Skip to content

Fix memories for recurrent training #3066

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Parent: Release 0.12.1
merged 1 commit into from
Dec 10, 2019
Merged

Fix memories for recurrent training #3066

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 5 additions & 2 deletions ml-agents/mlagents/trainers/ppo/policy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,10 +249,13 @@ def get_value_estimates(
]
if self.use_vec_obs:
feed_dict[self.model.vector_in] = [brain_info.vector_observations[idx]]
agent_id = brain_info.agents[idx]
if self.use_recurrent:
feed_dict[self.model.memory_in] = self.retrieve_memories([idx])
feed_dict[self.model.memory_in] = self.retrieve_memories([agent_id])
if not self.use_continuous_act and self.use_recurrent:
feed_dict[self.model.prev_action] = self.retrieve_previous_action([idx])
feed_dict[self.model.prev_action] = self.retrieve_previous_action(
[agent_id]
)
value_estimates = self.sess.run(self.model.value_heads, feed_dict)

value_estimates = {k: float(v) for k, v in value_estimates.items()}
Expand Down
12 changes: 6 additions & 6 deletions ml-agents/mlagents/trainers/tf_policy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,9 +56,9 @@ def __init__(self, seed, brain, trainer_parameters):
self.seed = seed
self.brain = brain
self.use_recurrent = trainer_parameters["use_recurrent"]
self.memory_dict: Dict[int, np.ndarray] = {}
self.memory_dict: Dict[str, np.ndarray] = {}
self.num_branches = len(self.brain.vector_action_space_size)
self.previous_action_dict: Dict[int, np.array] = {}
self.previous_action_dict: Dict[str, np.array] = {}
self.normalize = trainer_parameters.get("normalize", False)
self.use_continuous_act = brain.vector_action_space_type == "continuous"
if self.use_continuous_act:
Expand Down Expand Up @@ -181,14 +181,14 @@ def make_empty_memory(self, num_agents):
return np.zeros((num_agents, self.m_size), dtype=np.float)

def save_memories(
self, agent_ids: List[int], memory_matrix: Optional[np.ndarray]
self, agent_ids: List[str], memory_matrix: Optional[np.ndarray]
) -> None:
if memory_matrix is None:
return
for index, agent_id in enumerate(agent_ids):
self.memory_dict[agent_id] = memory_matrix[index, :]

def retrieve_memories(self, agent_ids: List[int]) -> np.ndarray:
def retrieve_memories(self, agent_ids: List[str]) -> np.ndarray:
memory_matrix = np.zeros((len(agent_ids), self.m_size), dtype=np.float)
for index, agent_id in enumerate(agent_ids):
if agent_id in self.memory_dict:
Expand All @@ -209,14 +209,14 @@ def make_empty_previous_action(self, num_agents):
return np.zeros((num_agents, self.num_branches), dtype=np.int)

def save_previous_action(
self, agent_ids: List[int], action_matrix: Optional[np.ndarray]
self, agent_ids: List[str], action_matrix: Optional[np.ndarray]
) -> None:
if action_matrix is None:
return
for index, agent_id in enumerate(agent_ids):
self.previous_action_dict[agent_id] = action_matrix[index, :]

def retrieve_previous_action(self, agent_ids: List[int]) -> np.ndarray:
def retrieve_previous_action(self, agent_ids: List[str]) -> np.ndarray:
action_matrix = np.zeros((len(agent_ids), self.num_branches), dtype=np.int)
for index, agent_id in enumerate(agent_ids):
if agent_id in self.previous_action_dict:
Expand Down