Allows specifying a --checkpoint-dir to which the training state will be persisted after every iteration. If a run is interrupted, it can be fully restored from the checkpoint directory. Also makes it possible to load policies after training has completed.
Allows specifying a
--checkpoint-dirto which the training state will be persisted after every iteration. If a run is interrupted, it can be fully restored from the checkpoint directory. Also makes it possible to load policies after training has completed.