Logistic Regression Tool

Simple binary logistic regression implementation using gradient descent optimization.

Installation

npm install @tpmjs/tools-logistic-regression

Usage

import { logisticRegressionTool } from '@tpmjs/tools-logistic-regression';

// Example: Predict binary outcome from features
const result = await logisticRegressionTool.execute({
  x: [
    [1.0, 2.0],
    [2.0, 3.0],
    [3.0, 4.0],
    [4.0, 5.0],
  ],
  y: [0, 0, 1, 1],
  iterations: 1000,
});

console.log(result);
// {
//   coefficients: [0.5, 0.3, -0.2], // [intercept, feature1, feature2]
//   predictions: [0, 0, 1, 1],
//   accuracy: 1.0,
//   iterations: 1000,
//   convergence: {
//     finalLoss: 0.123,
//     converged: true
//   }
// }

API

Input

• x (required): Feature matrix number[][] where each row is a sample
• y (required): Binary labels number[] (must be 0 or 1)
• iterations (optional): Number of gradient descent iterations (default: 1000)
• learningRate (optional): Learning rate (default: 0.1)

Output

• coefficients: Model weights including intercept
• predictions: Binary predictions for each sample
• accuracy: Classification accuracy (0 to 1)
• iterations: Number of iterations performed
• convergence: Loss and convergence status

Algorithm

Uses gradient descent to minimize binary cross-entropy loss:

1. Initialize coefficients to zero
2. For each iteration:
   • Calculate predictions using sigmoid function
   • Compute gradient of loss function
   • Update coefficients: θ = θ - α∇L
3. Return fitted model

The sigmoid function maps linear combinations to probabilities [0, 1].

License

MIT