Skip to content

Reduce tf.function retracing #100

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

Open
ymd-h wants to merge 4 commits into
base: master
Choose a base branch
from
Open

Reduce tf.function retracing #100

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
097ca2d
Ensure tf.Tensor for DDPG._get_action_body
ymd-h Aug 7, 2020
023331c
Ensure tf.Tensor for VPG._get_action_body
ymd-h Aug 7, 2020
12e581d
Ensure tf.Tensor SAC._get_action_body
ymd-h Aug 7, 2020
1ba62f8
Ensure tf.Tensor at DQN._train_body
ymd-h Aug 7, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion tf2rl/algos/ddpg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,7 +97,8 @@ def get_action(self, state, test=False, tensor=False):
state = np.expand_dims(state, axis=0).astype(
np.float32) if is_single_state else state
action = self._get_action_body(
tf.constant(state), self.sigma * (1. - test),
tf.constant(state),
tf.constant(self.sigma * (1. - test), dtype=tf.float32),
tf.constant(self.actor.max_action, dtype=tf.float32))
if tensor:
return action
Expand Down
3 changes: 2 additions & 1 deletion tf2rl/algos/dqn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,7 +192,8 @@ def train(self, states, actions, next_states, rewards, done, weights=None):
if weights is None:
weights = np.ones_like(rewards)
td_errors, q_func_loss = self._train_body(
states, actions, next_states, rewards, done, weights)
tf.constant(states), tf.constant(actions), tf.constant(next_states),
tf.constant(rewards), tf.constant(done), tf.constant(weights))

tf.summary.scalar(name=self.policy_name +
"/q_func_Loss", data=q_func_loss)
Expand Down
2 changes: 1 addition & 1 deletion tf2rl/algos/sac.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,7 +114,7 @@ def get_action(self, state, test=False):

state = np.expand_dims(state, axis=0).astype(
np.float32) if is_single_state else state
action = self._get_action_body(tf.constant(state), test)
action = self._get_action_body(tf.constant(state), tf.constant(test))

return action.numpy()[0] if is_single_state else action

Expand Down
3 changes: 2 additions & 1 deletion tf2rl/algos/vpg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,7 +93,8 @@ def get_action(self, state, test=False):
is_single_input = state.ndim == self._state_ndim
if is_single_input:
state = np.expand_dims(state, axis=0).astype(np.float32)
action, logp, _ = self._get_action_body(state, test)
action, logp, _ = self._get_action_body(tf.constant(state, dtype=tf.float32),
tf.constant(test))

if is_single_input:
return action.numpy()[0], logp.numpy()
Expand Down