Generic cdf and icdf implementation for scalar TransformedDistributions.#1853
Merged
fehiepsi merged 1 commit intopyro-ppl:masterfrom Aug 26, 2024
Merged
Conversation
…mplementation for `TransformedDistribution`s.
tillahoffmann
force-pushed
the
transformed-icdf
branch
from
90d8286 to
099e6cb
Compare
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 PR
signattribute toTransformwhich is the sign of the derivative for bijective scalar transforms. This matches thesignattribute of transforms in Pyro.cdfforTransformedDistributionby transforming from the domain of the transformed distribution to the domain of the base distribution and evaluating the cdf.icdfforTransformedDistributionby callingicdfof the base distribution and applying the transforms.This means cdfs for
LogNormal,InverseGamma,LogUniform, andParetodo not require a custom implementation. As a side effect, this PR addsicdfsupport forLogNormal,InverseGamma, andLogUniform; it also addscdfandicdfsupport forRelaxedBernoulliLogits.