Preface

15

Target Group

15

Requirements

15

Structure of the Book

16

How to Use This Book

18

Downloading Code and Additional Materials

18

Preparing the System

19

Installing Python

19

Installing an Integrated Development Environment

20

Installing Git

21

Downloading Course Material

21

Setting Up the Local Python Environment

22

Acknowledgements

24

Conventions Used in This Book

25

1 Introduction to Deep Learning

27

1.1 What is Deep Learning?

27

1.2 What Can You Use Deep Learning For?

28

1.3 How Does Deep Learning Work?

31

1.3.1 Model Training

31

1.3.2 Model Inference

32

1.4 Historical Development

33

1.5 Perceptrons

34

1.6 Network Structure and Layers

34

1.7 Activation Functions

35

1.7.1 Rectified Linear Units

36

1.7.2 Hyperbolic Tangents

37

1.7.3 Sigmoid

37

1.7.4 Softmax

37

1.8 Loss Functions

38

1.8.1 Regression Problems

39

1.8.2 Binary Classification

39

1.8.3 Multiclass Classification

40

1.9 Optimizers and Updating Parameters

40

1.9.1 Optimizers

40

1.9.2 Updating Parameters

41

1.10 Tensor Handling

42

1.10.1 What are Tensors?

43

1.10.2 Coding: Tensor Creation and Attributes

43

1.10.3 Coding: The Calculation Graph and Training

45

1.11 Summary

50

2 Creating Your First PyTorch Model

51

2.1 Data Preparation

51

2.1.1 Feature Types

52

2.1.2 Data Types

54

2.1.3 One-Hot Encoding

54

2.1.4 Exploratory Data Analysis

56

2.1.5 Data Scaling

59

2.2 Model Creation

60

2.2.1 Data Import

61

2.2.2 Model Training

62

2.2.3 Model Evaluation

64

2.2.4 Model Inference

65

2.3 The Model Class and the Optimizer

68

2.4 Batches

72

2.4.1 What Are Batches?

72

2.4.2 Advantages of Using Batches

72

2.4.3 Optimal Batch Size

73

2.4.4 Coding: Implementation of Batches

74

2.5 Coding: Implementation of Dataset and DataLoader

76

2.6 Loading and Saving a Model

80

2.6.1 Saving Model Parameters

80

2.6.2 Loading a Model

81

2.7 Data Sampling

83

2.7.1 What Is Data Sampling?

83

2.7.2 Cross-Validation

85

2.7.3 Why Is Data Sampling Needed?

85

2.7.4 Coding: Separation of Training and Validation Data

86

2.8 Summary

92

3 Classification Models

93

3.1 Classification Types

93

3.2 Confusion Matrix

95

3.3 Receiver Operator Characteristic Curve

97

3.4 Coding: Binary Classification

99

3.4.1 Data Preparation

99

3.4.2 Modeling

104

3.4.3 Evaluation

108

3.5 Coding: Multiclass Classification

112

3.5.1 Data Preparation

113

3.5.2 Modeling

117

3.5.3 Evaluation

121

3.6 Summary

124

4 Computer Vision

127

4.1 How Do Models Handle Images?

128

4.2 Network Architecture

129

4.2.1 Convolutional Neural Networks

130

4.2.2 Vision Transformers

132

4.3 Coding: Image Classification

134

4.3.1 Binary Classification

135

4.3.2 Multiclass Classification

151

4.4 Object Detection

163

4.4.1 How Does Object Detection Work?

164

4.4.2 Data Preparation

165

4.4.3 Model Training

171

4.4.4 Model Evaluation

172

4.4.5 Model Inference

175

4.5 Semantic Segmentation

178

4.5.1 How Does Semantic Segmentation Work?

179

4.5.2 Segmentation Masks

180

4.5.3 Data Preparation

181

4.5.4 Model Training

182

4.5.5 Model Evaluation

186

4.6 Style Transfer

188

4.6.1 How Does Style Transfer Work?

189

4.6.2 Data Preparation

190

4.6.3 Model Training

193

4.6.4 Model Evaluation

196

4.7 Summary

197

5 Recommendation Systems

199

5.1 Theoretical Foundations

199

5.1.1 Basic Concepts

199

5.1.2 Matrix Factorization

201

5.2 Coding: Recommendation Systems

202

5.2.1 Data Preparation

202

5.2.2 Modeling

206

5.2.3 Model Evaluation

209

5.3 Summary

218

6 Autoencoders

219

6.1 Architecture

220

6.2 Coding: Autoencoder

220

6.2.1 Data Preparation

221

6.2.2 Modeling

223

6.2.3 Model Training

226

6.2.4 Model Validation

228

6.3 Variational Autoencoders

230

