This repository has been archived by the owner on May 6, 2021. It is now read-only.
Fix bug in multi action ppo #169
Merged
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
|
Feel free to merge it first. The CI error is caused by a breaking change in GridWorlds.jl |
Co-authored-by: Jun Tian <[email protected]>
|
Cool, will do that then. I might also try to add some environment that has a multidimensional actions space so we can have a test for the algorithms that handles that. But I'll leave that for some later time. |
|
What type of merge do you usually use? Merge commit, squash and merge or rebase and merge? |
|
In most cases, I'd squash and merge. |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.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.
As mentioned in JuliaReinforcementLearning/ReinforcementLearning.jl#234 we had some strange behaviour. This was from the log_p`_a having a different size than log_p, (1, N) compared to (N,), creating a matrix for the ratio which was then reduced down.
I also found that the entropy loss was not defined to work with multi dimensional actions correctly, it was missing a multiple of how many dimensions there were in the actions space. Then it was also missing a division by 2 in one of the terms compared to how the entropy is defined on https://en.wikipedia.org/wiki/Multivariate_normal_distribution#Differential_entropy