[RLlib] POC: Run RLlib w/o Preprocessors setup. #17656
Closed
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
…le_batch_supports_complex_spaces
…eprocessors_soft # Conflicts: # rllib/policy/sample_batch.py
…le_batch_supports_complex_spaces
…le_batch_supports_complex_spaces # Conflicts: # rllib/policy/sample_batch.py
…le_batch_supports_complex_spaces
…le_batch_supports_complex_spaces
…le_batch_supports_complex_spaces
…eprocessors_soft # Conflicts: # rllib/agents/trainer.py # rllib/utils/annotations.py
…ecate_preprocessors_soft # Conflicts: # rllib/agents/trainer.py
…ecate_preprocessors_soft # Conflicts: # rllib/evaluation/collectors/simple_list_collector.py
…ecate_preprocessors_soft
…ecate_preprocessors_soft # Conflicts: # rllib/evaluation/collectors/simple_list_collector.py
sven1977
commented
Aug 21, 2021
rllib/execution/replay_buffer.py
Outdated
| @@ -402,7 +402,7 @@ def add_batch(self, batch: SampleBatchType) -> None: | |||
| # If SampleBatch has prio-replay weights, average | |||
| # over these to use as a weight for the entire | |||
| # sequence. | |||
| if "weights" in time_slice: | |||
| if "weights" in time_slice and time_slice["weights"]: | |||
There was a problem hiding this comment.
Avoid np.mean over empty list (yields: NaN).
…ecate_preprocessors_soft
…ecate_preprocessors_soft
sven1977
added
the
tests-ok
| @@ -416,6 +416,8 @@ def postprocess_nstep_and_prio(policy: Policy, | |||
| batch[SampleBatch.REWARDS], batch[SampleBatch.NEXT_OBS], | |||
| batch[SampleBatch.DONES]) | |||
|
|
|||
| # Create dummy prio-weights (1.0) in case we don't have any in | |||
There was a problem hiding this comment.
Prob should be in diff PR
| # env's action space before sending actions back to the env. | ||
| # (0.0 centered with small stddev; only affecting Box components). | ||
| # We will unsquash actions (and clip, just in case) to the bounds of | ||
| # the env's action space before sending actions back to the env. | ||
| "normalize_actions": True, | ||
| # If True, RLlib will clip actions according to the env's bounds | ||
| # before sending them back to the env. | ||
| # TODO: (sven) This option should be obsoleted and always be False. | ||
| "clip_actions": False, | ||
| # Whether to use "rllib" or "deepmind" preprocessors by default |
There was a problem hiding this comment.
Maybe describe what rllib or deepmind does in the comment.
| self.buffers[SampleBatch.AGENT_INDEX][0].append(agent_index) | ||
| self.buffers[SampleBatch.ENV_ID][0].append(env_id) | ||
| self.buffers[SampleBatch.T][0].append(t) | ||
| self.buffers[SampleBatch.EPS_ID][0].append(self.episode_id) |
There was a problem hiding this comment.
If self.episode_id and unroll_id is constant, why repeat adding the same data?
| if SampleBatch.EPS_ID in values: | ||
| assert values[SampleBatch.EPS_ID] == self.episode_id | ||
| del values[SampleBatch.EPS_ID] | ||
| self.buffers[SampleBatch.EPS_ID][0].append(self.episode_id) |
| shape=shape, | ||
| name=".".join([str(p) for p in path]), | ||
| ) | ||
|
|
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.
This PR was motivated in preparation for soon allowing individual observation components to be addressed by the trajectory view API, for example to enable frame-stacking for individual observation components within a complex observation space (Tuple|Dict). Also, soon soft-deprecating RLlib's Preprocessor API should increase transparency for the users and allow batched, model-based preprocessing of observations. Observations will arrive in the model exactly as they are returned by the env.
This PR is a POC that works for tf and torch.
preprocessor_pref: None; Set to None for disabling Preprocessors altogether (not even useNoPreprocessorclass anymore).Why are these changes needed?
Related issue number
Checks
scripts/format.shto lint the changes in this PR.