Add function for calculating quantiles of weighed samples. #3340
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
BenZickel
force-pushed
the
add_weighed_quantile
branch
from
d0aa2af to
fc00066
Compare
added 2 commits
March 17, 2024 12:50
|
@fehiepsi, I've fixed the testing errors and now |
fritzo
requested changes
Mar 17, 2024
There was a problem hiding this comment.
Hi @BenZickel looks great! I wish we had designed Pyro for more native support of weighted sets of samples, but this is a step in the right direction.
Could you add a simple unit test of this function, similar to test_quantiles() in tests/ops/test_stats.py?
fritzo
requested changes
Mar 17, 2024
fritzo
approved these changes
Mar 17, 2024
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Outdated suggestions cannot be applied.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 Problem
When calculating distribution quantiles based on samples sampled from a guide it is often desirable to correct for the non-uniform gap between the model log-probability and the guide log-probability (this gap is already computed by the
pyro.infer.importance.Importancemodule).Currently there is only support for the below:
pyro.ops.stats.quantilewhich interpolates quantiles over the sorted samples.pyro.distributions.empirical.Empiricalwhich does not support interpolation.The Proposed Solution
Add a function
pyro.ops.stats.weighed_quantilewhich supports calculating interpolated quantiles with non-uniform weights. The implementation is similar topyro.ops.stats.quantilebut with different weights generated for each index of the input, according to the indices that sort that specific index.