# ReinventWorkshopTools

## Re:Invent 2022 Workshop AIM342: Advancing Responsible AI

This project provides a Jupyter notebook and supporting utilities that let you explore unwanted bias in a multi-way classification model. 

Experiments in the notebook include: 
* Train a baseline model
* Measure classification accuracy
* Measure and visualize bias: disparities in accuracy for subsets of the data
* Compare bias metrics: relative vs. absolute, error vs. accuracy
* Compare test sets: balanced/unbalanced evaluation data and confidence intervals
* Address balanced/unbalanced training data with resampling
* Explore bias for high- and low-granularity data subsets
* Disentangle competing causes of bias (typical vs. atypical examples)



This library is licensed under the MIT-0 License. See the [LICENSE](LICENSE) file.