Fixed Jacobian computation of standardize transform #471
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Outdated suggestions cannot be applied.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.
I believe the Jacobian computation of standardize should have reversed sign.
y = (x - mu) / sigma
log p(y) = log p(x) - log(sigma)
Calling the adapter in forward direction should substract log(sigma), while inverse direction should add log(sigma).
Current implementation is switched. The PR fixes that.
BTW: It's really nice that Jacobian tracking adapter exists now in BF, thanks so much!