Do not include rvs in symbolic normalizing constant #7787
Merged
+29
−3
Conversation
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
There was a problem hiding this comment.
Pull Request Overview
This PR ensures that random variables (RVs) are not included in the symbolic normalizing constant graph by folding shapes to constants, adds shape inference for minibatch RVs, includes a test for the new behavior, and fixes a small typo.
- Use
constant_foldto derive batch shapes instead of carrying RVs intosymbolic_normalizing_constant - Implement
infer_shapeonMinibatchRandomVariableso shape propagation works correctly - Add a dedicated test (
assert_no_rvs) to confirm no RVs appear in the symbolic normalizing constant - Correct a typo in the
constant_foldcomment
Reviewed Changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| tests/variational/test_opvi.py | Added test_symbolic_normalizing_constant_no_rvs with assert_no_rvs |
| pymc/variational/opvi.py | Swapped direct .shape usage for constant_fold([...].shape) in scaling |
| pymc/variational/minibatch_rv.py | Added infer_shape method to propagate shapes without evaluation |
| pymc/pytensorf.py | Fixed typo in comment (constand_folding → constant_folding) |
Comments suppressed due to low confidence (3)
tests/variational/test_opvi.py:284
- [nitpick] The test verifies no RVs are in the graph but doesn't assert that the symbolic normalizing constant still produces the expected scalar or tensor shape. Consider adding an assertion on the returned value or shape to guard against regressions.
def test_symbolic_normalizing_constant_no_rvs():
pymc/variational/opvi.py:1109
- Calling
constant_foldinside the list comprehension for each RV will repeatedly clone and rewrite the graph, which may be costly. Consider computing all shapes once (e.g., collect inputs, callconstant_foldoutside the loop) or caching results before the comprehension.
get_scaling(
pymc/variational/opvi.py:1279
- This mirrored use of
constant_foldin another list comprehension also risks redundant graph rewriting. Extract a helper or hoist the folding step to improve efficiency and reduce duplicated logic.
get_scaling(
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #7787 +/- ##
=======================================
Coverage 92.84% 92.84%
=======================================
Files 107 107
Lines 18378 18380 +2
=======================================
+ Hits 17063 17065 +2
Misses 1315 1315
🚀 New features to boost your workflow:
|
jessegrabowski
approved these changes
May 17, 2025
ricardoV94
force-pushed
the
no_rvs_symbolic_normalizing_constant
branch
from
2551adf to
0867cde
Compare
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 is a more proper fix for the problem highlighted in #7778
The normalizing constant for MinibatchRVs included the graph of the shape of the RVs.
Even though the shape of the MinibatchRV can be derived without evaluating the draws, passing any graph with RVs to
pytensorf.compilewill automatically integrate the updates which requires evaluating the RV anyway. This PR makes sure we don't include the RVs only to get the symbolic normalizing constant.📚 Documentation preview 📚: https://pymc--7787.org.readthedocs.build/en/7787/