das.tracking#

Utilities for logging training runs.

We currenlty have integrations for tensorboard, and wandb.ai. While tensorboard is integrated with tensorflow. To use wandb you’ll have to install the wandb API: “pip install wandb” or “conda install wandb -c conda-forge”.

class das.tracking.Wandb(project: Optional[str] = None, api_token: Optional[str] = None, entity: Optional[str] = None, params: Optional[Dict] = None, infer_from_env: bool = False)[source]#

Utility class for logging to wandb.ai during training.

Parameters

  • project (Optional[str], optional) – Project to log to. Defaults to None.

  • api_token (Optional[str], optional) – api token. Defaults to None.

  • entity (Optional[str], optional) – Entity (user/team name). Defaults to None.

  • params (Optional[Dict], optional) – Dict to log to config. Defaults to None.

  • infer_from_env (bool, optional) – read project and api_token from environment variables WANDB_PROJECT and WANDB_API_TOKEN. Defaults to False.

callback(save_model=False)[source]#

Get callback for auto-logging from tensorfow/keras.

log_test_results(report: Dict)[source]#

Log final classification result from test data.

Parameters

report (Dict) – dictionary containing the classification report.