Enables readers to understand mathematical concepts behind data engineering and machine learning algorithms and apply them using open-source Python libraries
Machine Learning Theory and Applications delves into the realm of machine learning and deep learning, exploring their practical applications by comprehending mathematical concepts and implementing them in real-world scenarios using Python and renowned open-source libraries. This comprehensive guide covers a wide range of topics, including data preparation, feature engineering techniques, commonly utilized machine learning algorithms like support vector machines and neural networks, as well as generative AI and foundation models. To facilitate the creation of machine learning pipelines, a dedicated open-source framework named hephAIstos has been developed exclusively for this book. Moreover, the text explores the fascinating domain of quantum machine learning and offers insights on executing machine learning applications across diverse hardware technologies such as CPUs, GPUs, and QPUs. Finally, the book explains how to deploy trained models through containerized applications using Kubernetes and OpenShift, as well as their integration through machine learning operations (MLOps).
Additional topics covered in Machine Learning Theory and Applications include: - Current use cases of AI, including making predictions, recognizing images and speech, performing medical diagnoses, creating intelligent supply chains, natural language processing, and much more - Classical and quantum machine learning algorithms such as quantum-enhanced Support Vector Machines (QSVMs), QSVM multiclass classification, quantum neural networks, and quantum generative adversarial networks (qGANs) - Different ways to manipulate data, such as handling missing data, analyzing categorical data, or processing time-related data - Feature rescaling, extraction, and selection, and how to put your trained models to life and production through containerized applications
Machine Learning Theory and Applications is an essential resource for data scientists, engineers, and IT specialists and architects, as well as students in computer science, mathematics, and bioinformatics. The reader is expected to understand basic Python programming and libraries such as NumPy or Pandas and basic mathematical concepts, especially linear algebra.
Table of Contents
Foreword xiii
Acknowledgments xv
General Introduction xvii
1 Concepts, Libraries, and Essential Tools in Machine Learning and Deep Learning 1
1.1 Learning Styles for Machine Learning 2
1.1.1 Supervised Learning 2
1.1.1.1 Overfitting and Underfitting 3
1.1.1.2 K-Folds Cross-Validation 4
1.1.1.3 Train/Test Split 4
1.1.1.4 Confusion Matrix 5
1.1.1.5 Loss Functions 7
1.1.2 Unsupervised Learning 9
1.1.3 Semi-Supervised Learning 9
1.1.4 Reinforcement Learning 9
1.2 Essential Python Tools for Machine Learning 9
1.2.1 Data Manipulation with Python 10
1.2.2 Python Machine Learning Libraries 10
1.2.2.1 Scikit-learn 10
1.2.2.2 TensorFlow 10
1.2.2.3 Keras 12
1.2.2.4 PyTorch 12
1.2.3 Jupyter Notebook and JupyterLab 13
1.3 HephAIstos for Running Machine Learning on CPUs, GPUs, and QPUs 13
1.3.1 Installation 13
1.3.2 HephAIstos Function 15
1.4 Where to Find the Datasets and Code Examples 32
Further Reading 33
2 Feature Engineering Techniques in Machine Learning 35
2.1 Feature Rescaling: Structured Continuous Numeric Data 36
2.1.1 Data Transformation 37
2.1.1.1 StandardScaler 37
2.1.1.2 MinMaxScaler 39
2.1.1.3 MaxAbsScaler 40
2.1.1.4 RobustScaler 40
2.1.1.5 Normalizer: Unit Vector Normalization 42
2.1.1.6 Other Options 43
2.1.1.7 Transformation to Improve Normal Distribution 44
2.1.1.8 Quantile Transformation 48
2.1.2 Example: Rescaling Applied to an SVM Model 50
