rosikand / waterbirds-starter

Provides scaffolding for training and testing methods for robustness on the Waterbirds dataset. Uses torchplate.

Waterbirds starter code

Provides scaffolding for training models on the Waterbirds dataset for robustness techniques. Makes use of torchplate.

Directions:

1. Specify your config parameters (e.g., data paths, batch size) in configs.py.
2. Create custom experiments in experiments.py to test out new methods or use experiments.BaseExperiment.
3. Train and test the method using runner.py and specify the relevant CLI args. Example:

python runner.py -e BaseExp -c BaseConfig -train -test -num_epochs 10 -grad_accum 1