saving and Restoring Models......Page 265
  Visualizing the Graph and Training Curves Using TensorBoard......Page 266
  Name Scopes......Page 269
  Modularity......Page 270
  Sharing Variables......Page 272
  Exercises......Page 275
 Chapter 10. Introduction to Artificial Neural Networks......Page 277
  From Biological to Artificial Neurons......Page 278
   Biological Neurons......Page 279
   Logical Computations with Neurons......Page 280
   The Perceptron......Page 281
   Multi-Layer Perceptron and Backpropagation......Page 285
  Training an MLP with TensorFlow’s High-Level API......Page 288
   Construction Phase......Page 289
   Execution Phase......Page 293
  Fine-Tuning Neural Network Hyperparameters......Page 294
   Number of Hidden Layers......Page 295
   Activation Functions......Page 296
  Exercises......Page 297
  Vanishing/Exploding Gradients Problems......Page 299
   Xavier and He Initialization......Page 301
   Nonsaturating Activation Functions......Page 303
   Batch Normalization......Page 306
   Gradient Clipping......Page 310
   Reusing a TensorFlow Model......Page 311
   Reusing Models from Other Frameworks......Page 313
   Freezing the Lower Layers......Page 314
   Caching the Frozen Layers......Page 315
   Model Zoos......Page 316
   Unsupervised Pretraining......Page 317
   Pretraining on an Auxiliary Task......Page 318
   Momentum Optimization......Page 319
   Nesterov Accelerated Gradient......Page 321
   AdaGrad......Page 322
   Adam Optimization......Page 324
   Learning Rate Scheduling......Page 327
   ℓ1 and ℓ2 Regularization......Page 329
   Dropout......Page 331
   Max-Norm Regularization......Page 333
   Data Augmentation......Page 335
  Practical Guidelines......Page 336
  Exercises......Page 337
 Chapter 12. Distributing TensorFlow Across Devices and Servers......Page 339
   Installation......Page 340
   Managing the GPU RAM......Page 343
   Placing Operations on Devices......Page 344
   Parallel Execution......Page 347
  Multiple Devices Across Multiple Servers......Page 349
   The Master and Worker Services......Page 351
   Pinning Operations Across Tasks......Page 352
   Sharding Variables Across Multiple Parameter Servers......Page 353
   Sharing State Across Sessions Using Resource Containers......Page 354
   Asynchronous Communication Using TensorFlow Queues......Page 355
   Loading Data Directly from the Graph......Page 361
   One Neural Network per Device......Page 368
   In-Graph Versus Between-Graph Replication......Page 369
   Model Parallelism......Page 371
   Data Parallelism......Page 373
  Exercises......Page 378
 Chapter 13. Convolutional Neural Networks......Page 379
  The Architecture of the Visual Cortex......Page 380
  Convolutional Layer......Page 381
   Filters......Page 383
   Stacking Multiple Feature Maps......Page 384
   TensorFlow Implementation......Page 386
   Memory Requirements......Page 388
  Pooling Layer......Page 389
  CNN Architectures......Page 391
   LeNet-5......Page 392
   AlexNet......Page 393
   GoogLeNet......Page 395
   ResNet......Page 398
  Exercises......Page 402
 Chapter 14. Recurrent Neural Networks......Page 405
  Recurrent Neurons......Page 406
   Memory Cells......Page 408
   Input and Output Sequences......Page 409
  Basic RNNs in TensorFlow......Page 410
   Static Unrolling Through Time......Page 411
   Dynamic Unrolling Through Time......Page 413
   Handling Variable Length Input Sequences......Page 414
  Training RNNs......Page 415
   Training a Sequence Classifier......Page 416
   Training to Predict Time Series......Page 418
   Creative RNN......Page 422
  Deep RNNs......Page 423
   Distributing a Deep RNN Across Multiple GPUs......Page 424
   Applying Dropout......Page 425
   The Difficulty of Training over Many Time Steps......Page 426
  LSTM Cell......Page 427
  GRU Cell......Page 430
   Word Embeddings......Page 432
   An Encoder–Decoder Network for Machine Translation......Page 434
  Exercises......Page 437
 Chapter 15. Autoencoders......Page 439
  Efficient Data Representations......Page 440
  Performing PCA with an Undercomplete Linear Autoencoder......Page 441
  Stacked Autoencoders......Page 443
   TensorFlow Implementation......Page 444
   Tying Weights......Page 445
   Training One Autoencoder at a Time......Page 446
   Visualizing Features......Page 449
  Unsupervised Pretraining Using Stacked Autoencoders......Page 450
  Denoising Autoencoders......Page 452
   TensorFlow Implementation......Page 453
  Sparse Autoencoders......Page 454
   TensorFlow Implementation......Page 456
  Variational Autoencoders......Page 457
   Generating Digits......Page 460
  Other Autoencoders......Page 461
  Exercises......Page 462
 Chapter 16. Reinforcement Learning......Page 465
  Learning to Optimize Rewards......Page 466
  Policy Search......Page 468
  Introduction to OpenAI Gym......Page 469
  Neural Network Policies......Page 473
  Evaluating Actions: The Credit Assignment Problem......Page 475
  Policy Gradients......Page 476
  Markov Decision Processes......Page 481
  Temporal Difference Learning and Q-Learning......Page 485
   Exploration Policies......Page 487
   Approximate Q-Learning and Deep Q-Learning......Page 488
  Learning to Play Ms. Pac-Man Using the DQN Algorithm......Page 489
  Exercises......Page 497
  Thank You!......Page 498
 Chapter 1: The Machine Learning Landscape......Page 499
 Chapter 4: Training Models......Page 501
 Chapter 5: Support Vector Machines......Page 503
 Chapter 6: Decision Trees......Page 505
 Chapter 7: Ensemble Learning and Random Forests......Page 506
 Chapter 8: Dimensionality Reduction......Page 507
 Chapter 9: Up and Running with TensorFlow......Page 509
 Chapter 10: Introduction to Artificial Neural Networks......Page 511
 Chapter 11: Training Deep Neural Nets......Page 513
 Chapter 12: Distributing TensorFlow Across Devices and Servers......Page 515
 Chapter 13: Convolutional Neural Networks......Page 516
 Chapter 14: Recurrent Neural Networks......Page 519
 Chapter 15: Autoencoders......Page 520
 Chapter 16: Reinforcement Learning......Page 522
 Frame the Problem and Look at the Big Picture......Page 525
 Explore the Data......Page 526
 Prepare the Data......Page 527
 Fine-Tune the System......Page 528
 Present Your Solution......Page 529
 Launch!......Page 530
Appendix C. SVM Dual Problem......Page 531
 Manual Differentiation......Page 535
 Symbolic Differentiation......Page 536
 Numerical Differentiation......Page 537
 Forward-Mode Autodiff......Page 538
 Reverse-Mode Autodiff......Page 540
 Hopfield Networks......Page 543
 Boltzmann Machines......Page 544
 Restricted Boltzmann Machines......Page 546
 Deep Belief Nets......Page 547
 Self-Organizing Maps......Page 549
Index......Page 553
Colophon......Page 570