Speed up test_weakref_leak and test_step_args

[mcwanza]

Addresses #7686 — profiling-backed optimizations targeting tests where iteration reduction yields meaningful savings.

Changes

test_weakref_leak (89s → ~30s estimated on CI)

Object count profiling shows memory state stabilizes at iteration 2:

Iteration | Growing objects | Details
    0     | 10 growing      | function:+67554, tuple:+50400, dict:+30266
    1     |  1 growing      | list:+1
    2     |  0 growing      | STABLE
    3     |  0 growing      | STABLE
   ...    |  0 growing      | STABLE (all remaining iterations)

The original test ran 20 iterations but only checked after iteration 15 — 16 warmup iterations when 3 is sufficient. Reduced to 6 total (3 warmup + 3 check). Each conditional_logp call costs ~4.5s on CI, so removing 14 iterations saves ~63s.

test_step_args (62s → ~25s estimated on CI)

This test verifies target_accept argument plumbing, checking acceptance_rate.mean() ≈ 0.5 with decimal=1 precision. The default pm.sample() uses 1000 draws/tune, but 200 is more than sufficient for this loose tolerance (verified stable over 10 runs with different seeds).

Profiling: first pm.sample() with 1000 draws takes 2.20s locally, with 200 draws takes 0.29s. The test calls pm.sample() 5 times.

Changes NOT made (profiling showed negligible impact)

Test	CI time	Bottleneck	Iteration savings
test_default_value_transform_logprob	35s	Compile: 8.85s, Loop(10): 0.0014s	0.6ms total
TestMatchesScipy::test_interpolated	33s	Compile: 91.8%, Array: 7.9%	~296ms across 56 iters

Compilation dominates these tests — reducing iteration counts has near-zero impact, consistent with @ricardoV94's review feedback.

[ricardoV94]

Some changes are fine, other sacrifice too much on the tests. Left some commenst

[ricardoV94]

I don't think we should remove any of the parametrizations in test_multivariate. If these are slow the best thing is to analyze why instead

[mcwanza]

Reverted — all test_multivariate.py changes have been removed from this PR.

[mcwanza]

Reworked the entire PR based on this feedback. Profiled all the slow tests to understand where time is actually spent — you were right that compilation dominates.

Removed all the original changes (parametrization reduction, iteration count tweaks on compile-dominated tests). The PR now only contains 2 profiling-backed changes:

1. test_weakref_leak (20→6 iterations): Object count profiling shows state stabilizes at iteration 2 — the original 16 warmup iterations were overkill. Each conditional_logp call costs ~4.5s on CI, so this saves ~63s.

2. test_step_args (explicit draws=200, tune=200): This test checks acceptance_rate.mean() ≈ 0.5 with decimal=1 precision — 200 draws is sufficient. The default 1000 was unnecessary for this tolerance level.

[ricardoV94]

I would be very surprised if this played a role. The cost will be mostly compile time, not running 5 +- steps. Let me know if I am wrong

[mcwanza]

Reverted — the progress bar change has been removed from this PR.

[mcwanza]

Confirmed by profiling. The test_default_value_transform_logprob loop takes 0.0014s total for 10 iterations while compilation takes 8.85s — compile/loop ratio of 6480x. Removed this change from the PR.

[mcwanza]

Hi @ricardoV94 — friendly ping on this one. I addressed your feedback in the Feb 17 push (profiled compile vs. iteration cost, kept the parametrizations, justified each change in the PR description). Happy to make further changes if anything still looks off — just let me know.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[read-the-docs-community]

Documentation build overview

📚 pymc | 🛠️ Build #32735835 | 📁 Comparing 236a227 against latest (a65e8b6)

  🔍 Preview build  

2 files changed

± glossary.html
± contributing/jupyter_style.html

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.76%. Comparing base (169e901) to head (236a227).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8107      +/-   ##
==========================================
- Coverage   91.78%   91.76%   -0.02%     
==========================================
  Files         125      125              
  Lines       20428    20428              
==========================================
- Hits        18749    18745       -4     
- Misses       1679     1683       +4     

see 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

[welcome]

]
Congrats on merging your first pull request! 🎉 We here at PyMC are proud of you! 💖 Thank you so much for your contribution 🎁

[ricardoV94]

Thanks!