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

tf_agents.agents.ppo.ppo_utils.make_timestep_mask

View source on GitHub

Create a mask for transitions and optionally final incomplete episodes.

tf_agents.agents.ppo.ppo_utils.make_timestep_mask(
    batched_next_time_step, allow_partial_episodes=False
)

Args:

  • batched_next_time_step: Next timestep, doubly-batched [batch_dim, time_dim, ...].
  • allow_partial_episodes: If true, then steps on incomplete episodes are allowed.

Returns:

A mask, type tf.float32, that is 0.0 for all between-episode timesteps (batched_next_time_step is FIRST). If allow_partial_episodes is set to False, the mask has 0.0 for incomplete episode at the end of the sequence.