You're reading from Debugging Machine Learning Models with Python Develop high-performance, low-bias, and explainable machine learning and deep learning models

Product type Paperback

Published in Sep 2023

Publisher Packt

ISBN-13 9781800208582

Length 344 pages

Edition 1st Edition

Languages

Python

Tools

TensorFlow

Concepts

Deep Learning

Author (1):

Ali Madani

View More author details

Table of Contents (26) Chapters

Preface

1. Part 1:Debugging for Machine Learning Modeling

2. Chapter 1: Beyond Code Debugging FREE CHAPTER

3. Chapter 2: Machine Learning Life Cycle

4. Chapter 3: Debugging toward Responsible AI

5. Part 2:Improving Machine Learning Models

6. Chapter 4: Detecting Performance and Efficiency Issues in Machine Learning Models

7. Chapter 5: Improving the Performance of Machine Learning Models

8. Chapter 6: Interpretability and Explainability in Machine Learning Modeling

9. Chapter 7: Decreasing Bias and Achieving Fairness

10. Part 3:Low-Bug Machine Learning Development and Deployment

11. Chapter 8: Controlling Risks Using Test-Driven Development

12. Chapter 9: Testing and Debugging for Production

13. Chapter 10: Versioning and Reproducible Machine Learning Modeling

14. Chapter 11: Avoiding and Detecting Data and Concept Drifts

15. Part 4:Deep Learning Modeling

16. Chapter 12: Going Beyond ML Debugging with Deep Learning

17. Chapter 13: Advanced Deep Learning Techniques

18. Chapter 14: Introduction to Recent Advancements in Machine Learning

19. Part 5:Advanced Topics in Model Debugging

20. Chapter 15: Correlation versus Causality

21. Chapter 16: Security and Privacy in Machine Learning

22. Chapter 17: Human-in-the-Loop Machine Learning

23. Assessments

24. Index

Why subscribe?

25. Other Books You May Enjoy

Modeling data preparation

In this stage of a machine learning life cycle, we need to finalize the features and data points we want to use for modeling, as well as our model evaluation and testing strategies.

Feature selection and extraction

The original features that were normalized and scaled in previous steps can be now processed further to increase the likelihood of having a high-performance model. In general, features can either be sub-selected, meaning some of the features get thrown out, using a feature selection method, or be used to generate new features, which is traditionally called feature extraction.

Feature selection

The goal of feature selection is to reduce the number of features, or the dimensionality of your data, and keep features that are information-rich. For example, if we have 20,000 features and 500 data points, there is a high chance that most of the original 20,000 features are not informative when used to build a supervised learning model. The...