[Feature] Add scheduler for alpha/beta parameters of PrioritizedSampler #2452
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/2452
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 8 New Failures, 2 Unrelated FailuresAs of commit 4b2897a with merge base 33e86c5 ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
vmoens
reviewed
Sep 25, 2024
Comment on lines
+221
to
+224
| if self._step_cnt % self.n_steps == 0: | ||
| return self.operator(current_val, self.gamma) | ||
| else: | ||
| return current_val |
test/test_rb.py
Outdated
Comment on lines
3036
to
3041
| INIT_ALPHA = 0.7 | ||
| INIT_BETA = 0.6 | ||
| GAMMA = 0.1 | ||
| EVERY_N_STEPS = 10 | ||
| LINEAR_STEPS = 100 | ||
| TOTAL_STEPS = 200 |
There was a problem hiding this comment.
let's maybe make these args to the func?
test/test_rb.py
Outdated
Comment on lines
3059
to
3062
| expected_alpha_vals = np.linspace(INIT_ALPHA, 0.0, num=LINEAR_STEPS + 1) | ||
| expected_alpha_vals = np.pad( | ||
| expected_alpha_vals, (0, TOTAL_STEPS - LINEAR_STEPS), constant_values=0.0 | ||
| ) |
test/test_rb.py
Outdated
Comment on lines
3070
to
3075
| assert np.isclose( | ||
| rb.sampler.alpha, expected_alpha_vals[i] | ||
| ), f"expected {expected_alpha_vals[i]}, got {rb.sampler.alpha}" | ||
| assert np.isclose( | ||
| rb.sampler.beta, expected_beta_vals[i] | ||
| ), f"expected {expected_beta_vals[i]}, got {rb.sampler.beta}" |
There was a problem hiding this comment.
let's use torch.testing.assert_close
Co-authored-by: Vincent Moens <[email protected]>
vmoens
reviewed
Sep 30, 2024
| self.initial_val = getattr(self.sampler, self.param_name) | ||
| self._step_cnt = 0 | ||
|
|
||
| def state_dict(self): |
Comment on lines
149
to
153
| def _step(self): | ||
| if self._step_cnt < self.num_steps: | ||
| return self.initial_val + (self._delta * self._step_cnt) | ||
| else: | ||
| return self.final_val |
There was a problem hiding this comment.
yeah that's fine, maybe let's add a comment to let someone know in the future that this should be fixed
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.
Description
Add scheduler for alpha/beta parameters of PrioritizedSampler.
Motivation and Context
close #1575
Following the suggestions made by @vmoens in issue #1575, this PR adds different Scheduler classes through which the user can adjust the alpha and beta parameters of the
PrioritizedSamplerduring training when using thePrioritizedReplayBuffer. This is explicitly suggested in the paper "Schaul, T.; Quan, J.; Antonoglou, I.; and Silver, D. 2015. Prioritized experience replay".The main reason to use separate scheduler classes for the annealing instead of a simple linear annealing (as also suggested by @vmoens in issue #1575) is the greater flexibility for the users. This way, the annealing can take place for example after taking a sample from the replay buffer or after a full training epoch (depending on where the user places the
scheduler.step()command). Also, through theLinearScheduler,StepSchedulerandLambdaSchedulerdifferent annealing schemes can be used (or new ones can be easily created).Types of changes
What types of changes does your code introduce? Remove all that do not apply:
Checklist
Go over all the following points, and put an
xin all the boxes that apply.If you are unsure about any of these, don't hesitate to ask. We are here to help!