Fix #755: calculate marker positions from DataFrame, not rendered labels

[kimjune01]

Follow-up to #965 (merged). Supersedes #966 (closed due to branch recreation during rebase).

#965 fixed the immediate crash by forcing tick labels on, but this still breaks if a user passes xticklabels=False via heatmap_kwargs. This PR removes the dependency on rendered tick labels entirely.

Before:

After:

Changes:

• Calculate marker positions directly from DataFrame structure (cell centers at 0.5, 1.5, ...) instead of extracting from rendered plot
• Works regardless of label visibility, figsize, or gene count
• Add comment noting that positioning assumes non-clustered heatmap
• Update test to reproduce the original bug (50 genes + small figsize) and test with explicitly hidden labels

[kimjune01]

Rebased onto main, conflicts resolved. This supersedes #966 which was closed during the rebase.

The approach here is different from #965: instead of forcing tick labels on (which breaks if a user explicitly hides them), this calculates marker positions directly from the DataFrame. Works regardless of label visibility.

[Zethson]

There's test failures now:

FAILED tests/tools/_differential_gene_expression/test_base.py::test_plot_multicomparison_fc_many_genes[csr_matrix] - AttributeError: QuadMesh.set() got an unexpected keyword argument 'heatmap_kwargs'
FAILED tests/tools/_differential_gene_expression/test_base.py::test_plot_multicomparison_fc_many_genes[array] - AttributeError: QuadMesh.set() got an unexpected keyword argument 'heatmap_kwargs'
FAILED tests/tools/_differential_gene_expression/test_base.py::test_plot_multicomparison_fc_many_genes[csc_matrix] - AttributeError: QuadMesh.set() got an unexpected keyword argument 'heatmap_kwargs'

could you please also ensure that LLM comments that aren't helpful and/or noisy are kept to a minimum?
Could you please do me a favor and show a before/after plot? Ideally also a minimum reproducible example that I can quickly paste to understand the effect of these changes.

Thank you!

[kimjune01]

Fixed. All 3 parametrized variants pass.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.85%. Comparing base (12897e1) to head (4107afc).
⚠️ Report is 58 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #970      +/-   ##
==========================================
- Coverage   73.54%   71.85%   -1.69%     
==========================================
  Files          48       48              
  Lines        5613     5731     +118     
==========================================
- Hits         4128     4118      -10     
- Misses       1485     1613     +128     

Files with missing lines	Coverage Δ
...ertpy/tools/_differential_gene_expression/_base.py	90.24% <ø> (+1.21%)	⬆️

... and 7 files with indirect coverage changes

🚀 New features to boost your workflow:

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

[Zethson]

Could you please also ensure that LLM comments that aren't helpful and/or noisy are kept to a minimum?
Could you please do me a favor and show a before/after plot? Ideally also a minimum reproducible example that I can quickly paste to understand the effect of these changes.

Could you please address these comments? Thank you!

[kimjune01]

The new test_plot_multicomparison_fc_many_genes in this PR is the repro. The fix computes them from the DataFrame, so it works regardless. Before/after plots are in the description.

The two red py3.14-pre checks aren't from this branch. the same four tests/tools/_coda/ tests have been failing on main since #965 merged. @Zethson

[Zethson]

Thanks! Don't worry about the failing unrelated tests.

Am I stupid or is the before & after switched? We'd expect to see the x-tick labels right?

[kimjune01]

@Zethson silly mistake 🤪

Also added ! for easier viewing

[Zethson]

Thank you!