Practical Explainable AI Using Python: Artificial Intelligence Model Explanations Using Python-based Libraries, Extensions, and Frameworks

✍ Scribed by Pradeepta Mishra

• Publisher: Apress
• Year: 2022
• Tongue: English
• Leaves: 356
• Edition: 1
• Category: Library

✦ Synopsis

Learn the ins and outs of decisions, biases, and reliability of AI algorithms and how to make sense of these predictions. This book explores the so-called black-box models to boost the adaptability, interpretability, and explainability of the decisions made by AI algorithms using frameworks such as Python XAI libraries, TensorFlow 2.0+, Keras, and custom frameworks using Python wrappers.

You'll begin with an introduction to model explainability and interpretability basics, ethical consideration, and biases in predictions generated by AI models. Next, you'll look at methods and systems to interpret linear, non-linear, and time-series models used in AI. The book will also cover topics ranging from interpreting to understanding how an AI algorithm makes a decision

Further, you will learn the most complex ensemble models, explainability, and interpretability using frameworks such as Lime, SHAP, Skater, ELI5, etc. Moving forward, you will be introduced to model explainability for unstructured data, classification problems, and natural language processing–related tasks. Additionally, the book looks at counterfactual explanations for AI models. Practical Explainable AI Using Python shines the light on deep learning models, rule-based expert systems, and computer vision tasks using various XAI frameworks.

What You'll Learn

• Review the different ways of making an AI model interpretable and explainable
• Examine the biasness and good ethical practices of AI models
• Quantify, visualize, and estimate reliability of AI models
• Design frameworks to unbox the black-box models
• Assess the fairness of AI models
  
• Understand the building blocks of trust in AI models
  
• Increase the level of AI adoption

Who This Book Is For

AI engineers, data scientists, and software developers involved in driving AI projects/ AI products.

✦ Table of Contents

Table of Contents
About the Author
About the Technical Reviewers
Acknowledgments
Introduction
Chapter 1: Model Explainability and Interpretability
Establishing the Framework
Artificial Intelligence
Need for XAI
Explainability vs. Interpretability
Explainability Types
Tools for Model Explainability
SHAP
LIME
ELI5
Skater
Skope_rules
Methods of XAI for ML
XAI Compatible Models
XAI Meets Responsible AI
Evaluation of XAI
Conclusion
Chapter 2: AI Ethics, Biasness, and Reliability
AI Ethics Primer
Biasness in AI
Data Bias
Algorithmic Bias
Bias Mitigation Process
Interpretation Bias
Training Bias
Reliability in AI
Conclusion
Chapter 3: Explainability for Linear Models
Linear Models
Linear Regression
VIF and the Problems It Can Generate
Final Model
Model Explainability
Trust in ML Model: SHAP
Local Explanation and Individual Predictions in a ML Model
Global Explanation and Overall Predictions in ML Model
LIME Explanation and ML Model
Skater Explanation and ML Model
ELI5 Explanation and ML Model
Logistic Regression
Interpretation
LIME Inference
Conclusion
Chapter 4: Explainability for Non-Linear Models
Non-Linear Models
Decision Tree Explanation
Data Preparation for the Decision Tree Model
Creating the Model
Decision Tree – SHAP
Partial Dependency Plot
PDP Using Scikit-Learn
Non-Linear Model Explanation – LIME
Non-Linear Explanation – Skope-Rules
Conclusion
Chapter 5: Explainability for Ensemble Models
Ensemble Models
Types of Ensemble Models
Why Ensemble Models?
Using SHAP for Ensemble Models
Using the Interpret Explaining Boosting Model
Ensemble Classification Model: SHAP
Using SHAP to Explain Categorical Boosting Models
Using SHAP Multiclass Categorical Boosting Model
Using SHAP for Light GBM Model Explanation
Conclusion
Chapter 6: Explainability for Time Series Models
Time Series Models
Knowing Which Model Is Good
Strategy for Forecasting
Confidence Interval of Predictions
What Happens to Trust?
Time Series: LIME
Conclusion
Chapter 7: Explainability for NLP
Natural Language Processing Tasks
Explainability for Text Classification
Dataset for Text Classification
Explaining Using ELI5
Calculating the Feature Weights for Local Explanation
Local Explanation Example 1
Local Explanation Example 2
Local Explanation Example 3
Explanation After Stop Word Removal
N-gram-Based Text Classification
Multi-Class Label Text Classification Explainability
Local Explanation Example 1
Local Explanation Example 2
Local Explanation Example 1
Conclusion
Chapter 8: AI Model Fairness Using a What-If Scenario
What Is the WIT?
Installing the WIT
Evaluation Metric
Conclusion
Chapter 9: Explainability for Deep Learning Models
Explaining DL Models
Using SHAP with DL
Using Deep SHAP
Using Alibi
SHAP Explainer for Deep Learning
Another Example of Image Classification
Using SHAP
Deep Explainer for Tabular Data
Conclusion
Chapter 10: Counterfactual Explanations for XAI Models
What Are CFEs?
Implementation of CFEs
CFEs Using Alibi
Counterfactual for Regression Tasks
Conclusion
Chapter 11: Contrastive Explanations for Machine Learning
What Is CE for ML?
CEM Using Alibi
Comparison of an Original Image vs. an Autoencoder-Generated Image
CEM for Tabular Data Explanations
Conclusion
Chapter 12: Model-Agnostic Explanations by Identifying Prediction Invariance
What Is Model Agnostic?
What Is an Anchor?
Anchor Explanations Using Alibi
Anchor Text for Text Classification
Anchor Image for Image Classification
Conclusion
Chapter 13: Model Explainability for Rule-Based Expert Systems
What Is an Expert System?
Backward and Forward Chaining
Rule Extraction Using Scikit-Learn
Need for a Rule-Based System
Challenges of an Expert System
Conclusion
Chapter 14: Model Explainability for Computer Vision
Why Explainability for Image Data?
Anchor Image Using Alibi
Integrated Gradients Method
Conclusion
Index