Add logsigmoid (numerically stable) and softshrink #4663
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
dzhwinter
reviewed
Oct 10, 2017
paddle/operators/activation_op.h
Outdated
| @@ -95,6 +95,31 @@ struct SigmoidGradFunctor : public BaseActivationFunctor<T> { | |||
| } | |||
| }; | |||
|
|
|||
| // Originally: logsigmoid(x) = -log (1 + exp(-x)) | |||
| // For numerical stability: logsigmoid(x) = - (max(-x, 0) + log(exp(-max(-x, 0)) | |||
| // + exp(-x - max(-x, 0)))) | |||
There was a problem hiding this comment.
maybe we can put the source URL into the comment? (https://hips.seas.harvard.edu/blog/2013/01/09/computing-log-sum-exp/).
For the follow-up maintainers, he may not know the trick, he even cannot check the formula correctness.
There was a problem hiding this comment.
Thanks, I added more detailed comments. Please let me know if this seems okay now.
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Outdated suggestions cannot be applied.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 closes #4622, by adding 2 basic activation functions: logsigmoid and softshrink
However, to make the computation numerically stable for large negative values of x, the well-known "log-sum-exp" trick is employed. (https://hips.seas.harvard.edu/blog/2013/01/09/computing-log-sum-exp/).