Module epiclass.utils.my_logging

Implement logging of specific data through comet logger.

Functions

def log_dset_composition(my_data: DataSet, logdir: str | Path | None, logger: pl_loggers.CometLogger | None, split_nb: int)

Logs dataset composition to logger and file.

Either the 'logdir' or 'logger' parameter must be provided, otherwise ValueError is raised.

This function logs - training/validation set size - total number of unique files (training + validation), - training/valid unique md5 composition (to a file)

Args

my_data : DataSet

The dataset to log.

logdir (str | Path | None): The directory where logs will be stored. None if logger is given.

logger (pl_loggers.CometLogger | None): The logger object to use for logging. None if logdir is given.

split_nb : int

The split number to log.

Raises

ValueError if: - both 'logdir' and 'logger' parameters are None - train and validation sets overlap.

def log_pre_training(logger: pl_loggers.CometLogger, to_log: Dict[str, str], step: int | None)

Log a bunch of stuff in comet logger. Return experience key (str).

to_log expects keys: - category - hdf5_resolution - loading_time (initial, for hdf5) When step is an int: - split_time (generator yield time)