model.js

// Client-side model runtime: intercepts the app's fetch() calls and answers
// them from item factors shipped with the page. Generated by export_static.py.
(() => {
  const ALPHA = 20.0, REG = 0.1, F = 96;
  const PERSONAS = [{"name": "Sci-fi night", "titles": ["Matrix, The", "Star Wars: Episode IV - A New Hope", "Terminator 2: Judgment Day", "Alien", "Blade Runner"]}, {"name": "Rom-com classics", "titles": ["Sleepless in Seattle", "You've Got Mail", "Pretty Woman", "Four Weddings and a Funeral", "Notting Hill"]}, {"name": "Animated favorites", "titles": ["Toy Story", "Bug's Life, A", "Aladdin", "Lion King, The", "Beauty and the Beast"]}, {"name": "Horror night", "titles": ["Shining, The", "Halloween", "Exorcist, The", "Scream", "Psycho"]}];
  let catalog = null, V = null, VtV = null, unit = null, byId = null;

  async function load() {
    if (catalog) return;
    catalog = await (await realFetch("catalog.json")).json();
    const buf = await (await realFetch("factors.bin")).arrayBuffer();
    V = new Float32Array(buf);
    byId = new Map(catalog.map((m, i) => [m.movie_id, i]));
    VtV = new Float64Array(F * F);
    const n = catalog.length;
    for (let i = 0; i < n; i++) {
      const o = i * F;
      for (let a = 0; a < F; a++) {
        const va = V[o + a];
        for (let b = a; b < F; b++) VtV[a * F + b] += va * V[o + b];
      }
    }
    for (let a = 0; a < F; a++) for (let b = 0; b < a; b++) VtV[a * F + b] = VtV[b * F + a];
    unit = new Float32Array(V.length);
    for (let i = 0; i < n; i++) {
      const o = i * F;
      let s = 0;
      for (let a = 0; a < F; a++) s += V[o + a] * V[o + a];
      const inv = s > 0 ? 1 / Math.sqrt(s) : 0;
      for (let a = 0; a < F; a++) unit[o + a] = V[o + a] * inv;
    }
  }

  function solve(A, b) {   // Gaussian elimination with partial pivoting
    const n = b.length, x = Float64Array.from(b), M = Float64Array.from(A);
    for (let c = 0; c < n; c++) {
      let p = c;
      for (let r = c + 1; r < n; r++) if (Math.abs(M[r * n + c]) > Math.abs(M[p * n + c])) p = r;
      if (p !== c) {
        for (let k = c; k < n; k++) { const t = M[c * n + k]; M[c * n + k] = M[p * n + k]; M[p * n + k] = t; }
        const t = x[c]; x[c] = x[p]; x[p] = t;
      }
      const piv = M[c * n + c];
      for (let r = c + 1; r < n; r++) {
        const f = M[r * n + c] / piv;
        if (f === 0) continue;
        for (let k = c; k < n; k++) M[r * n + k] -= f * M[c * n + k];
        x[r] -= f * x[c];
      }
    }
    for (let r = n - 1; r >= 0; r--) {
      let s = x[r];
      for (let k = r + 1; k < n; k++) s -= M[r * n + k] * x[k];
      x[r] = s / M[r * n + r];
    }
    return x;
  }

  const movie = i => {
    const m = catalog[i];
    return { movie_id: m.movie_id, title: m.title, year: m.year, genres: m.genres };
  };

  function recommend(movieIds, topN) {
    const idx = [...new Set(movieIds.map(id => byId.get(id)))].filter(i => i !== undefined);
    const A = Float64Array.from(VtV), b = new Float64Array(F);
    for (const i of idx) {
      const o = i * F;
      for (let a = 0; a < F; a++) {
        b[a] += (1 + ALPHA) * V[o + a];
        for (let c = 0; c < F; c++) A[a * F + c] += ALPHA * V[o + a] * V[o + c];
      }
    }
    for (let a = 0; a < F; a++) A[a * F + a] += REG;
    const u = solve(A, b);
    const n = catalog.length, scores = new Float64Array(n);
    for (let i = 0; i < n; i++) {
      const o = i * F;
      let s = 0;
      for (let a = 0; a < F; a++) s += V[o + a] * u[a];
      scores[i] = s;
    }
    const excluded = new Set(idx);
    const order = [...scores.keys()].filter(i => !excluded.has(i))
      .sort((x, y) => scores[y] - scores[x]).slice(0, topN);
    const maxScore = Math.max(scores[order[0]], 1e-9);
    const recs = order.map((i, r) => {
      let best = idx[0], bestSim = -2;
      for (const p of idx) {
        let s = 0;
        for (let a = 0; a < F; a++) s += unit[i * F + a] * unit[p * F + a];
        if (s > bestSim) { bestSim = s; best = p; }
      }
      return { ...movie(i), rank: r + 1, score: Math.round(scores[i] * 1e4) / 1e4,
               match: Math.round(scores[i] / maxScore * 1e4) / 1e4, because: catalog[best].title };
    });
    return { picks: idx.map(movie), recommendations: recs };
  }

  const respond = data => Promise.resolve({ ok: true, status: 200, json: async () => data });
  const realFetch = window.fetch.bind(window);

  window.fetch = async (url, opts) => {
    const u = String(url);
    if (u.startsWith("search?")) {
      await load();
      const q = new URLSearchParams(u.split("?")[1]).get("q").trim().toLowerCase();
      if (q.length < 2) return respond([]);
      const hits = catalog.filter(m => m.title.toLowerCase().includes(q))
        .sort((a, b) => b.n_likes - a.n_likes).slice(0, 20)
        .map(m => movie(byId.get(m.movie_id)));
      return respond(hits);
    }
    if (u.startsWith("sample?")) {
      await load();
      const i = parseInt(new URLSearchParams(u.split("?")[1]).get("index") || "0", 10);
      const p = PERSONAS[i % PERSONAS.length];
      return respond({ name: p.name,
        movies: p.titles.filter(t => catalog.some(m => m.title === t))
          .map(t => movie(byId.get(catalog.find(m => m.title === t).movie_id))) });
    }
    if (u === "recommend") {
      await load();
      const body = JSON.parse(opts.body);
      return respond(recommend(body.movie_ids, body.n || 10));
    }
    return realFetch(url, opts);
  };
})();