| 96 | self.returns = self.advantages + self.values |
| 97 | |
| 98 | def add(self, |
| 99 | obs_dict: Dict[str, np.ndarray], |
| 100 | action: np.ndarray, |
| 101 | reward: np.ndarray, |
| 102 | done: np.ndarray, |
| 103 | value: np.ndarray, |
| 104 | log_prob: np.ndarray, |
| 105 | mu: np.ndarray, |
| 106 | sigma: np.ndarray, |
| 107 | infos) -> None: |
| 108 | |
| 109 | for k, v in obs_dict.items(): |
| 110 | self.observations[k][self.pos] = v |
| 111 | self.actions[self.pos] = action |
| 112 | self.rewards[self.pos] = reward |
| 113 | self.dones[self.pos] = done |
| 114 | self.values[self.pos] = value |
| 115 | self.log_probs[self.pos] = log_prob |
| 116 | self.mus[self.pos] = mu |
| 117 | self.sigmas[self.pos] = sigma |
| 118 | |
| 119 | for i in range(self.n_envs): |
| 120 | self.reward_debugs[i].append(infos[i]['reward_debug']['debug_texts']) |
| 121 | self.terminal_debugs[i].append(infos[i]['terminal_debug']['debug_texts']) |
| 122 | |
| 123 | n_steps = infos[i]['terminal_debug']['exploration_suggest']['n_steps'] |
| 124 | if n_steps > 0: |
| 125 | n_start = max(0, self.pos-n_steps) |
| 126 | self.exploration_suggests[n_start:self.pos, i] = \ |
| 127 | infos[i]['terminal_debug']['exploration_suggest']['suggest'] |
| 128 | |
| 129 | self.pos += 1 |
| 130 | if self.pos == self.buffer_size: |
| 131 | self.full = True |
| 132 | |
| 133 | def update_values(self, policy): |
| 134 | for i in range(self.buffer_size): |