mlberkeley / climatehack2023 Goto Github PK

This repository contains the project files for ML@B's ClimateHack 2023 contest submissions.

Specify the model name in config.model.name with the model name in build.py. We use a yaml file for each model, for easy access.

python main.py -n run_name -c config_filepath

Specify run_name to for wandb logging. Run without flags to use defaults. Run name and configs must be specified.

The model weights and a json copy of the config file used will be saved in ckpts/{run_name}/.

Local eval:

python main.py -n run_name -c config_filepath -t eval

(default behaviour is that main.py will train, not eval)

DOXA local eval:

python doxa_local.py ckpts/run_name

This automatically copies the model weights and config from the folder ckpts/{run_name}/ to the submissions folder, then runs the eval on the model. We recommend running this before submission to make sure everything works as intended!

bash submit.sh ckpts/run_name

Logs into DOXA and submits model.