25103f21c5fb2a0b97fd92f95a1ca870f27ce09c,tf_agents/drivers/dynamic_episode_driver.py,DynamicEpisodeDriver,run,#DynamicEpisodeDriver#Any#Any#Any#Any#,110

Before Change

    // collection.
    batch_dims = nest_utils.get_outer_shape(
        time_step, self.env.time_step_spec())
    counter = tf.zeros(batch_dims, tf.int32)

    num_episodes = num_episodes or self._num_episodes
    [_, time_step, policy_state] = tf.while_loop(
        cond=self._loop_condition_fn(num_episodes),
        body=self._loop_body_fn(),
        loop_vars=[
            counter,
            time_step,
            policy_state],
        back_prop=False,
        parallel_iterations=1,
        maximum_iterations=maximum_iterations,
        name="driver_loop"
    )
    return time_step, policy_state

After Change

      time_step: TimeStep named tuple with final observation, reward, etc.
      policy_state: Tensor with final step policy state.
    
    return self._run_fn(time_step=time_step,
                        policy_state=policy_state,
                        num_episodes=num_episodes,
                        maximum_iterations=maximum_iterations)

  def _run(self,
           time_step=None,
           policy_state=None,

Instances