Optional BatchNorm integration in NatureCNN #2132
Open
Conversation
This file contains hidden or 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
added 2 commits
May 3, 2025 17:17
When enabled via , this stabilizes feature distributions and reduces internal covariate shift. On Pong, it boosts avg. reward by ~2.3 points at 200k timesteps vs. the default extractor, with all existing tests still passing.
When enabled via , this stabilizes feature distributions and reduces internal covariate shift. On Pong, it boosts avg. reward by ~2.3 points at 200k timesteps vs. the default extractor, with all existing tests still passing.
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Outdated suggestions cannot be applied.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
Added an optional
BatchNormintegration to theNatureCNNarchitecture used in the feature extractor module of Stable-Baselines3. This enhancement introduces ause_batch_normflag to toggle Batch Normalization after each convolutional layer. This change provides a performance and stability improvement option for image-based environments.Motivation and Context
This change will solve the exploding gradients problem and in case it is set to False it does not change anything, however if set to True it will help converge much faster and enables us to use higher learning rates.
N/A N/AFurther than that this change allows users to optionally enable Batch Normalization in NatureCNN, which can improve training stability and convergence, especially in environments with high variance in pixel input. I initially explored alternatives (LayerNorm, GroupNorm). BatchNorm showed the best trade-off of speed and stability and convergence.
Types of changes
Checklist
make format(required)make check-codestyleandmake lint(required)make pytestandmake typeboth pass. (required)make doc(required)Note: You can run most of the checks using
make commit-checks.Note: we are using a maximum length of 127 characters per line