Stochastic Gradient Descent, but this time in Java.

sgdlinreg.java

package com.nikhilgopal.spark;

/**
 * Created by nikhilgopal on 3/24/17.
 */
public class SGDLinReg {
    public static void main(String[] args) {
        double[] coefficients = {0.4, 0.8};
        double[][] dataset = {
                {1.0, 1.0},
                {2.0, 3.0},
                {4.0, 3.0},
                {3.0, 2.0},
                {5.0, 5.0}
        };
        double[] newcoef = coef_sgd(dataset, 0.001, 500);
        System.out.println(newcoef[0] + " " + newcoef[1]);
    }

    private static double[] coef_sgd(double[][] train, double l_rate, double n_epoch) {
        double[] coefficients = {0.0, 0.0};
        for (int e = 0; e < n_epoch; e++) {
            double sum_error = 0.0;
            for (int w = 0; w < train.length; w++) {
                double yhat = predict(train[w], coefficients);
                double error = yhat - train[w][1];
                sum_error += error*error;
                coefficients[0] = coefficients[0] = l_rate*error;
                for (int k = 0; k < (train[w].length-1); k++) {
                    coefficients[k+1] = coefficients[k+1] - l_rate * error * train[w][k];
                }
            }
            System.out.println("EPOCH " + e + " LRATE " + l_rate + " ERR " + sum_error);
        }
        return coefficients;
    }

    private static double predict(double[] row, double[] coef) {
        double yhat = coef[0];
        for (int i = 0; i < (coef.length-1); i++) {
            yhat += coef[i+1] * row[i];
        }
        return yhat;
    }
}