ഇന്ററാക്ടീവ് ഗൗസിയൻ മിശ്രിത മോഡലുകൾ

function randn() {
  let u = 0, v = 0;
  while(u === 0) u = Math.random();
  while(v === 0) v = Math.random();
  return Math.sqrt(-2.0 * Math.log(u)) * Math.cos(2.0 * Math.PI * v);
}

function gaussianPDF(x, mean, variance) {
  const std = Math.sqrt(variance);
  const coeff = 1.0 / (std * Math.sqrt(2 * Math.PI));
  const exponent = -0.5 * Math.pow((x - mean)/std, 2);
  return coeff * Math.exp(exponent);
}

function generateSeparatedMeans(C) {
  let candidate = [];
  for (let i = 0; i < C; i++) {
    candidate.push(Math.random());
  }
  candidate.sort((a,b) => a - b);
  let means = candidate.map(x => -5 + x*10);
  for (let i = 1; i < C; i++) {
    if (means[i] - means[i-1] < 0.5) {
      means[i] = means[i-1] + 0.5;
    }
  }
  return means;
}

function generateData(C, N=1000) {
  let means = generateSeparatedMeans(C);
  let variances = [];
  let weights = [];
  for (let i = 0; i < C; i++) {
    variances.push(0.5 + 1.5*Math.random());
    weights.push(1.0/C);
  }

  let data = [];
  for (let i = 0; i < N; i++) {
    const comp = Math.floor(Math.random() * C);
    const x = means[comp] + Math.sqrt(variances[comp])*randn();
    data.push(x);
  }
  return {data, means, variances, weights};
}

function decentInitialGuess(C, data) {
  const N = data.length;
  let means = [];
  let variances = [];
  let weights = [];
  for (let c = 0; c < C; c++) {
    means.push(data[Math.floor(Math.random()*N)]);
    variances.push(1.0);
    weights.push(1.0/C);
  }
  return {means, variances, weights};
}

function emGMM(data, C, maxIter=100, tol=1e-4) {
  const N = data.length;
  let init = decentInitialGuess(C, data);
  let means = init.means.slice();
  let variances = init.variances.slice();
  let weights = init.weights.slice();

  let logLikOld = -Infinity;
  let paramsHistory = [];

  for (let iter = 0; iter < maxIter; iter++) {
    let resp = new Array(N).fill(0).map(() => new Array(C).fill(0));
    for (let i = 0; i < N; i++) {
      let total = 0;
      for (let c = 0; c < C; c++) {
        const val = weights[c]*gaussianPDF(data[i], means[c], variances[c]);
        resp[i][c] = val;
        total += val;
      }
      for (let c = 0; c < C; c++) {
        resp[i][c] /= (total + 1e-15);
      }
    }

    for (let c = 0; c < C; c++) {
      let sumResp = 0;
      let sumMean = 0;
      let sumVar = 0;
      for (let i = 0; i < N; i++) {
        sumResp += resp[i][c];
        sumMean += resp[i][c]*data[i];
      }
      const newMean = sumMean / (sumResp + 1e-15);

      for (let i = 0; i < N; i++) {
        let diff = data[i] - newMean;
        sumVar += resp[i][c]*diff*diff;
      }
      const newVar = sumVar/(sumResp + 1e-15);

      means[c] = newMean;
      variances[c] = Math.max(newVar, 1e-6);
      weights[c] = sumResp/N;
    }

    let logLik = 0;
    for (let i = 0; i < N; i++) {
      let p = 0;
      for (let c = 0; c < C; c++) {
        p += weights[c]*gaussianPDF(data[i], means[c], variances[c]);
      }
      logLik += Math.log(p + 1e-15);
    }

    paramsHistory.push({
      means: means.slice(),
      variances: variances.slice(),
      weights: weights.slice()
    });

    if (Math.abs(logLik - logLikOld) < tol) {
      break;
    }
    logLikOld = logLik;
  }
  return paramsHistory;
}