2.2 Strategies to Work with Categorical (Discrete) Data 57
2.2.1 Ordinal Encoding 59
2.2.2 One-Hot Encoding 61
2.2.3 Label Encoding 62
2.2.4 Helmert Encoding 63
2.2.5 Binary Encoding 64
2.2.6 Frequency Encoding 65
2.2.7 Mean Encoding 66
2.2.8 Sum Encoding 68
2.2.9 Weight of Evidence Encoding 68
2.2.10 Probability Ratio Encoding 70
2.2.11 Hashing Encoding 71
2.2.12 Backward Difference Encoding 72
2.2.13 Leave-One-Out Encoding 73
2.2.14 James-Stein Encoding 74
2.2.15 M-Estimator Encoding 76
2.2.16 Using HephAIstos to Encode Categorical Data 77
2.3 Time-Related Features Engineering 77
2.3.1 Date-Related Features 79
2.3.2 Lag Variables 79
2.3.3 Rolling Window Feature 82
2.3.4 Expending Window Feature 84
2.3.5 Understanding Time Series Data in Context 85
2.4 Handling Missing Values in Machine Learning 88
2.4.1 Row or Column Removal 89
2.4.2 Statistical Imputation: Mean, Median, and Mode 90
2.4.3 Linear Interpolation 91
2.4.4 Multivariate Imputation by Chained Equation Imputation 92
2.4.5 KNN Imputation 93
2.5 Feature Extraction and Selection 97
2.5.1 Feature Extraction 97
2.5.1.1 Principal Component Analysis 98
2.5.1.2 Independent Component Analysis 102
2.5.1.3 Linear Discriminant Analysis 110
2.5.1.4 Locally Linear Embedding 115
2.5.1.5 The t-Distributed Stochastic Neighbor Embedding Technique 123
2.5.1.6 More Manifold Learning Techniques 125
2.5.1.7 Feature Extraction with HephAIstos 130
2.5.2 Feature Selection 131
2.5.2.1 Filter Methods 132
2.5.2.2 Wrapper Methods 146
2.5.2.3 Embedded Methods 154
2.5.2.4 Feature Importance Using Graphics Processing Units (GPUs) 167
2.5.2.5 Feature Selection Using HephAIstos 168
Further Reading 170
3 Machine Learning Algorithms 175
3.1 Linear Regression 176
3.1.1 The Math 176
3.1.2 Gradient Descent to Optimize the Cost Function 177
3.1.3 Implementation of Linear Regression 182
3.1.3.1 Univariate Linear Regression 182
3.1.3.2 Multiple Linear Regression: Predicting Water Temperature 185
3.2 Logistic Regression 202
3.2.1 Binary Logistic Regression 202
3.2.1.1 Cost Function 203
3.2.1.2 Gradient Descent 204
3.2.2 Multinomial Logistic Regression 204
3.2.3 Multinomial Logistic Regression Applied to Fashion MNIST 204
3.2.3.1 Logistic Regression with scikit-learn 205
3.2.3.2 Logistic Regression with Keras on TensorFlow 208
3.2.4 Binary Logistic Regression with Keras on TensorFlow 210
3.3 Support Vector Machine 211
3.3.1 Linearly Separable Data 212
3.3.2 Not Fully Linearly Separable Data 214
3.3.3 Nonlinear SVMs 216
3.3.4 SVMs for Regression 217
3.3.5 Application of SVMs 219
3.3.5.1 SVM Using scikit-learn for Classification 220
3.3.5.2 SVM Using scikit-learn for Regression 222
3.4 Artificial Neural Networks 223
3.4.1 Multilayer Perceptron 224
3.4.2 Estimation of the Parameters 225
3.4.2.1 Loss Functions 225
3.4.2.2 Backpropagation: Binary Classification 226
3.4.2.3 Backpropagation: Multi-class Classification 227
3.4.3 Convolutional Neural Networks 230
3.4.4 Recurrent Neural Network 232
3.4.5 Application of MLP Neural Networks 233
3.4.6 Application of RNNs: LST Memory 242
3.4.7 Building a CNN 246
3.5 Many More Algorithms to Explore 249
3.6 Unsupervised Machine Learning Algorithms 251
3.6.1 Clustering 251
3.6.1.1 K-means 253
3.6.1.2 Mini-batch K-means 255
3.6.1.3 Mean Shift 257
3.6.1.4 Affinity Propagation 259
3.6.1.5 Density-based Spatial Clustering of Applications with Noise 262
3.7 Machine Learning Algorithms with HephAIstos 264
References 270
Further Reading 270
4 Natural Language Processing 273
4.1 Classifying Messages as Spam or Ham 274
4.2 Sentiment Analysis 281
4.3 Bidirectional Encoder Representations from Transformers 286
4.4 BERT’s Functionality 287
4.5 Installing and Training BERT for Binary Text Classification Using TensorFlow 288
