Fix shape errors in scalar_solve_to_division rewrite

[ricardoV94]

This showed up in the PyMC tests. There were some mistakes with the alignment of a, b batch and core dimensions.

The tests are failing for CholeskySolve, it seems to come back with flipped signs. Is this expected?

---

📚 Documentation preview 📚: https://pytensor--1516.org.readthedocs.build/en/1516/

[ricardoV94]

If a had shape (1, 1, 1) and b had shape (5, 1) and b_ndim=1 (meaning 5 is a batch dimension for b), we were returning something with the wrong shape as it became (1, 5, 1), after broadcasting with A, and then (1, 5) after squeezing below

[jessegrabowski]

oops :D

[jessegrabowski]

I approved too quickly. Regarding the flipped signs, I need to look more closely. There might be sign ambiguity due to the squared square root, or there might be a bug.

[ricardoV94]

Tests are not passing so no accident of merging

[jessegrabowski]

I think the problem might be the test. Numerical cholesky routines always compute such that the main diagonal is positive, which resolves any sign ambiguity in cho_solve. Here we're drawing the test value from a random normal, so there is sign ambiguity in the test.

I think if you square a_val, at least in the cho_solve case, tests will pass.

[ricardoV94]

Squaring worked like a charm

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.03%. Comparing base (2ab60b3) to head (e8a0b0b).
⚠️ Report is 118 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1516      +/-   ##
==========================================
+ Coverage   81.98%   82.03%   +0.05%     
==========================================
  Files         231      231              
  Lines       52274    52350      +76     
  Branches     9206     9215       +9     
==========================================
+ Hits        42856    42946      +90     
+ Misses       7106     7095      -11     
+ Partials     2312     2309       -3     

Files with missing lines	Coverage Δ
pytensor/tensor/rewriting/linalg.py	92.08% <100.00%> (+0.03%)	⬆️

... and 11 files with indirect coverage changes

🚀 New features to boost your workflow:

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