Fix KNN Regressor weighted prediction

[andrewdalpino]

src/Regressors/KNNRegressor.php:211-224 (root cause :232-246) — HIGH In weighted mode, predictSample builds $weights in distance order but array_values($labels) in original-index order, and Stats::weightedMean pairs them positionally — so each label is multiplied by the weight of a different neighbor.

Verified: train [[1],[2],[3]], labels [0,10,30], k=3 weighted, predict 2.0 → returns 10.0, correct is 12.5.
Verified: [[1],[5],[10]], [0,5,100], k=3 weighted, predict 10 → 8.55, correct is 79.60.
The classifier sibling KNearestNeighbors.php:229-233 does this correctly (aligns by key: $weights[$labels[$i]] += ...).
Fix: build weights keyed by neighbor index and pass $labels + $weights un-reordered.

[Copilot]

Pull request overview

This PR fixes incorrect weighted predictions in KNNRegressor by ensuring the per-neighbor weights remain aligned with the corresponding neighbor labels when distances are sorted by proximity.

Changes:

• Key the computed $weights by the original neighbor index (as preserved by asort(...)+array_slice(..., true)), instead of building a reindexed list.
• Pass $labels to Stats::weightedMean() without reindexing, so weights/labels are paired by key rather than by position.
• Add regression tests covering the misalignment case, a boundary/edge case, and correct enforcement of the k limit in weighted mode.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/Regressors/KNNRegressor.php	Fixes weighted prediction by aligning weights with label keys preserved through distance sorting/slicing.
tests/Regressors/KNNRegressorTest.php	Adds regression tests to ensure weighted predictions are correct and stable for key alignment + k limiting.

---

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

[apphp]

LGTM