4.6 Utilizing BERT for Text Summarization 294
4.7 Utilizing BERT for Question Answering 296
Further Reading 297
5 Machine Learning Algorithms in Quantum Computing 299
5.1 Quantum Machine Learning 303
5.2 Quantum Kernel Machine Learning 306
5.3 Quantum Kernel Training 328
5.4 Pegasos QSVC: Binary Classification 333
5.5 Quantum Neural Networks 337
5.5.1 Binary Classification with EstimatorQNN 338
5.5.2 Classification with a SamplerQNN 343
5.5.3 Classification with Variational Quantum Classifier 348
5.5.4 Regression 351
5.6 Quantum Generative Adversarial Network 352
5.7 Quantum Algorithms with HephAIstos 368
References 372
Further Reading 373
6 Machine Learning in Production 375
6.1 Why Use Docker Containers for Machine Learning? 375
6.1.1 First Things First: The Microservices 375
6.1.2 Containerization 376
6.1.3 Docker and Machine Learning: Resolving the “It Works in My Machine” Problem 376
6.1.4 Quick Install and First Use of Docker 377
6.1.4.1 Install Docker 377
6.1.4.2 Using Docker from the Command Line 378
6.1.5 Dockerfile 380
6.1.6 Build and Run a Docker Container for Your Machine Learning Model 381
6.2 Machine Learning Prediction in Real Time Using Docker and Python REST APIs with Flask 389
6.2.1 Flask-RESTful APIs 390
6.2.2 Machine Learning Models 392
6.2.3 Docker Image for the Online Inference 393
6.2.4 Running Docker Online Inference 394
6.3 From DevOps to MLOPS: Integrate Machine Learning Models Using Jenkins and Docker 396
6.3.1 Jenkins Installation 397
6.3.2 Scenario Implementation 399
6.4 Machine Learning with Docker and Kubernetes: Install a Cluster from Scratch 405
6.4.1 Kubernetes Vocabulary 405
6.4.2 Kubernetes Quick Install 406
6.4.3 Install a Kubernetes Cluster 407
6.4.4 Kubernetes: Initialization and Internal Network 410
6.5 Machine Learning with Docker and Kubernetes: Training Models 415
6.5.1 Kubernetes Jobs: Model Training and Batch Inference 415
6.5.2 Create and Prepare the Virtual Machines 415
6.5.3 Kubeadm Installation 415
6.5.4 Create a Kubernetes Cluster 416
6.5.5 Containerize our Python Application that Trains Models 418
6.5.6 Create Configuration Files for Kubernetes 422
6.5.7 Commands to Delete the Cluster 424
6.6 Machine Learning with Docker and Kubernetes: Batch Inference 424
6.6.1 Create Configuration Files for Kubernetes 427
6.7 Machine Learning Prediction in Real Time Using Docker, Python Rest APIs with Flask, and Kubernetes: Online Inference 428
6.7.1 Flask-RESTful APIs 428
6.7.2 Machine Learning Models 431
6.7.3 Docker Image for Online Inference 432
6.7.4 Running Docker Online Inference 433
6.7.5 Create and Prepare the Virtual Machines 434
6.7.6 Kubeadm Installation 434
6.7.7 Create a Kubernetes Cluster 435
6.7.8 Deploying the Containerized Machine Learning Model to Kubernetes 437
6.8 A Machine Learning Application that Deploys to the IBM Cloud Kubernetes Service: Python, Docker, Kubernetes 440
6.8.1 Create Kubernetes Service on IBM Cloud 440
6.8.2 Containerization of a Machine Learning Application 443
6.8.3 Push the Image to the IBM Cloud Registry 446
6.8.4 Deploy the Application to Kubernetes 448
6.9 Red Hat OpenShift to Develop and Deploy Enterprise ML/DL Applications 452
6.9.1 What is OpenShift? 453
6.9.2 What Is the Difference Between OpenShift and Kubernetes? 453
6.9.3 Why Red Hat OpenShift for ML/DL? To Build a Production-Ready ML/DL Environment 454
6.10 Deploying a Machine Learning Model as an API on the Red Hat OpenShift Container Platform: From Source Code in a GitHub Repository with Flask, Scikit-Learn, and Docker 454
6.10.1 Create an OpenShift Cluster Instance 455
6.10.1.1 Deploying an Application from Source Code in a GitHub Repository 457
Further Reading 463
Conclusion: The Future of Computing for Data Science? 465
Index 477
