Experiment tracking with Weights & Biases

Weights & Biases can track experiments, visualize results, and compare model fits.

Setup Weights & Biases

  • Create account at https://wandb.ai

  • Install the wandb python package and log in:

conda install wandb -c conda-forge
wandb login

Log your DAS runs:

  • CLI:

das train OTHER_CLI_ARGS --wandb-token MY_SUPER_SECRET_TOKEN --wandb-entity django_reinhardt --wandb-project test_run

  • Python:

das.train.train(..., wandb-token="MY_SUPER_SECRET_TOKEN", wandb-entity="django_reinhardt", wandb_project="test_run")

  • Args:

    • wandb_api_token: API token from wandb. See the wandb docs on how to generate the token.

    • wandb_project: Project to log to - useful for grouping runs.

    • wandb_entity: User or team name to log to.