(states, rewards, actions, values)
| 70 | _train = trainer.apply_gradients(grads_and_params) |
| 71 | |
| 72 | def train(states, rewards, actions, values): |
| 73 | advs = rewards - values |
| 74 | feed_dict = {train_model.X: states, A: actions, ADV: advs, R: rewards, LR: lr} |
| 75 | policy_loss, value_loss, policy_entropy, _ = sess.run( |
| 76 | [pg_loss, vf_loss, entropy, _train], |
| 77 | feed_dict |
| 78 | ) |
| 79 | return policy_loss, value_loss, policy_entropy |
| 80 | |
| 81 | def save(save_path): |
| 82 | ps = sess.run(params) |