DataHour: Interpreting Machine Learning Models with Python

Using machine learning models without knowing how they make predictions poses tremendous risks to the organization’s finance and reputation, and more importantly, to the subjects being served or scrutinized by those models. That is why interpreting machine learning models has become an integral part of their development and deployment.

In this DataHour, we’ll discuss various interpretability methods that we can use to explain the predictions of machine learning models, highlighting their advantages and shortcomings. We’ll explore how to explain intrinsically explainable models by understanding their components and parameters. Then we’ll introduce different post-hot interpretability methods that we can leverage to understand the decisions made by less transparent models. We’ll also introduce local explainability methods, like the individual conditional expectation, LIME and SHAP. We will understand the uses and limitations of post-hoc interpretability methods, and how to deploy them effectively using open-source.

Soledad Galli is a Data Scientist, Instructor, and Software Developer. She developed and put into production machine learning models to assess insurance claims, credit risk, and prevent fraud. Sole teaches various online courses on machine learning, which have enrolled 55,000+ students worldwide and consistently receive good student reviews. She is the author of Packt’s “Python Feature Engineering Cookbook” and Leanpub’s “Feature Selection in Machine Learning” book. Sole is also the developer and maintainer of the open-source Python library Feature-engine, which features 150k+ monthly downloads, 3M total downloads and 1.6k stars on Github.