Code and Data: Install

View on and Install via moonshape_subpopulation

Repo Structure Overview

.
├── README.md
├── main.py
├── utils.py
├── figures/
├── datasets/
│   └── metashift/metashift_prepare.ipynb
│   └── modified-cifar4/cifar4_prepare.ipynb
│   └── officehome/officehome_prepare.ipynb
│   └── pacs/PACS_prepare.ipynb
│   └── waterbireds/waterbireds_prepare.ipynb

We organize the code in the orders of the figures as presented in the paper. As the folder name indicated, the datasets folder provides the code for generating datasets.

Citation

@misc{liang2023accuracy,
      title={Accuracy on the Curve: On the Nonlinear Correlation of ML Performance Between Data Subpopulations},
      author={Weixin Liang and Yining Mao and Yongchan Kwon and Xinyu Yang and James Zou},
      year={2023},
      eprint={2305.02995},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}