Fix limit of log1mexp gradient at zero and improve numerical stablity
#725
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
ricardoV94
force-pushed
the
fix_log1mexp_grad
branch
from
ffa4128
to
2f6c33d
Compare
Codecov Report
@@ Coverage Diff @@
## main #725 +/- ##
=======================================
Coverage 78.17% 78.17%
=======================================
Files 152 152
Lines 47716 47718 +2
Branches 10893 10893
=======================================
+ Hits 37300 37302 +2
Misses 7848 7848
Partials 2568 2568
|
ricardoV94
changed the title
log1mexp gradient at zero and improve numerical precisionlog1mexp gradient at zero and improve numerical stablity
brandonwillard
added
bug
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Outdated suggestions cannot be applied.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.
The grad of
log1mexpwas evaluating to+inffor inputs of0, due to specific number/order of operations and the fact that1/-0.0 == -np.infvs1/0.0 == np.inf.This PR fixes this edge case and also improves numerical stability for values close to zero, relative to the previous implementation and relative to the gradient obtained from the equivalent symbolic graph
at.log(1-at.exp(x))This showed up here first: pymc-devs/pymc#5289