Model interpretability is a seminal topic in the field of data science. The field is moving quickly. Practitioners can leverage a range of tools to provide explainability to their models from traditional approaches to the latest in deep neural network explainers.
This project contains starter code for a few of these approaches and some educational material on xAI.
- SHAP_and_LIME.ipynb - a how-to notebook
- rtemis.R - a how-to R file
- traditional_methods.ipynb - a how-to notebook
- Navigating Interpretable and Predictive Models.pdf - slides from the Domino tech talk on xAI which includes many links to further research
- Explore the SHAP_and_LIME.ipynb notebook
- Learn more about xAI by browsing the instructional pdf
This project uses standard python libraries and any base Domino image should work well. The additional Python libraries needed are shap, lime, and pycebox.
There are several additional R libraries needed to run rtemis. This library changes frequently, sometimes breaking dependencies. See the R script included in this project for details.