Paperback
-
PICK UP IN STORECheck Availability at Nearby Stores
Available within 2 business hours
Related collections and offers
Overview
If you’ve been curious about artificial intelligence and machine learning but didn’t know where to start, this is the book you’ve been waiting for. Focusing on the subfield of machine learning known as deep learning, it explains core concepts and gives you the foundation you need to start building your own models. Rather than simply outlining recipes for using existing toolkits, Practical Deep Learning teaches you the why of deep learning and will inspire you to explore further.
All you need is basic familiarity with computer programming and high school math—the book will cover the rest. After an introduction to Python, you’ll move through key topics like how to build a good training dataset, work with the scikit-learn and Keras libraries, and evaluate your models’ performance.
You’ll also learn:
- How to use classic machine learning models like k-Nearest Neighbors, Random Forests, and Support Vector Machines
- How neural networks work and how they’re trained
- How to use convolutional neural networks
- How to develop a successful deep learning model from scratch
The perfect introduction to this dynamic, ever-expanding field, Practical Deep Learning will give you the skills and confidence to dive into your own machine learning projects.
Product Details
| ISBN-13: | 9781718500747 |
|---|---|
| Publisher: | No Starch Press |
| Publication date: | 02/23/2021 |
| Pages: | 464 |
| Sales rank: | 613,142 |
| Product dimensions: | 7.00(w) x 9.50(h) x 1.30(d) |
About the Author
Table of Contents
Foreword xix
Acknowledgments xxiii
Introduction xxv
Who Is This Book For? xxvi
What Can You Expect to Learn? xxvii
About This Book xxvii
1 Getting started 1
The Operating Environment 1
NumPy 2
Scikit-learn 2
Keras with TensorFlow 2
Installing the Toolkits 3
Basic Linear Algebra 4
Vectors 4
Matrices 5
Multiplying Vectors and Matrices 5
Statistics and Probability 6
Descriptive Statistics 6
Probability Distributions 7
Statistical Tests 8
Graphics Processing Units 9
Summary 9
2 Using python 11
The Python Interpreter 11
Statements and Whitespace 12
Variables and Basic Data Structures 13
Representing Numbers 13
Variables 14
Strings 14
Lists 15
Dictionaries 18
Control Structures 19
If-elif-else Statements 19
For Loops 19
While Loops 22
Break and continue Statements 22
With Statement 23
Handling Errors with try-except Blocks 24
Functions 24
Modules 26
Summary 27
3 Using NumPy 29
Why NumPy? 29
Arrays vs. Lists 30
Testing Array and List Speed 31
Basic Arrays 33
Defining an Array with np.array 33
Defining Arrays with 0s and Is 36
Accessing Elements in an Array 37
Indexing into an Array 37
Slicing an Array 39
The Ellipsis 41
Operators and Broadcasting 42
Array Input and Output 45
Random Numbers 48
NumPyand Images 48
Summary 50
4 Working with data 51
Classes and Labels 51
Features and Feature Vectors 51
Types of Features 53
Feature Selection and the Curse of Dimensionality 55
Features of a Good Dataset 57
Interpolation and Extrapolation 58
The Parent Distribution 60
Prior Class Probabilities 60
Confuses 61
Dataset Size 62
Data Preparation 63
Scaling Features 63
Missing Features 67
Training, Validation, and Test Data 68
The Three Subsets 68
Partitioning the Dataset 69
k-Fold Cross Validation 74
Look at Your Data 76
Searching for Problems in the Data 76
Cautionary Tales 80
Summary 81
5 Building datasets 83
Irises 84
Breast Cancer 86
MNIST Digits 88
CIFAR-10 90
Data Augmentation 92
Why Should You Augment Training Data? 93
Ways to Augment Training Data 94
Augmenting the Iris Dataset 95
Augmenting the CIFAR-10 Dataset 101
Summary 105
6 Classical machine learning 107
Nearest Centroid 108
k-Nearest Neighbors 112
Naïve Bayes 113
Decision Trees and Random Forests 117
Recursion Primer 120
Building Decision Trees 121
Random Forests 122
Support Vector Machines 124
Margins 124
Support Vectors 126
Optimization 126
Kernels 127
Summary 128
7 Experiments with classical models 129
Experiments with the Iris Dataset 129
Testing the Classical Models 130
Implementing a Nearest Centroid Classifier 133
Experiments with the Breast Cancer Dataset 135
Two Initial Test Runs 135
The Effect of Random Splits 138
Adding k-fold Validation 140
Searching for Hyperparameters 145
Experiments with the MNIST Dataset 150
Testing the Classical Models 150
Analyzing Runtimes 156
Experimenting with PCA Components 158
