CQL rllib 1.7.2 backport #170
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…proper train iteration size)
dmlyubim
force-pushed
the
dmlyubim/cql-1.7.2-port
branch
from
07481dc
to
8598a97
Compare
dmlyubim
force-pushed
the
dmlyubim/cql-1.7.2-port
branch
from
d09381a
to
acc2dde
Compare
dmlyubim
force-pushed
the
dmlyubim/cql-1.7.2-port
branch
from
acc2dde
to
e099a1d
Compare
The test passes for me in command line but fails in the pipeline where it fails to locate the json data file.
* set recursive mod 777 on /home/vsts/work/_temp/_bazel_vsts directory prior to build * use $TEST_TMPDIR env variable instead of literal directory name
…o dmlyubim/cql-1.7.2-port
dmlyubim
force-pushed
the
dmlyubim/cql-1.7.2-port
branch
from
15df96a
to
08bc679
Compare
dmlyubim
force-pushed
the
dmlyubim/cql-1.7.2-port
branch
from
08bc679
to
84bf8ae
Compare
This reverts commit 84bf8ae.
* set recursive mod 777 on /home/vsts/work/_temp/_bazel_vsts directory prior to build * use $TEST_TMPDIR env variable instead of literal directory name * explicitly set MACOSX_DEPLOYMENT_TARGET env variable * removed minor version of Python; renamed steps to relect correct Python version * get latest pip version to test MacOs wheels * updated hash * undid changes to info,yml * unbounded setuptools * undid change * Fix MacOs version if bdist_wheel generates incorrect MacOS version tag for wheel * undid changes * undid changes * undid changes * force reinstall tune and upstream requirements * updatd CI hash * updated dependencies * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated ci folder hash * updated requirements * updated requirements * updates CI hash * updated requirements * updated requirements * updated requirements * updated requirements * undid requirement changes * updated ci folder hash * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated dependencies * updated requirements * updated dependencies * apt update * fixed GCC download, set Ubuntu 20.04 as default OS for pipeline * updated requirements * updated requirements * fixed setup.py * updated ci hash * fixed setup.py * fixed setup.py * fixed setup.py * updated requirements * fixed setup.py * force reintall of torch and torchvision * updated ci hash * fixed rllib requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated requirements * updated dependencies * updated dependencies * updated requirements * updated requirements * updated requirements * explicitly set locale in MacOS to fix test_signal
abhiksingla
reviewed
Feb 8, 2023
| CQL_SAC = (cql.CQLSACTrainer, cql.CQLSAC_DEFAULT_CONFIG) | ||
| CQL_APEX_SAC = (cql.CQLApexSACTrainer, cql.CQLAPEXSAC_DEFAULT_CONFIG) | ||
| CQL_DQN = (cql.CQLDQNTrainer, cql.CQLDQN_DEFAULT_CONFIG) | ||
| CQL = (cql.CQLTrainer, cql.CQL_DEFAULT_CONFIG) |
There was a problem hiding this comment.
RLlib documentation mentions that CQL does not support discrete actions. Are we supporting discrete actions?
There was a problem hiding this comment.
i don't think so. this is backported code. I am not sure exactly how rllib hanldes that restriction, but we have ability to restrict it elsewhere in the outer code. I would not deviate from original rllib coding unless absolutely incorrect, makes further backporting merges easier.
| action_dist_class = _get_dist_class(policy, policy.config, | ||
| policy.action_space) | ||
| action_dist_class = _get_dist_class( | ||
| # policy, |
There was a problem hiding this comment.
Good to clean this to avoid confusion later.
| [cat.deterministic_sample() for cat in self.cats], axis=1) | ||
| if isinstance(self.action_space, gym.spaces.Box): |
There was a problem hiding this comment.
Given that it is categorical distribution and will be used for discrete action, is this statement valid?
There was a problem hiding this comment.
Also, it is not clear to me why extra dim is required for Box space only but not for others.
|
|
||
| @override(ActionDistribution) | ||
| def logp(self, actions: TensorType) -> TensorType: | ||
| # If tensor is provided, unstack it into list. | ||
| if isinstance(actions, tf.Tensor): | ||
| if isinstance(self.action_space, gym.spaces.Box): |
|
|
||
| @staticmethod | ||
| @override(ActionDistribution) | ||
| def required_model_output_shape( | ||
| action_space: gym.Space, | ||
| model_config: ModelConfigDict) -> Union[int, np.ndarray]: | ||
| return np.sum(action_space.nvec) | ||
| # Int Box. | ||
| if isinstance(action_space, gym.spaces.Box): |
| "requires": true, | ||
| "packages": { |
There was a problem hiding this comment.
I am not aware what this is. Ignoring it. Will suggest to get this reviewed by Ruofan or Kiko.
abhiksingla
approved these changes
Feb 8, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Why are these changes needed?
Related issue number
Checks
scripts/format.shto lint the changes in this PR.