Apache Math 3.6.1的多项式回归

Question

Can someone let me know how I can do Polynomial Regression with Apache Maths 3.6.1

Below are the data points I used for my testing

60735214881.391304  1520254800000.000000
60697824142.469570  1520258400000.000000
60651182200.208694  1520262000000.000000
60684367132.939130  1520265600000.000000
60676588613.008700  1520269200000.000000
60641816564.869570  1520272800000.000000
60604714824.233510  1520276400000.000000
60580042814.330440  1520280000000.000000
60536134542.469570  1520283600000.000000
60566323732.034780  1520287200000.000000
60578775249.252174  1520290800000.000000
60547382844.104350  1520294400000.000000
60536776546.802160  1520298000000.000000
60474342718.330440  1520301600000.000000
60452725477.286960  1520305200000.000000
60486821569.669560  1520308800000.000000
60247997139.995674  1520312400000.000000
60248432181.426090  1520316000000.000000
60217476247.373920  1520319600000.000000
60170744493.634780  1520323200000.000000

My code looks like below

private void polynomialFitter(List<List<Double>> pointlist) {
        final PolynomialCurveFitter fitter = PolynomialCurveFitter.create(2);
        final WeightedObservedPoints obs = new WeightedObservedPoints();
        for (List<Double> point : pointlist) {
            obs.add(point.get(1), point.get(0));
        }
        double[] fit = fitter.fit(obs.toList());
        System.out.printf("\nCoefficient %f, %f, %f", fit[0], fit[1], fit[2]);
    }

The coefficients are reported as

Coefficient 12.910025, 0.000000, 0.000000

But these does not seem to be quite correct. If I use the same dataset in Online Polynimal Regression and in archanoid online regression - both reports same value as 654623237474.68250993904929103762, 28.75921919628759991574, -0.00000000023885199278

Can someone let me know what is going wrong? I have seen this question but that is not helping me.

Answer 1

This has been answered in apache-commons mailing list

Polynomial regression is not the same as curve fitting. To do polynomial regression in Commons Math, use the OLSMultipleLinearRegression class, using, X, X^2 etc as the independent variables (as your second reference above shows).

A sample code is like below

private OLSMultipleLinearRegression getMultipleLinearRegression(List<List<Double>> pointlist) {
    OLSMultipleLinearRegression regression = new OLSMultipleLinearRegression();
    double y[] = new double[pointlist.size()];
    double x[][] = new double[pointlist.size()][2];
    int c = 0;
    for (List<Double> point : pointlist) {
        y[c] = point.get(0);
        x[c][0] = point.get(1);
        x[c][1] = Math.pow(point.get(1), 2);
        regression.newSampleData(y, x);
        c++;
    }
    System.out.printf("\tR2 = %f", regression.calculateRSquared());
    return regression;
}

Answer 2

With Commons Math 3.6.1, I get almost identical parameters when fitting a cubic to data using PolynomialCurveFitter and tuk's OLSMultipleLinearRegression suggestion.

The curve gives correct interpolations in the range of the data, but you've got to be careful about extrapolations.

The PolynomialCurveFitter code was half the number of lines of the OLSMultipleLinearRegression code.