Skip to content
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.

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

[rllib] Revert [rllib] Port DDPG to the build_tf_policy pattern #5626

Merged
merged 8 commits into from
Sep 5, 2019

Conversation

ericl
Copy link
Contributor

@ericl ericl commented Sep 3, 2019

This reverts #5242 to fix #5604

Two issues: (1) blowup of Q-values during training of MountainCar (and also Pendulum if you use higher tau values), (2) eager mode acts differently from graph mode somehow.

These are likely due to a subtle bug in the port to the build_tf_policy pattern. We should re-visit the port, but this PR reverts it quickly for 0.7.4.

@ericl
Copy link
Contributor Author

ericl commented Sep 3, 2019

@pcmoritz can we also cherry pick this into the 0.7.4. branch?

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16753/
Test FAILed.

@pcmoritz
Copy link
Contributor

pcmoritz commented Sep 3, 2019

Sounds good, yeah! Shall we also merge it into master?

@ericl
Copy link
Contributor Author

ericl commented Sep 3, 2019 via email

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16765/
Test FAILed.

@pcmoritz
Copy link
Contributor

pcmoritz commented Sep 4, 2019

There is another Jenkins error (it might involve copying a lot of code back, since it is shared by the SAC agent).

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16781/
Test FAILed.

@pcmoritz
Copy link
Contributor

pcmoritz commented Sep 4, 2019

Jenkins retest this please

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16785/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16790/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16791/
Test FAILed.

@pcmoritz
Copy link
Contributor

pcmoritz commented Sep 4, 2019

Jenkins retest this please

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16792/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/16793/
Test FAILed.

@pcmoritz pcmoritz merged commit 19bbf1e into ray-project:master Sep 5, 2019
@pcmoritz pcmoritz deleted the revert-ddpg-port branch September 5, 2019 04:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[rllib] mountaincarcontinous-ddpg regression
4 participants