Missed TensorFlow Dev Summit? Check out the video playlist. Watch recordings

tf_agents.utils.common.Checkpointer

View source on GitHub

Checkpoints training state, policy state, and replay_buffer state.

tf_agents.utils.common.Checkpointer(
    ckpt_dir, max_to_keep=20, **kwargs
)

Used in the notebooks

Used in the tutorials

Args:

  • ckpt_dir: The directory to save checkpoints.
  • max_to_keep: Maximum number of checkpoints to keep (if greater than the max are saved, the oldest checkpoints are deleted).
  • **kwargs: Items to include in the checkpoint.

Attributes:

  • checkpoint_exists

Methods

initialize_or_restore

View source

initialize_or_restore(
    session=None
)

Initialize or restore graph (based on checkpoint if exists).

save

View source

save(
    global_step
)

Save state to checkpoint.