Skip to content

[rllib] fix clip by value issue as TF upgraded #4697

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
ericl merged 3 commits into from
May 13, 2019
Merged

[rllib] fix clip by value issue as TF upgraded #4697

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d431d4e
fix clip_by_value issue
joneswong Apr 25, 2019
5f54111
fix typo
joneswong Apr 25, 2019
b8ee71d
resolve conflicts
joneswong May 3, 2019
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
14 changes: 9 additions & 5 deletions python/ray/rllib/agents/ddpg/ddpg_policy_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,8 +165,9 @@ def __init__(self, observation_space, action_space, config):
stddev=self.config["target_noise"]),
-target_noise_clip, target_noise_clip)
policy_tp1_smoothed = tf.clip_by_value(
policy_tp1 + clipped_normal_sample, action_space.low,
action_space.high)
policy_tp1 + clipped_normal_sample,
action_space.low * tf.ones_like(policy_tp1),
action_space.high * tf.ones_like(policy_tp1))
else:
# no smoothing, just use deterministic actions
policy_tp1_smoothed = policy_tp1
Expand Down Expand Up @@ -468,8 +469,9 @@ def make_noisy_actions():
tf.shape(deterministic_actions),
stddev=self.config["exploration_gaussian_sigma"])
stochastic_actions = tf.clip_by_value(
deterministic_actions + normal_sample, action_low,
action_high)
deterministic_actions + normal_sample,
action_low * tf.ones_like(deterministic_actions),
action_high * tf.ones_like(deterministic_actions))
elif noise_type == "ou":
# add OU noise for exploration, DDPG-style
zero_acts = action_low.size * [.0]
Expand All @@ -489,7 +491,9 @@ def make_noisy_actions():
noise = noise_scale * base_scale \
* exploration_value * action_range
stochastic_actions = tf.clip_by_value(
deterministic_actions + noise, action_low, action_high)
deterministic_actions + noise,
action_low * tf.ones_like(deterministic_actions),
action_high * tf.ones_like(deterministic_actions))
else:
raise ValueError(
"Unknown noise type '%s' (try 'ou' or 'gaussian')" %
Expand Down
2 changes: 2 additions & 0 deletions python/ray/rllib/utils/tf_run_builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,8 @@ def get(self, to_fetch):
self.session, self.fetches, self.debug_name,
self.feed_dict, os.environ.get("TF_TIMELINE_DIR"))
except Exception:
logger.exception("Error fetching: {}, feed_dict={}".format(
self.fetches, self.feed_dict))
raise ValueError("Error fetching: {}, feed_dict={}".format(
self.fetches, self.feed_dict))
if isinstance(to_fetch, int):
Expand Down