def add_path(self, path):
obs = path["observations"]
actions = path["actions"]
next_obs = path["next_observations"]
rewards = path["rewards"]
terminals = path["terminals"]
path_len = len(terminals)
# assert path_len == self.max_path_length
actions = flatten_n(actions)
obs = flatten_dict(obs, self.ob_keys_to_save)
next_obs = flatten_dict(next_obs, self.ob_keys_to_save)
self._actions[self._top][:path_len] = actions
rewards = np.squeeze(rewards)
rewards = np.expand_dims(rewards, -1)
self._environment_rewards[self._top][:path_len] = rewards
self._terminals[self._top][:path_len] = terminals
if len(np.argwhere(terminals)):
episode_length = np.argwhere(terminals)[:, 0][0] + 1
assert episode_length == path_len
else:
episode_length = self.max_path_length
self._episode_lengths[self._top] = episode_length
for key in self.ob_keys_to_save:
self._obs[key][self._top][:path_len] = obs[key]
self._next_obs[key][self._top][:path_len] = next_obs[key]
self._actions[self._top][:path_len] = actions
self._advance()
def add_path(self, path):
obs = path["observations"]
actions = path["actions"]
next_obs = path["next_observations"]
terminals = path["terminals"]
path_len = len(terminals)
assert path_len == self.max_path_length
actions = flatten_n(actions)
obs = flatten_dict(obs, self.ob_keys_to_save)
next_obs = flatten_dict(next_obs, self.ob_keys_to_save)
self._actions[self._top] = actions
self._terminals[self._top] = terminals
for key in self.ob_keys_to_save:
self._obs[key][self._top] = obs[key]
self._next_obs[key][self._top] = next_obs[key]
self._actions[self._top] = actions
self._advance()
def add_decoded_vae_goals_to_path(self, path):
# decoding the self-sampled vae images should be done in batch (here)
# rather than in the env for efficiency
desired_goals = flatten_dict(
path["observations"],
[self.desired_goal_key])[self.desired_goal_key]
desired_decoded_goals = self.env._decode(desired_goals)
desired_decoded_goals = desired_decoded_goals.reshape(
len(desired_decoded_goals), -1)
for idx, next_obs in enumerate(path["observations"]):
path["observations"][idx][
self.decoded_desired_goal_key] = desired_decoded_goals[idx]
path["next_observations"][idx][
self.decoded_desired_goal_key] = desired_decoded_goals[idx]
def add_decoded_vae_goals_to_path(self, path):
# decoding the self-sampled vae images should be done in batch (here)
# rather than in the env for efficiency
if 'latent_desired_goal' in path['observations'][0].keys():
desired_encoded_goals = flatten_dict(
path['observations'],
['latent_desired_goals']
)['latent_desired_goals']
desired_decoded_goals = self.env._decode(desired_encoded_goals)
desired_decoded_goals = desired_decoded_goals.reshape(
len(desired_decoded_goals),
-1
)
for idx, next_obs in enumerate(path['observations']):
path['observations'][idx][self.decoded_desired_goal_key] = \
desired_decoded_goals[idx]
path['next_observations'][idx][self.decoded_desired_goal_key] = \
desired_decoded_goals[idx]