als-statistics/lib/analyze/regression/logistic.js

Modular JS statistics toolkit for Node.js and the browser: descriptive stats, correlations (Pearson/Spearman/Kendall), t-tests & ANOVA (Student/Welch), reliability (Cronbach’s alpha), regression (linear/logistic), clustering (DBSCAN/HDBSCAN), and table/co

import { RegressionBase } from "./regression-base.js"
const sigmoid = (z) => 1 / (1 + Math.exp(-z))

class LogisticRegression extends RegressionBase {
   tol = 1e-6;
   constructor(samples, yName, xNames, step, learningRate = 0.01, epochs = 1000) {
      super(samples, yName, xNames, step)
      this.learningRate = learningRate
      this.epochs = epochs
   }

   calculate() {
      super.calculate()
      const correct = this.yHat.reduce((acc, pred, i) => acc + (pred === this.y[i] ? 1 : 0), 0)
      this.accuracy = correct / this.n
      return this
   }

   computeCoefficients() {
      this.coefficients = Array(this.k).fill(0)
      for (let epoch = 0; epoch < this.epochs; epoch++) { // Градиентный спуск
         const preds = this.predictProba(this.X)
         const errors = preds.map((p, i) => p - this.y[i])

         const grads = Array(this.k).fill(0) // Градиенты
         for (let i = 0; i < this.n; i++) {
            for (let j = 0; j < this.k; j++) {
               grads[j] += this.X[i][j] * errors[i]
            }
         }
         for (let j = 0; j < this.k; j++) { // Обновление коэффициентов
            this.coefficients[j] -= this.learningRate * grads[j] / this.n
         }
         const gradNorm = grads.reduce((sum, g) => sum + g * g, 0);
         if (gradNorm < this.tol * this.tol) break; // Early stop
      }
   }

   predictProba(X) {
      return X.map(row => sigmoid(row.reduce((sum, val, j) => sum + val * this.coefficients[j], 0)))
   }

   predict(X, threshold = 0.5) {
      return this.predictProba(X).map(p => (p >= threshold ? 1 : 0))
   }

   get result() { return { ...super.result, Accuracy: this.accuracy } }
}

export default LogisticRegression