# tf_agents.environments.TrajectoryReplay

<br />

|---|
| [![](https://www.tensorflow.org/images/GitHub-Mark-32px.png) View source on GitHub](https://github.com/tensorflow/agents/blob/v0.19.0/tf_agents/environments/trajectory_replay.py#L29-L275) |

A helper that replays a policy against given `Trajectory` observations.

#### View aliases


**Main aliases**

[`tf_agents.environments.trajectory_replay.TrajectoryReplay`](https://www.tensorflow.org/agents/api_docs/python/tf_agents/environments/TrajectoryReplay)

<br />

    tf_agents.environments.TrajectoryReplay(
        policy, time_major=False
    )

<br />

<br />

<br />

| ## Args ||
|---|---|
| `policy` | A tf_policy.TFPolicy policy. |
| `time_major` | If `True`, the tensors in `trajectory` passed to method `run` are assumed to have shape `[time, batch, ...]`. Otherwise (default) they are assumed to have shape `[batch, time, ...]`. |

<br />

<br />

<br />

<br />

| ## Raises ||
|---|---|
| `ValueError` | If policy is not an instance of tf_policy.TFPolicy. |

<br />

## Methods

### `run`

[View source](https://github.com/tensorflow/agents/blob/v0.19.0/tf_agents/environments/trajectory_replay.py#L52-L275)

    run(
        trajectory, policy_state=None
    )

Apply the policy to trajectory steps and store actions/info.

If `self.time_major == True`, the tensors in `trajectory` are assumed to
have shape `[time, batch, ...]`. Otherwise they are assumed to
have shape `[batch, time, ...]`.

<br />

<br />

<br />

| Args ||
|---|---|
| `trajectory` | The `Trajectory` to run against. If the replay class was created with `time_major=True`, then the tensors in trajectory must be shaped `[time, batch, ...]`. Otherwise they must be shaped `[batch, time, ...]`. |
| `policy_state` | (optional) A nest Tensor with initial step policy state. |

<br />

<br />

<br />

<br />

| Returns ||
|---|---|
| `output_actions` | A nest of the actions that the policy took. If the replay class was created with `time_major=True`, then the tensors here will be shaped `[time, batch, ...]`. Otherwise they'll be shaped `[batch, time, ...]`. |
| `output_policy_info` | A nest of the policy info that the policy emitted. If the replay class was created with `time_major=True`, then the tensors here will be shaped `[time, batch, ...]`. Otherwise they'll be shaped `[batch, time, ...]`. |
| `policy_state` | A nest Tensor with final step policy state. |

<br />

<br />

<br />

<br />

| Raises ||
|---|---|
| `TypeError` | If `policy_state` structure doesn't match `self.policy.policy_state_spec`, or `trajectory` structure doesn't match `self.policy.trajectory_spec`. |
| `ValueError` | If `policy_state` doesn't match `self.policy.policy_state_spec`, or `trajectory` structure doesn't match `self.policy.trajectory_spec`. |
| `ValueError` | If `trajectory` lacks two outer dims. |

<br />