Scrambling Our Dataset 161
Classical Model Summary 162
Nearest Centroid 162
k-Nearest Neighbors 163
Naïve Bayes 163
Decision Trees 164
Random Forests 164
Support Vector Machines 165
When to Use Classical Models 165
Handling Small Datasets 165
Dealing with Reduced Computational Requirements 165
Having Explainable Models 166
Working with Vector Inputs 166
Summary 167
8 Introduction to neural networks 169
Anatomy of a Neural Network 170
The Neuron 171
Activation Functions 172
Architecture of a Network 176
Output Layers 178
Representing Weights and Biases 180
Implementing a Simple Neural Network 182
Building the Dataset 182
Implementing the Neural Network 183
Training and Testing the Neural Network 185
Summary 188
9 Training a neural network 189
A High-Level Overview 190
Gradient Descent 190
Finding Minimums 192
Updating the Weights 193
Stochastic Gradient Descent 194
Batches and Minibatches 195
Convex vs. Nonconvex Functions 196
Ending Training 197
Updating the Learning Rate 198
Momentum 199
Backpropagafion 200
Backprop, Take 1 200
Backprop, Take 2 204
Loss Functions 208
Absolute and Mean Squared Error Loss 209
Cross-Entropy Loss 210
Weight Initialization 211
Overfilling and Regularization 213
Understanding Overfilling 213
Understanding Regularization 215
L2 Regularization 216
Dropout 217
Summary 219
10 Experiments with neural networks 221
Our Dataset 222
The MLPClassifier Class 222
Architecture and Activation Functions 223
The Code 223
The Results 227
Batch Size 231
Base Learning Rate 235
Training Set Size 238
L2 Regularization 239
Momentum 242
Weight Initialization 243
Feature Ordering 247
Summary 249
11 Evaluating models 251
Definitions and Assumptions 251
Why Accuracy Is Not Enough 252
The 2 × 2 Confusion Matrix 254
Metrics Derived from the 2 × 2 Confusion Matrix 257
Deriving Metrics from the 2 × 2 Table 257
Using Our Metrics to Interpret Models 260
More Advanced Metrics 262
Informedness and Markedness 262
F1 Score 263
Cohen's Kappa 263
Matthews Correlation Coefficient 264
Implementing Our Metrics 264
The Receiver Operating Characteristics Curve 266
Gathering Our Models 266
Plotting Our Metrics 268
Exploring the ROC Curve 269
Comparing Models with ROC Analysis 271
Generating an ROC Curve 273
The Precision-Recall Curve 275
Handling Multiple Classes 276
Extending the Confusion Matrix 276
Calculating Weighted Accuracy 279
Multiclass Matthews Correlation Coefficient 281
Summary 282
12 Introduction to convolutional neural networks 283
Why Convolutional Neural Networks? 284
Convolution 284
Scanning with the Kernel 285
Convolution for Image Processing 287
Anatomy of a Convolutional Neural Network 288
Different Types of Layers 289
Passing Data Through the CNN 291
Convolutional Layers 292
How a Convolution Layer Works 292
Using a Convolutional Layer 295
Multiple Convolutional Layers 298
Initializing a Convolutional Layer 299
Pooling Layers 299
Fully Connected Layers 301
Fully Convolutional Layers 302
Step by Step 304
Summary 308
13 Experiments with keras and MNIST 309
Building CNNs in Keras 310
Loading the MNIST Data 310
Building Our Model 312
Training and Evaluating the Model 314
Plotting the Error 317
Basic Experiments 319
Architecture Experiments 319
Training Set Size, Minibatches, and Epochs 323
Optimizers 326
Fully Convolutional Networks 328
Building and Training the Model 328
Making the Test Images 331
Testing the Model 333
Scrambled MNIST Digits 340
Summary 342
14 Experiments with CIFAR-10 343
A CIFAR-10 Refresher 343
Working with the Full CIFAR-10 Dataset 344
Building the Models 345
Analyzing the Models 348
Animal or Vehicle? 352
Binary or Multiclass? 357
Transfer Learning 361
Fine-Tuning a Model 367
Building Our Datasets 368
Adapting Our Model for Fine-Tuning 371
Testing Our Model 373
Summary 375
15 A case study: Classifying Audio Samples 377
Building the Dataset 378
Augmenting the Dataset 379
Preprocessing Our Data 383
Classifying the Audio Features 385
Using Classical Models 385
Using a Traditional Neural Network 388
Using a Convolutional Neural Network 389
Spectrograms 394
Classifying Spectrograms 398
Initialization, Regularization, and Batch Normalization 402
Examining the Confusion Matrix 403
Ensembles 404
Summary 408
16 Going further 411
Going Further with CNNs 411
Reinforcement Learning and Unsupervised Learning 412
Generative Adversarial Networks 413
Recurrent Neural Networks 414
Online Resources 414
Conferences 415
The Book 416
So Long and Thanks for All the Fish 416
Index 417