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

tf_agents.utils.common.get_contiguous_sub_episodes

View source on GitHub

Computes mask on sub-episodes which includes only contiguous components.

tf_agents.utils.common.get_contiguous_sub_episodes(
    next_time_steps_discount
)

Args:

  • next_time_steps_discount: Tensor of shape [batch_size, total_steps] corresponding to environment discounts on next time steps (i.e. next_time_steps.discount).

Returns:

A float Tensor of shape [batch_size, total_steps] specifying mask including only contiguous components. Each row will be of the form [1.0] * a + [0.0] * b, where a >= 1 and b >= 0, and in which the initial sequence of ones corresponds to a contiguous sub-episode.