Hands-On Unsupervised Learning Using Python. How to Build Applied Machine Learning Solutions from Unlabeled Data (e-book) Katowice

Many industry experts consider unsupervised learning the next frontier in artificial intelligence, one that may hold the key to general artificial intelligence. Since the majority of the world's data is unlabeled, conventional supervised learning cannot be applied. Unsupervised learning, on the other hand, can be applied to unlabeled datasets to discover meaningful patterns buried deep in the data, patterns that may be near impossible for humans to uncover.Author Ankur Patel shows you how to apply unsupervised learning using two simple, production-ready Python frameworks: Scikit-learn and TensorFlow using Keras. With code and hands-on examples, data scientists will identify difficult-to-find patterns in data and gain deeper business insight, detect anomalies, perform automatic feature engineering and selection, and generate synthetic datasets. All you need is programming and some machine learning experience to get started.Compare the strengths and weaknesses of the different machine learning approaches: supervised, unsupervised, and reinforcement learningSet up and manage machine learning projects end-to-endBuild an anomaly detection system to catch credit card fraudClusters users into distinct and homogeneous groupsPerform semisupervised learningDevelop movie recommender systems using restricted Boltzmann machinesGenerate synthetic images using generative adversarial networks Spis treści: Preface A Brief History of Machine Learning AI Is Back, but Why Now? The Emergence of Applied AI Major Milestones in Applied AI over the Past 20 Years From Narrow AI to AGI Objective and Approach Prerequisites Roadmap Conventions Used in This Book Using Code Examples OReilly Online Learning How to Contact Us I. Fundamentals of Unsupervised Learning 1. Unsupervised Learning in the Machine Learning Ecosystem Basic Machine Learning Terminology Rules-Based vs. Machine Learning Supervised vs. Unsupervised The Strengths and Weaknesses of Supervised Learning The Strengths and Weaknesses of Unsupervised Learning Using Unsupervised Learning to Improve Machine Learning Solutions Insufficient labeled data Overfitting Curse of dimensionality Feature engineering Outliers Data drift A Closer Look at Supervised Algorithms Linear Methods Linear regression Logistic regression Neighborhood-Based Methods k-nearest neighbors Tree-Based Methods Single decision tree Bagging Random forests Boosting Support Vector Machines Neural Networks A Closer Look at Unsupervised Algorithms Dimensionality Reduction Linear projection Principal component analysis (PCA) Singular value decomposition (SVD) Random projection Manifold learning Isomap t-distributed stochastic neighbor embedding (t-SNE) Dictionary learning Independent component analysis Latent Dirichlet allocation Clustering k-means Hierarchical clustering DBSCAN Feature Extraction Autoencoders Feature extraction using supervised training of feedforward networks Unsupervised Deep Learning Unsupervised pretraining Restricted Boltzmann machines Deep belief networks Generative adversarial networks Sequential Data Problems Using Unsupervised Learning Reinforcement Learning Using Unsupervised Learning Semisupervised Learning Successful Applications of Unsupervised Learning Anomaly Detection Group segmentation Conclusion 2. End-to-End Machine Learning Project Environment Setup Version Control: Git Clone the Hands-On Unsupervised Learning Git Repository Scientific Libraries: Anaconda Distribution of Python Neural Networks: TensorFlow and Keras Gradient Boosting, Version One: XGBoost Gradient Boosting, Version Two: LightGBM Clustering Algorithms Interactive Computing Environment: Jupyter Notebook Overview of the Data Data Preparation Data Acquisition Download the data Import the necessary libraries Read the data Preview the data Data Exploration Generate summary statistics Identify nonnumerical values by feature Identify distinct values by feature Generate Feature Matrix and Labels Array Create the feature matrix X and the labels array Y Standardize the feature matrix X Feature Engineering and Feature Selection Check correlation of features Data Visualization Model Preparation Split into Training and Test Sets Select Cost Function Create k-Fold Cross-Validation Sets Machine Learning Models (Part I) Model #1: Logistic Regression Set hyperparameters Train the model Evaluate the results Evaluation Metrics Confusion Matrix Precision-Recall Curve Receiver Operating Characteristic Evaluating the logistic regression model Machine Learning Models (Part II) Model #2: Random Forests Set the hyperparameters Train the model Evaluate the results Model #3: Gradient Boosting Machine (XGBoost) Set the hyperparameters Train the model Evaluate the results Model #4: Gradient Boosting Machine (LightGBM) Set the hyperparameters Train the model Evaluate the results Evaluation of the Four Models Using the Test Set Logistic regression Random forests XGBoost gradient boosting LightGBM gradient boosting Ensembles Stacking Combine layer one predictions with the original training dataset Set the hyperparameters Train the model Evaluate the results Final Model Selection Production Pipeline Conclusion II. Unsupervised Learning Using Scikit-Learn 3. Dimensionality Reduction The Motivation for Dimensionality Reduction The MNIST Digits Database Data acquisition and exploration Load the MNIST datasets Verify shape of datasets Create Pandas DataFrames from the datasets Explore the