Fixes calculation of accuracy for the signal game with GumbelSoftmax #257
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Outdated suggestions cannot be applied.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
Fix the calculation of the accuracy when running the signal game with GumbelSoftmax.
Here, the mean of the accuracies for the batch was calculated already in the loss function, although it should be only calculated in the
forward
call of the game.Related Issue (if any)
Issue #256
Motivation and Context
Signal game can be run using GumbelSoftmax (see Issue #256)
How Has This Been Tested?
Error is fixed and the agents are trained on the data, using GumbelSoftmax.