Manager

class hdnnpy.training.manager.Manager(tag, trainer, result, is_snapshot=True)[source]

Bases: contextlib.AbstractContextManager

Context manager to take trainer snapshot and decide whether to train or not.

Parameters:
  • tag (str) – Tag of dataset used for training.
  • trainer (Trainer) – Trainer object to be managed.
  • result (dict) – Dictionary object containing total elapsed time and metrics value corresponding to the type of loss function. Even when training is stopped / resumed, it is retained.
  • is_snapshot (bool, optional) – Take trainer snapshot if True.
__enter__()[source]

Replace signal handler of SIGINT and SIGTERM.

__exit__(type_, value, traceback)[source]

Restore signal handler of SIGINT and SIGTERM, and record the result of training.

check_to_resume(resume_tag)[source]

Decide whether to train or not.

If current tag of dataset is equal to resume_tag, restore the state of trainer from snapshot file.

Parameters:resume_tag (str) – Tag of dataset when snapshot was taken last time.
allow_to_run

Whether the given trainer can train with the dataset.