Implement specialized Hurdle distribution #7810

ricardoV94 · 2025-06-04T12:16:21Z

It indirectly addresses the issue reported in in pymc-devs/nutpie#163

The new objects have a logp that handles the discrete + continuous process correctly, without requiring the arbitrary truncation of the latter at epsilon. This provides a cheaper and more stable logp / logcdf.
For discrete variables we keep using a truncation

~~Also added special logic to truncate a Hurdle distribution which solves bambinos/bambi#768~~, this is not the desired behavior, reverted it

CC @zwelitunyiswa

📚 Documentation preview 📚: https://pymc--7810.org.readthedocs.build/en/7810/

ricardoV94 · 2025-06-04T15:05:32Z

tests/distributions/test_distribution.py

+        dist
+        for dist in dists
+        if (
+            getattr(dist, "rv_type", None) is not None


This was too restrictive, a subclass also inherits the dispatch function, and need not be in the registry explicitly

codecov · 2025-06-04T15:32:02Z

Codecov Report

Attention: Patch coverage is 91.42857% with 6 lines in your changes missing coverage. Please review.

Project coverage is 92.88%. Comparing base (3b62f82) to head (1e38719).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
pymc/distributions/mixture.py	91.17%	6 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #7810   +/-   ##
=======================================
  Coverage   92.88%   92.88%           
=======================================
  Files         107      107           
  Lines       18377    18389   +12     
=======================================
+ Hits        17069    17081   +12     
  Misses       1308     1308

Files with missing lines	Coverage Δ
pymc/distributions/moments/means.py	`100.00% <100.00%> (ø)`
pymc/distributions/mixture.py	`95.25% <91.17%> (+0.23%)`	⬆️

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

pymc/distributions/truncated.py

zwelitunyiswa · 2025-06-04T16:07:49Z

It indirectly addresses the issue reported in in pymc-devs/nutpie#163

The new objects have a logp that handles the discrete + continuous process correctly, without requiring the arbitrary truncation of the latter at epsilon. This provides a cheaper and more stable logp / logcdf. For discrete variables we keep using a truncation

Also added special logic to truncate a Hurdle distribution which solves bambinos/bambi#768

CC @zwelitunyiswa

📚 Documentation preview 📚: https://pymc--7810.org.readthedocs.build/en/7810/

@ricardoV94 This is amazing. Thank you so much for this!

zaxtax

LGTM! We should probably add a test similar to one that motivate this PR in the first class

ricardoV94 · 2025-06-05T20:21:54Z

LGTM! We should probably add a test similar to one that motivate this PR in the first class

We have the pre-existing hurdlesl tests, in a sense this is just a refactor/optimization. Can't think of anything ~~reasonable~~ obvious to test here?

zaxtax · 2025-06-05T20:25:08Z

LGTM! We should probably add a test similar to one that motivate this PR in the first class

We have the pre-existing hurdlesl tests, in a sense this is just a refactor/optimization. Can't think of anything ~~reasonable~~ obvious to test here?

What caused the error originally reported here? pymc-devs/nutpie#163 Does that have a test already?

ricardoV94 · 2025-06-05T20:29:23Z

LGTM! We should probably add a test similar to one that motivate this PR in the first class

We have the pre-existing hurdlesl tests, in a sense this is just a refactor/optimization. Can't think of anything ~~reasonable~~ obvious to test here?

What caused the error originally reported here? pymc-devs/nutpie#163 Does that have a test already?

That was fixed sometime ago in PyTensor: pymc-devs/pytensor#1137

The performance question when in numba is addressed by pymc-devs/pytensor#1445

Neither is PyMC specific

zaxtax · 2025-06-06T07:39:58Z

LGTM! We should probably add a test similar to one that motivate this PR in the first class

We have the pre-existing hurdlesl tests, in a sense this is just a refactor/optimization. Can't think of anything ~~reasonable~~ obvious to test here?

What caused the error originally reported here? pymc-devs/nutpie#163 Does that have a test already?

That was fixed sometime ago in PyTensor: pymc-devs/pytensor#1137

The performance question when in numba is addressed by pymc-devs/pytensor#1445

Neither is PyMC specific

Feel free to merge whenever you feel comfortable! I think it's good to go

lucianopaz

Looks good to me. I just left two very minor comments

pymc/distributions/mixture.py

lucianopaz · 2025-06-06T07:44:54Z

pymc/distributions/mixture.py

-    )
-
-    return mix_logp
+    mix_support_point = pt.sum(weights * support_point_components, axis=mix_axis)


Why not use the logsumexp and have log scale weights here? Is it because the weights are already in the 0-1 range and taking the log won’t help with precision?

