WILDs is a benchmark of in-the-wild distribution shifts spanning diverse data modalities and applications, from tumor identification to wildlife monitoring to poverty mapping.
This repo contains my infrastructure to maintain experiments on the dataset.
First, download docker.
Second, run for an interactive terminal
make run
make run -- python src/train.py
--experiment-id 'test_model'
- See example test experiment at
src/experiments/test_model.py
- Inside of
src/experiments/experiments.py
import the new experiment - Run
make run -- python src/train.py --experiment-id '<experiment_name>'
make lint
make test
@article{wilds-project,
title={Wilds project},
author={Adrian Orenstein},
journal={GitHub. Note: https://github.com/AdrianOrenstein/wilds-project},
year={2021}
}
git push public public:master