tf_agents.trajectories.to_transition

Create a transition from a trajectory or two adjacent trajectories.

View aliases

Main aliases

tf_agents.trajectories.trajectory.to_transition

tf_agents.trajectories.to_transition(
    trajectory: tf_agents.trajectories.Trajectory,
    next_trajectory: Optional[tf_agents.trajectories.Trajectory] = None
) -> tf_agents.trajectories.Transition

Note: If next_trajectory is not provided, tensors of trajectory are sliced along their second (time) dimension; for example:

time_steps.step_type = trajectory.step_type[:,:-1]
time_steps.observation = trajectory.observation[:,:-1]
next_time_steps.observation = trajectory.observation[:,1:]
next_time_steps. step_type = trajectory. next_step_type[:,:-1]
next_time_steps.reward = trajectory.reward[:,:-1]
next_time_steps. discount = trajectory. discount[:,:-1]

Notice that reward and discount for time_steps are undefined, therefore filled with zero.

Args
trajectory	An instance of Trajectory. The tensors in Trajectory must have shape [B, T, ...] when next_trajectory is None. discount is assumed to be a scalar float; hence the shape of trajectory.discount must be [B, T].
next_trajectory	(optional) An instance of Trajectory.

Returns
A tuple (time_steps, policy_steps, next_time_steps). The reward and discount fields of time_steps are filled with zeros because these cannot be deduced (please do not use them).

Raises
ValueError	if discount rank is not within the range [1, 2].