This repository contains the code for training and evaluating reinforcement learning-based algorithms that may be used for the mitigation of gender bias in occupation prediction. The data cannot be published here, but the link to crawl it is supplied.
The models can be trained using the notebook RL_gender_bias. In order to run it, change the path names to the location of the data and run all cells. The predictions made by the different models may be used to calculate TPR gender gaps, a measure for the amount of gender bias that a model has: Code for this is also found in the notebook.
The report can be found via this link:
The data can be crawled by following the steps as described in https://github.com/Microsoft/biosbias