Detecting Bias in Datasets

What is it

When to use it

The Coverage class should be used when you have lots of images, but only a small fraction from certain regimes/labels.

The Parity class and similar should be used when you would like to determine if two datasets have statistically independent labels.

Theory behind it