| Publisher | Packt Publishing (March 26, 2021) |
| Language | English |
| Paperback | 736 pages |
| ISBN-10 | 180020390X |
| ISBN-13 | 978-1800203907 |
| Item Weight | 2.77 pounds |
| Dimensions | 9.25 x 7.5 x 1.51 inches |
| Best Sellers Rank | #1,361,110 in Books (See Top 100 in Books) #219 in Computer Simulation (Books) #375 in Artificial Intelligence Expert Systems #571 in Computer Neural Networks |
| Customer Reviews | 4.6 out of 5 stars 56Reviews |
With AI systems replacing or complementing human decision-makers with machine learning models designed for the most complex tasks, trust is mission-critical. And thus, understanding how your ML model generates an outcome by complying with the principles of interpretable ML ensures the reliability of the ML model.
This book is a comprehensive hands-on guide to all things machine learning interpretability, presenting its topics with the help of real-world examples. Interpretable Machine Learning with Python takes you through the fundamentals and challenges in interpretation to help you design your systems with fairness, accountability, and transparency - the core principles of interpretable ML synonymous with Explainable Artificial Intelligence (XAI). This book will help you to mitigate the risks associated with poor predictions.
Interpretable Machine Learning with Python is an extensive guide that tackles both sides of the equation: the diagnosis and the treatment of interpretability concerns. It goes beyond transparency to cover fairness and accountability, which are often ignored or underplayed by most practitioner-oriented books on the topic.
This book is mission-centric. Every chapter takes you on a journey to discover a wide range of topics using case studies that are as realistic as possible. Therefore, ‘toy datasets’ such as MNIST, Iris, and Titanic, which are too clean to depict real-world conditions, are not included.
In my 15 years of development experience, I've learned that for a product to be adopted and embraced, you have to trust it, and to trust it, you have to understand it.
In web development in particular, even if a website is up 99.9% of the time, stakeholders remember more the times that the website was down than those times it wasn't. I realized how important it was to explain predictions and assure a degree of reliability or, at least, anticipate points of failure. However, unlike software, complex ML models can't be debugged line by line. Even ML models with high predictive performance still get it wrong sometimes, and understanding the ways they could fail can help improve outcomes or, at least, manage expectations.
This book allows you to look under the hood and demystify the "black-box" ML model so that you can make assurances to stakeholders and mitigate issues such as overfitting, unfair outcomes, uncertainty, and lack of adversarial robustness.
Interpretation is often seen as an essential skill for descriptive analytics; however, it's also very much leveraged in predictive and prescriptive analytics. With this book, you'll realize that training a good machine learning model is more than just optimizing predictive performance. The goodness of a model can be measured in many ways, such as those encompassed by concepts of fairness and robustness. Interpretable machine learning is not limited to a toolset for making complex models explainable, instead you can learn from a model and improve it in more ways than with predictive performance. Interpretable ML is also how Ethical AI, Responsible AI, and Fair AI are implemented by ML practitioners.
