package com.imsl.test.example.datamining.supportvectormachine;

import com.imsl.datamining.supportvectormachine.*;

/**
 * <p>
 * Illustrates the use of case weights on the
 * training data.
 * </p>
 * <p>
 * This example is the same as {@link SupportVectorMachineEx4} except that it
 * also shows how to apply case weights to the training data. Compare the fitted
 * values between the two examples (the predicted values do not change).
 * </p>
 *
 * @see <a href="SupportVectorMachineEx5.java">Code</a>
 * @see <a href="doc-files/SupportVectorMachineEx5.out">Output</a>
 *
 */
public class SupportVectorMachineEx5 {

    public static void main(String[] args) throws Exception {

        SVRegression.VariableType[] ex4DataType = {
            SVRegression.VariableType.CATEGORICAL,
            SVRegression.VariableType.QUANTITATIVE_CONTINUOUS,
            SVRegression.VariableType.QUANTITATIVE_CONTINUOUS};

        String dashes
                = "--------------------------------------------------------------";

        double C = 50., nu = .01;
        double[][] xyTrain = {
            {1, 0.19, 0.61}, {1, 0.156, 0.564}, {1, 0.224, 0.528},
            {1, 0.178, 0.51}, {1, 0.234, 0.578}, {2, 0.394, 0.296},
            {2, 0.478, 0.254}, {2, 0.454, 0.294}, {2, 0.48, 0.358},
            {2, 0.398, 0.336}
        };
        double[][] xyTest = {
            {1, 0.316, 0.556}, {1, 0.278, 0.622}, {2, 0.562, 0.336},
            {2, 0.522, 0.412}
        };

        double[] trainingWts = {10, 1, 1, 1, 1, 1, 1, 1, 1, 1};

        /* Construct a Support Vector Machine. */
        SVRegression svm1 = new SVRegression(xyTrain, 0, ex4DataType);
        svm1.setNuFormulation(true);
        svm1.setNuParameter(nu);
        svm1.setRegularizationParameter(C);
        svm1.setWeights(trainingWts);
        svm1.fitModel();

        double[] fittedValues = svm1.predict();

        System.out.println("\n" + dashes);
        System.out.println(" NU SVR: Training data predicted (fitted) values");
        System.out.println(" Actual  Fitted value | Difference");
        for (int i = 0; i < fittedValues.length; i++) {
            System.out.printf("  %2.1f      %5.4f       %5.4f\n", xyTrain[i][0],
                    fittedValues[i], (fittedValues[i] - xyTrain[i][0]));
        }
        System.out.println("\n" + dashes);

        double[] testPredictedValues = svm1.predict(xyTest);

        System.out.println("\n NU SVR:   Test data predictions");
        System.out.println("  Actual  Prediction  | Difference");
        for (int i = 0; i < testPredictedValues.length; i++) {
            System.out.printf("  %2.1f      %5.4f      %5.4f\n", xyTest[i][0],
                    testPredictedValues[i],
                    (testPredictedValues[i] - xyTest[i][0]));
        }
        /* Now use the categorical version and compare results. */
        SVClassification svm2 = new SVClassification(xyTrain, 0, ex4DataType);
        svm2.setNuFormulation(true);
        svm2.setNuParameter(nu);
        svm2.setRegularizationParameter(C);
        svm2.setWeights(trainingWts);
        svm2.fitModel();

        fittedValues = svm2.predict();

        System.out.println("\n" + dashes);
        System.out.println(" NU SVC:  Training data predicted (fitted) values");
        System.out.println(" Actual  Fitted value | Difference");
        for (int i = 0; i < fittedValues.length; i++) {
            System.out.printf("  %2.1f      %5.4f       %5.4f\n", xyTrain[i][0],
                    fittedValues[i], (fittedValues[i] - xyTrain[i][0]));
        }
        System.out.println("\n" + dashes);

        testPredictedValues = svm2.predict(xyTest);

        System.out.println("\n NU SVC:  Test data predictions");
        System.out.println(" Actual  Prediction  | Difference");
        for (int i = 0; i < testPredictedValues.length; i++) {
            System.out.printf("  %2.1f      %5.4f      %5.4f\n", xyTest[i][0],
                    testPredictedValues[i],
                    (testPredictedValues[i] - xyTest[i][0]));
        }
    }
}