Attached is the code supporting the ICML submission "FairPFN: A Tabular Foundation Model for Causal Fairness."
Create a conda environment with python version 3.10, activate it, and install the requirements in requirements.txt
conda create -n fairpfn_env python=3.10
conda activate fairpfn_env
pip install -r requirements.txt
Run inference_example.py to play around with FairPFN on any of our real-world and synthetic benchmark datasets by changing the dataset variable (runs on Law School by default)
Run prior_data_example.py to sample data from our prior. You can change the values of listed hyperparameters in config_sample_ to change the MLP/SCM that the data is sampled is from.