We're not computing any log quantities nor starting with any log quantities so I don't think it would help. Also the initial point is not so critical?

This does not require the arbitrary truncation of continuous distribution in the logp/logcdf

ricardoV94 requested a review from tomicapretto June 4, 2025 12:16

ricardoV94 force-pushed the hurdle_mixtures branch 2 times, most recently from 77ed668 to 9c65ca1 Compare June 4, 2025 14:02

ricardoV94 changed the title ~~Implement specialized Hurdle distribution~~ Implement specialized Hurdle distribution and allow truncating it Jun 4, 2025

ricardoV94 added the enhancements label Jun 4, 2025

ricardoV94 force-pushed the hurdle_mixtures branch 3 times, most recently from 3d5772c to bdb3f12 Compare June 4, 2025 14:15

ricardoV94 requested a review from lucianopaz June 4, 2025 14:15

ricardoV94 force-pushed the hurdle_mixtures branch 2 times, most recently from ac73b55 to 9a65487 Compare June 4, 2025 14:52

pymc-devs deleted a comment from review-notebook-app bot Jun 4, 2025

ricardoV94 force-pushed the hurdle_mixtures branch from 9a65487 to 0917343 Compare June 4, 2025 15:04

ricardoV94 commented Jun 4, 2025

View reviewed changes

pymc/distributions/truncated.py Outdated Show resolved Hide resolved

ricardoV94 marked this pull request as draft June 4, 2025 15:59

ricardoV94 force-pushed the hurdle_mixtures branch from 0917343 to 3f7efa1 Compare June 5, 2025 08:21

ricardoV94 changed the title ~~Implement specialized Hurdle distribution and allow truncating it~~ Implement specialized Hurdle distribution Jun 5, 2025

ricardoV94 marked this pull request as ready for review June 5, 2025 12:34

zaxtax approved these changes Jun 5, 2025

View reviewed changes

lucianopaz approved these changes Jun 6, 2025

View reviewed changes

ErikRingen mentioned this pull request Jun 13, 2025

BUG: HurdleGamma results in large number of divergences, even under the correct model #7630

Closed

ricardoV94 force-pushed the hurdle_mixtures branch from 3944eb3 to 3e0ab8c Compare June 16, 2025 08:20

Implement specialized Hurdle class

1e38719

This does not require the arbitrary truncation of continuous distribution in the logp/logcdf

ricardoV94 force-pushed the hurdle_mixtures branch from 3e0ab8c to 1e38719 Compare June 16, 2025 20:00

ricardoV94 merged commit 0f1bfa9 into pymc-devs:main Jun 16, 2025
40 of 42 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Implement specialized Hurdle distribution #7810

Implement specialized Hurdle distribution #7810

Uh oh!

ricardoV94 commented Jun 4, 2025 •

edited

Loading

Uh oh!

ricardoV94 Jun 4, 2025

Uh oh!

codecov bot commented Jun 4, 2025 •

edited

Loading

Uh oh!

Uh oh!

zwelitunyiswa commented Jun 4, 2025

Uh oh!

zaxtax left a comment

Uh oh!

ricardoV94 commented Jun 5, 2025 •

edited

Loading

Uh oh!

zaxtax commented Jun 5, 2025

Uh oh!

ricardoV94 commented Jun 5, 2025

Uh oh!

zaxtax commented Jun 6, 2025

Uh oh!

lucianopaz left a comment

Uh oh!

Uh oh!

lucianopaz Jun 6, 2025

Uh oh!

ricardoV94 Jun 6, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Implement specialized Hurdle distribution #7810

Implement specialized Hurdle distribution #7810

Uh oh!

Conversation

ricardoV94 commented Jun 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ricardoV94 Jun 4, 2025

Choose a reason for hiding this comment

Uh oh!

codecov bot commented Jun 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

zwelitunyiswa commented Jun 4, 2025

Uh oh!

zaxtax left a comment

Choose a reason for hiding this comment

Uh oh!

ricardoV94 commented Jun 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

zaxtax commented Jun 5, 2025

Uh oh!

ricardoV94 commented Jun 5, 2025

Uh oh!

zaxtax commented Jun 6, 2025

Uh oh!

lucianopaz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

lucianopaz Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

ricardoV94 Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

ricardoV94 commented Jun 4, 2025 •

edited

Loading

codecov bot commented Jun 4, 2025 •

edited

Loading

ricardoV94 commented Jun 5, 2025 •

edited

Loading