Install the required Python packages by running:
conda env create -f environment.yml
conda activate cc19
Or build the Docker image by:
docker build -t cc19:latest .
To do a hyperparam tuning, copy the example config files from configs
to your
<log-dir>
, modify them accordingly and execute:
python tune.py --logdir <log-dir> --num-samples <search-samples> --max-epochs <max-epochs>
You can use -s
to run a smoke test (only 2 samples with 2 epochs) or pass in
--subset <fraction>
to run the experiments with only fraction*len(dataset)
data points, i.e. 0.5 is half of the original ds.