This repo provides supporting Python code for the paper
Xu, Z., Dan, C., Khim, J., Ravikumar, P. (2020). Class-Weighted Classification: Trade-offs and Robust Approaches. arXiv preprint arXiv:2005.12914.
Requires conda with Python 3.7.
- Install conda dependencies in the environment:
conda env create -f environment.yml
- Run
download_uci_data.sh
from the repo main directory to download the Covertype dataset. - Activate the conda enviroment with
conda activate robust_weighting
- Setup up a wandb account and create a project named
extreme-classification
(or rename theproject
argument inside thewandb.init
call insidesrc/main.py
)
Navigate to the root of the repo.
Run ./power_exp.sh
for the synthetic experiment results.
Run ./uci_exp.sh
for the real world dataset (Covertype) results.
View results in the wandb website.