// cmj-pca.jsx v1 — CMJ 向心曲线形态监控（PC3 趋势）
// 职责：从已存 trial.curve.f + trial.keyPts.e 提取向心段 → PCA → PC3 时序图
// 依赖：无。由 DSIPanel 传入 cmjSessions 数组。
// 注：每周监控场景下 PC3 反映跨-session 曲线形态一致性，非急性疲劳分类。
// 参考：Gathercole et al. 2019 PLOS ONE PMC6619745
(function () {
const { useState, useMemo } = React;
const N_INTERP = 50; // concentric curve interpolation length
const MIN_SESS = 3; // minimum sessions for PCA
// ── helpers ──────────────────────────────────────────────────────────────
function interp(arr, nOut) {
const n = arr.length;
if (n < 2) return Array(nOut).fill(arr[0] ?? 0);
return Array.from({ length: nOut }, (_, i) => {
const t = i / (nOut - 1) * (n - 1);
const lo = Math.floor(t);
const hi = Math.min(n - 1, lo + 1);
return arr[lo] + (arr[hi] - arr[lo]) * (t - lo);
});
}
function dot(a, b) { return a.reduce((s, x, i) => s + x * b[i], 0); }
function matvec(M, v) { return M.map(row => dot(row, v)); }
// Power iteration: returns dominant eigenvector of a symmetric matrix
function powerIter(M, iters = 120) {
const n = M.length;
// Deterministic seed: normalize first non-zero row
let v = M.find(row => row.some(Boolean)) || M[0];
v = v.slice();
const nm = Math.sqrt(dot(v, v)) || 1;
v = v.map(x => x / nm);
for (let k = 0; k < iters; k++) {
const Mv = matvec(M, v);
const len = Math.sqrt(dot(Mv, Mv)) || 1;
v = Mv.map(x => x / len);
}
return v;
}
// Deflate: remove one eigenvector component from symmetric matrix
function deflate(M, v) {
const lambda = dot(v, matvec(M, v));
return M.map((row, i) => row.map((x, j) => x - lambda * v[i] * v[j]));
}
// ── PCA computation ───────────────────────────────────────────────────────
// Returns { scores: [{date, pc3}], mean, sd } or null if data insufficient.
function computePCA(sessions) {
// 1. Extract concentric curves from each session's representative trial
const rows = [];
sessions.forEach(sess => {
const trial = sess.trials?.find(item =>
item && String(item.index) === String(sess.representative?.index)
) ?? sess.trials?.[0];
if (!trial?.curve?.f || !trial?.keyPts?.e) return;
const f = trial.curve.f;
const start = Math.round(trial.keyPts.e * (f.length - 1)); // zeroCross index
const conc = f.slice(start);
if (conc.length < 5) return;
rows.push({ date: sess.date, vec: interp(conc, N_INTERP) });
});
if (rows.length < MIN_SESS) return null;
const N = rows.length;
const D = N_INTERP;
// 2. Column-wise mean centering
const mean_col = Array(D).fill(0);
rows.forEach(r => r.vec.forEach((v, j) => { mean_col[j] += v / N; }));
const X = rows.map(r => r.vec.map((v, j) => v - mean_col[j]));
// 3. Covariance matrix D×D
const cov = Array.from({ length: D }, () => Array(D).fill(0));
X.forEach(row => {
for (let i = 0; i < D; i++)
for (let j = i; j < D; j++) {
const v = row[i] * row[j] / (N - 1);
cov[i][j] += v;
if (i !== j) cov[j][i] += v;
}
});
// 4. Extract PC1, PC2, PC3 via power iteration + deflation
const pc1 = powerIter(cov);
const cov2 = deflate(cov, pc1);
const pc2 = powerIter(cov2);
const cov3 = deflate(cov2, pc2);
const pc3 = powerIter(cov3);
// 5. Project onto PC3
const scores = rows.map((r, i) => ({ date: r.date, pc3: dot(X[i], pc3) }));
// 6. Global mean / SD for reference band
const mu = scores.reduce((s, r) => s + r.pc3, 0) / N;
const sd = Math.sqrt(scores.reduce((s, r) => s + (r.pc3 - mu) ** 2, 0) / N) || 1;
return { scores, mean: mu, sd };
}
// ── Chart component ───────────────────────────────────────────────────────
function CMJPCAChart({ cmjSessions }) {
const sorted = useMemo(
() => [...cmjSessions].sort((a, b) => a.date.localeCompare(b.date)),
[cmjSessions]
);
const result = useMemo(() => computePCA(sorted), [sorted]);
const [hover, setHover] = useState(null);
// Not enough data
if (!result) {
const validN = sorted.filter(s => s.trials?.[0]?.curve?.f).length;
return (
需至少 {MIN_SESS} 次含完整曲线的 CMJ session（MARS 汇总导入不计）。当前有效：{validN} 次。
);
}
const { scores, mean, sd } = result;
const N = scores.length;
const W = 640, H = 160, ML = 44, MR = 12, MT = 14, MB = 28;
const PW = W - ML - MR, PH = H - MT - MB;
// Visible range: mean ± 2.5 SD, but always include all actual scores
const allPc3 = scores.map(s => s.pc3);
const visLo = Math.min(mean - 2.5 * sd, Math.min(...allPc3));
const visHi = Math.max(mean + 2.5 * sd, Math.max(...allPc3));
const span = visHi - visLo || 1;
const xAt = i => ML + (N < 2 ? PW / 2 : i / (N - 1) * PW);
const yAt = v => MT + PH * (1 - (Math.max(visLo, Math.min(visHi, v)) - visLo) / span);
const dotColor = v => {
const z = Math.abs(v - mean) / sd;
return z > 1.5 ? 'var(--neg)' : z > 1.0 ? 'var(--warn)' : 'var(--pos)';
};
// Band y-coords
const yMean = yAt(mean);
const y1hi = yAt(mean + sd), y1lo = yAt(mean - sd);
const y15hi = yAt(mean + 1.5 * sd), y15lo = yAt(mean - 1.5 * sd);
return (
PC3 = 向心段力-时曲线第3主成分得分 · 绿带=±1σ正常 · 橙带=1–1.5σ · 红点=偏移>1.5σ
· 每周监控场景：PC3 体现曲线形态跨-session 一致性，非急性疲劳量化 · Gathercole 2019 PLOS ONE
);
}
window.CMJPCAChart = CMJPCAChart;
})();