Workshop materials for ODSC 2019 Interpretability Talk
The Notebooks here are to show:
- Reading Skills Variable Importance, which shows how split based importance is misleading compared to permutation based approach.
- Lack of Fidelity-Diversity using LIME on Adult, which shows how the Adult dataset on LIME provides the same top explanation for every explanation.
- Lack of Consistency using LIME on Adult, which shows how the same datapoint can have different explanations.
- Lack of Consistency LIME Boston, which shows how the same datapoint can have different explanations.
For more on learning about permutation, partial dependence, and explanations, take a look at the excellent Kaggle mini course on explainability as well as the Intepretable Machine Learning book.