c28b2e4cfd60a5b131a1f8752d484b9c99d05e81,main.py,,main,#,44

Before Change

                    // Reshape to do in a single forward pass for all steps
                    values, action_log_probs, dist_entropy = actor_critic.evaluate_actions(Variable(observations_batch), Variable(actions_batch))

                    _, old_action_log_probs, _ = old_model.evaluate_actions(Variable(observations_batch, volatile=True), Variable(actions_batch, volatile=True))

                    ratio = torch.exp(action_log_probs - Variable(old_action_log_probs.data))
                    adv_targ = Variable(advantages.view(-1, 1)[indices])
                    surr1 = ratio * adv_targ

After Change

                    observations_batch = rollouts.observations[:-1].view(-1, *obs_shape)[indices]
                    actions_batch = rollouts.actions.view(-1, action_shape)[indices]
                    return_batch = rollouts.returns[:-1].view(-1, 1)[indices]
                    old_action_log_probs_batch = rollouts.action_log_probs.view(-1, 1)[indices]

                    // Reshape to do in a single forward pass for all steps
                    values, action_log_probs, dist_entropy = actor_critic.evaluate_actions(Variable(observations_batch), Variable(actions_batch))

Instances