feat: evaluation metrics

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

tests/eval/notes.test.ts

@@ -0,0 +1,37 @@
+import { expect, test } from "bun:test";
+import { matchNotes, maxAssignment } from "../../src/eval/notes.ts";
+import type { TimedNote } from "../../src/core/types.ts";
+
+const n = (onset: number, pitch: number, string?: number): TimedNote => ({ onset, offset: onset + 0.2, pitch, string });
+const ref = [n(1.0, 60), n(1.5, 62), n(2.0, 64)];
+
+test("tp / octave errors", () => {
+  const est = [n(1.03, 60), n(1.5, 74), n(2.2, 64)];
+  expect(matchNotes(ref, est).tp).toBe(1);
+  expect(matchNotes(ref, est).octaveErrors).toBe(1);
+});
+test("perfect and empty cases", () => {
+  expect(matchNotes(ref, ref).f1).toBe(1);
+  expect(matchNotes([], []).f1).toBeNull();
+  expect(matchNotes(ref, []).recall).toBe(0);
+  expect(matchNotes(ref, []).precision).toBeNull();
+  expect(matchNotes([], ref).recall).toBeNull();
+  expect(matchNotes([], ref).f1).toBe(0);
+});
+test("matching is maximum, not greedy", () => {
+  // greedy nearest would pair r1-e1 (0.00) leaving r2 unmatched; optimum pairs r1-e2, r2-e1
+  const r = [n(1.0, 60), n(1.04, 60)];
+  const e = [n(1.0, 60), n(0.96, 60)];
+  expect(matchNotes(r, e, { onsetTol: 0.045 }).tp).toBe(2);
+  const r2 = [n(0.0, 60), n(0.08, 60)];
+  const e2 = [n(0.04, 60), n(-0.01, 60)];
+  expect(matchNotes(r2, e2, { onsetTol: 0.05 }).tp).toBe(2);
+});
+test("pitchString key", () => {
+  expect(matchNotes([n(1, 60, 1)], [n(1, 60, 2)], { key: "pitchString" }).tp).toBe(0);
+  expect(matchNotes([n(1, 60, 1)], [n(1, 60, 2)]).tp).toBe(1);
+});
+test("maxAssignment", () => {
+  expect(maxAssignment([[5, 4], [4, 0]])).toEqual([1, 0]);
+  expect(maxAssignment([[3], [5]])).toEqual([-1, 0]);
+});