[raphaelvallat]

Summary

compare_channels compares every pair of channels by calling the public yasa.compare_detection, which checks, sorts and removes duplicate events on every call. The events of each channel are therefore re-checked and re-sorted 2 × (n_chan - 1) times. This PR splits compare_detection into three private steps (_check_parameters, _prepare_events and _compare_prepared) so that compare_channels prepares each channel once, then only runs the matching for each pair. The public function and its outputs are unchanged.

On 64 channels with 5,000 events each, compare_channels goes from 6.3 s to 2.4 s with method="distance" (4.8 s to 4.7 s with method="intersection"), with identical scores.

One small behaviour change: compare_channels now validates iou_threshold and max_distance_sec before comparing any pair of channels. Before, invalid values were silently ignored on a single-channel detection, because no comparison was run.

Stacked on #278, through a copy of its branch in this repo (feat/interval-iou-comparison). Once #278 is merged, this PR will be retargeted to master.

AI assistance

Model used: Claude Opus 5.5 via Claude Code

Checklist

• An issue was opened or discussed first (preferred)
• The branch is up to date with master, and I checked open PRs for conflicts
• The PR is small and focused, or stacked
• The description starts with my own summary
• Only necessary tests were added, and all tests pass
• The documentation is updated
• The changelog was not modified (it is updated by the maintainer before each release)

If AI-assisted:

• The exact model used is reported
• An AI review was run, the issues were fixed, and a summary was posted
• /simplify was run (preferred, Claude Code)

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.60%. Comparing base (0dbdd9a) to head (e48072f).

Additional details and impacted files

@@                      Coverage Diff                      @@
##           feat/interval-iou-comparison     #282   +/-   ##
=============================================================
  Coverage                         98.60%   98.60%           
=============================================================
  Files                                13       13           
  Lines                              3159     3165    +6     
  Branches                            429      431    +2     
=============================================================
+ Hits                               3115     3121    +6     
  Misses                               23       23           
  Partials                             21       21           

☔ 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.

[raphaelvallat]

AI review summary (/code-review with Claude Opus 5.5, on 784fe70)

No correctness issues found. Matching results are identical to compare_detection, which I confirmed on a 64-channel benchmark for both methods.

Fixed in e48072f:

• Channels were validated twice in compare_channels. Now _events_by_channel handles validation, and compare_channels only removes duplicate start times in distance mode.
• The method was checked twice per call. Each entry point now checks it once.
• _compare_prepared didn't say that it needs sorted, unique distance indices. This is now in its docstring.
• Invalid iou_threshold / max_distance_sec now raise even on single-channel detections. This is kept as a fix, tested, and noted in the description.

Not changed:

• The compare_detection method still uses the public function. Removing duplicates up front would change n_self / n_other, and that method only runs one comparison per channel.
• _match_intervals still calls argsort for every pair. The input is already sorted, so this costs about 0.08 s in total for 64 channels, which isn't worth a "sorted" flag.

🤖 Generated with Claude Code