6.4 Coding: Variational Autoencoder

231

6.4.1 Network Architecture

231

6.4.2 Loss Function

235

6.5 Summary

240

7 Graph Neural Networks

241

7.1 Introduction to Graph Theory

241

7.1.1 Graphs and the Adjacency Matrix

242

7.1.2 Features

243

7.1.3 Passing Messages

244

7.1.4 Use Cases

245

7.2 Coding: Developing a Graph

246

7.3 Coding: Training a Graph Neural Network

250

7.4 Summary

259

8 Time Series Forecasting

261

8.1 Modeling Approaches

261

8.1.1 Special Features of Time Series Models

261

8.1.2 Data Modeling

262

8.1.3 Long Short-Term Memory

263

8.1.4 One-Dimensional Convolutional Neural Networks

264

8.1.5 Transformer

265

8.2 Coding: Custom Model

266

8.2.1 Data Preparation

266

8.2.2 Long Short-Term Memory

271

8.2.3 Convolutional Neural Network

274

8.2.4 Transformers

277

8.3 Coding: Using PyTorch Forecasting

280

8.3.1 Data Preparation

280

8.3.2 Model Training

284

8.3.3 Evaluation

286

8.4 Summary

288

9 Language Models

289

9.1 Using Large Language Models with Python

290

9.1.1 Coding: Using OpenAI

291

9.1.2 Coding: Using Groq

294

9.1.3 Coding: Multimodal Models

297

9.1.4 Coding: Using Large Language Models Locally

300

9.2 Model Parameters

304

9.2.1 Model Temperature

304

9.2.2 Top-p and Top-k

305

9.2.3 Best Practices

307

9.3 Model Selection

307

9.3.1 Performance

307

9.3.2 Cutoff Date

308

9.3.3 On-Premise Versus Cloud Hosting

309

9.3.4 Open-Source, Open-Weight, and Proprietary Models

309

9.3.5 Cost

309

9.3.6 Context Window

310

9.3.7 Latency

310

9.4 Message Types

310

9.4.1 User/Human Messages

310

9.4.2 System Messages

311

9.4.3 Assistant Messages

311

9.5 Prompt Templates

311

9.5.1 Coding: ChatPromptTemplates

312

9.5.2 Coding: Improving a Prompt with LangChain Hub

313

9.6 Chains

315

9.6.1 A Simple Sequential Chain

315

9.6.2 Coding: A Simple Sequential Chain

316

9.7 Structured Outputs

317

9.7.1 What Are Structured Outputs?

317

9.7.2 Coding: Structured Outputs

318

9.8 Deep Dive: How Do Transformers Work?

320

9.8.1 Tokenization

321

9.8.2 Word Embeddings

323

9.8.3 Positional Encoding

324

9.8.4 Attention

325

9.9 Summary

327

10 Pretrained Networks and Fine-Tuning

329

10.1 Pretrained Networks with Hugging Face

329

10.2 Transfer Learning

332

10.2.1 Advantages of Transfer Learning

332

10.2.2 Transfer Learning Approaches

334

10.3 Coding: Fine-Tuning a Computer Vision Model

335

10.3.1 Data Preparation

336

10.3.2 Model Training

339

10.3.3 Model Evaluation

341

10.4 Coding: Fine-Tuning a Language Model

343

10.4.1 Data Preparation

344

10.4.2 Model Training

346

10.4.3 Model Evaluation

348

10.5 Summary

348

11 PyTorch Lightning

351

11.1 PyTorch Versus PyTorch Lightning

351

11.2 Coding: Model Training

352

11.3 Callbacks

359

11.3.1 Model Checkpoints

359

11.3.2 Early Stopping

361

11.4 Summary

362

12 Model Evaluation, Logging, and Monitoring

363

12.1 TensorBoard

363

12.1.1 How It Works

364

12.1.2 Using TensorBoard

364

12.1.3 TensorBoard Dashboard

370

12.2 MLflow

372

12.2.1 Data Logging

373

12.2.2 Saving and Loading the Model

375

12.2.3 MLflow Dashboard

376

12.3 Weights & Biases: WandB

377

12.3.1 Initialization

378

12.3.2 Logging Metrics

379

12.3.3 Logging Artifacts

380

12.3.4 Sweeps

382

12.4 Summary

384

13 Deployment

385

13.1 Deployment Strategies

385

13.2 Local Deployment

387

13.2.1 API Development

388

13.2.2 Deployment

391

13.2.3 Test

391

13.3 Heroku

393

13.3.1 Command Line Interface and Login

393

13.3.2 Deployment

394

13.3.3 Test

397

13.3.4 Stopping and Deleting the App

398

13.4 Microsoft Azure

399

13.4.1 First Steps

399

13.4.2 App Development

401

13.4.3 Local Test

403

13.4.4 Function App Development in the Azure Portal

404

13.4.5 Cloud Deployment

406

13.5 Summary

407