Match T4/T5 tuning metrics to ground truth by cell type

[marisbasha]

Summary

correlation_to_known_tuning_curves and angular_distance_to_known compare each T4/T5 cell with the known tuning by position rather than by cell type. time_window also fails on a floating point comparison for some edge offsets.

Changes

Tuning curve correlation

The selected T4/T5 cells were correlated with the known tuning curves in the fixed order T4a..T5d. The data keeps its own neuron order, so any other order (a custom connectome, a cell_index subset, isel or sortby) correlated a cell with another type's curve. The ground truth is now built in the order of the selected cells. With the default connectome the result is unchanged.

Preferred direction distance

angular_distance_to_known zipped the selected cells with a fixed list of four angles, with the same assumption. It now looks up each cell type in groundtruth_utils.preferred_directions, which holds the same angles.

time_window assertion

time_window asserts abs(end_in_columns) >= abs(to_column). With the default to_degree of peak_responses, (offsets[1] - 1) * 2.25, both sides are equal in exact arithmetic but can differ in the last bit. So peak_responses, direction_selectivity_index and preferred_direction raised a bare AssertionError for edge offsets ending at 12, 16 or 19, such as (-10, 12), (-11, 12) or (-12, 16). The default offsets (-10, 11) are not affected. The difference is at most 9e-16, and the comparison now allows 1e-9.

Testing

New tests/test_moving_bar_responses.py builds a synthetic moving-edge dataset in which the eight central T4/T5 cells respond with their published tuning curves. Both metrics are checked on the original neuron order and with T4a and T4b swapped, and peak_responses is run for the three offsets above. All five tests fail on main (correlation -0.4 instead of 1 for the swapped cells, angular distances near pi, AssertionError) and pass with the change.

I also scored the moving edge responses of the pretrained model flow/0000/000. With the default neuron order, correlation_to_known_tuning_curves, angular_distance_to_known and direction_selectivity_index are bit-identical to main. With the neuron order reversed, the branch gives the same value for each cell type. Main returns the correlations without the cell_type coordinate, each against another type's curve: for T4a at ON edges, -0.24 instead of 0.95.

Full suite (-m "not require_download and not require_large_download and not gpu", test_examples.py and test_sintel.py excluded): 196 passed, 12 skipped on main; 201 passed, 12 skipped on the branch. require_download tests against the pretrained models: 21 passed on both. tests/test_sintel.py, run offline with the mock data: 14 passed on both. ruff check and ruff format (0.5.5) are clean on the changed files.

[Copilot]

Copilot review overview

🟢 Approval recommended

The targeted fixes are consistent with existing data structures and are adequately covered by regression tests.

Review effort: Balanced
Findings: None

What changed in this PR

Aligns T4/T5 tuning metrics with cell-type coordinates and tolerates harmless floating-point error in time-window bounds.

Changes:

• Builds ground-truth curves in selected neuron order.
• Looks up preferred directions by cell type.
• Adds regression tests for reordered neurons and edge offsets.

File	Description
flyvis/​analysis/​moving_bar_responses.py	Fixes metric alignment and time-window tolerance.
tests/​test_moving_bar_responses.py	Adds synthetic regression coverage.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 39.93%. Comparing base (92b3845) to head (e93f038).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #28      +/-   ##
==========================================
+ Coverage   38.68%   39.93%   +1.24%     
==========================================
  Files          75       75              
  Lines        9738     9739       +1     
==========================================
+ Hits         3767     3889     +122     
+ Misses       5971     5850     -121     

Flag	Coverage Δ
unittests	39.93% <100.00%> (+1.24%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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