data Display the images Dimensionality Reduction Algorithms Linear Projection vs. Manifold Learning Principal Component Analysis PCA, the Concept PCA in Practice Set the hyperparameters Apply PCA Evaluate PCA Visualize the separation of points in space Incremental PCA Sparse PCA Kernel PCA Singular Value Decomposition Random Projection Gaussian Random Projection Sparse Random Projection Isomap Multidimensional Scaling Locally Linear Embedding t-Distributed Stochastic Neighbor Embedding Other Dimensionality Reduction Methods Dictionary Learning Independent Component Analysis Conclusion 4. Anomaly Detection Credit Card Fraud Detection Prepare the Data Define Anomaly Score Function Define Evaluation Metrics Define Plotting Function Normal PCA Anomaly Detection PCA Components Equal Number of Original Dimensions Search for the Optimal Number of Principal Components Sparse PCA Anomaly Detection Kernel PCA Anomaly Detection Gaussian Random Projection Anomaly Detection Sparse Random Projection Anomaly Detection Nonlinear Anomaly Detection Dictionary Learning Anomaly Detection ICA Anomaly Detection Fraud Detection on the Test Set Normal PCA Anomaly Detection on the Test Set ICA Anomaly Detection on the Test Set Dictionary Learning Anomaly Detection on the Test Set Conclusion 5. Clustering MNIST Digits Dataset Data Preparation Clustering Algorithms k-Means k-Means Inertia Evaluating the Clustering Results k-Means Accuracy k-Means and the Number of Principal Components k-Means on the Original Dataset Hierarchical Clustering Agglomerative Hierarchical Clustering The Dendrogram Evaluating the Clustering Results DBSCAN DBSCAN Algorithm Applying DBSCAN to Our Dataset HDBSCAN Conclusion 6. Group Segmentation Lending Club Data Data Preparation Load libraries Explore the data Transform String Format to Numerical Format Impute Missing Values Engineer Features Select Final Set of Features and Perform Scaling Designate Labels for Evaluation Goodness of the Clusters k-Means Application Hierarchical Clustering Application HDBSCAN Application Conclusion III. Unsupervised Learning Using TensorFlow and Keras 7. Autoencoders Neural Networks TensorFlow TensorFlow example Keras Autoencoder: The Encoder and the Decoder Undercomplete Autoencoders Overcomplete Autoencoders Dense vs. Sparse Autoencoders Denoising Autoencoder Variational Autoencoder Conclusion 8. Hands-On Autoencoder Data Preparation The Components of an Autoencoder Activation Functions Our First Autoencoder Loss Function Optimizer Training the Model Evaluating on the Test Set Two-Layer Undercomplete Autoencoder with Linear Activation Function Increasing the Number of Nodes Adding More Hidden Layers Nonlinear Autoencoder Overcomplete Autoencoder with Linear Activation Overcomplete Autoencoder with Linear Activation and Dropout Sparse Overcomplete Autoencoder with Linear Activation Sparse Overcomplete Autoencoder with Linear Activation and Dropout Working with Noisy Datasets Denoising Autoencoder Two-Layer Denoising Undercomplete Autoencoder with Linear Activation Two-Layer Denoising Overcomplete Autoencoder with Linear Activation Two-Layer Denoising Overcomplete Autoencoder with ReLu Activation Conclusion 9. Semisupervised Learning Data Preparation Supervised Model Unsupervised Model Semisupervised Model The Power of Supervised and Unsupervised Conclusion IV. Deep Unsupervised Learning Using TensorFlow and Keras 10. Recommender Systems Using Restricted Boltzmann Machines Boltzmann Machines Restricted Boltzmann Machines Recommender Systems Collaborative Filtering The Netflix Prize MovieLens Dataset Data Preparation Define the Cost Function: Mean Squared Error Perform Baseline Experiments Matrix Factorization One Latent Factor Three Latent Factors Five Latent Factors Collaborative Filtering Using RBMs RBM Neural Network Architecture Build the Components of the RBM Class Train RBM Recommender System Conclusion 11. Feature Detection Using Deep Belief Networks Deep Belief Networks in Detail MNIST Image Classification Restricted Boltzmann Machines Build the Components of the RBM Class Generate Images Using the RBM Model View the Intermediate Feature Detectors Train the Three RBMs for the DBN Examine Feature Detectors View Generated Images The Full DBN How Training of a DBN Works Train the DBN How Unsupervised Learning Helps Supervised Learning Generate Images to Build a Better Image Classifier Image Classifier Using LightGBM Supervised Only Unsupervised and Supervised Solution Conclusion 12. Generative Adversarial Networks GANs, the Concept The Power of GANs Deep Convolutional GANs Convolutional Neural Networks DCGANs Revisited Generator of the DCGAN Discriminator of the DCGAN Discriminator and Adversarial Models DCGAN for the MNIST Dataset MNIST DCGAN in Action Synthetic Image Generation Conclusion 13. Time Series Clustering ECG Data Approach to Time Series Clustering k-Shape Time Series Clustering Using k-Shape on ECGFiveDays Data Preparation Training and Evaluation Time Series Clustering Using k-Shape on ECG5000 Data Preparation Training and Evaluation Time Series Clustering Using k-Means on ECG5000 Time Series Clustering Using Hierarchical DBSCAN on ECG5000 Comparing the Time Series Clustering Algorithms Full Run with k-Shape Full Run with k-Means Full Run with HDBSCAN Comparing All Three Time Series Clustering Approaches Conclusion 14. Conclusion Supervised Learning Unsupervised Learning Scikit-Learn TensorFlow and Keras Reinforcement Learning Most Promising Areas of Unsupervised Learning Today The Future of Unsupervised Learning Final Words Index