Example usage for org.apache.commons.math.stat.regression SimpleRegression getMeanSquareError

Introduction

In this page you can find the example usage for org.apache.commons.math.stat.regression SimpleRegression getMeanSquareError.

Prototype

public double getMeanSquareError() 

Source Link

Document

Returns the sum of squared errors divided by the degrees of freedom, usually abbreviated MSE.

Usage

From source file:playground.boescpa.converters.vissim.tools.TripMatcher.java

@Override
public HashMap<Id<Trip>, Integer> matchTrips(HashMap<Id<Trip>, Long[]> msTrips,
        HashMap<Id<Trip>, Long[]> amTrips) {

    int matchesWithHighScores = 0;
    int matchesWithVeryHighScores = 0;
    int progressCounter = 0;
    int progressChecker = 2;

    HashMap<Id<Trip>, Integer> countsPerAnmTrip = new HashMap<>();
    for (Id<Trip> amTrip : amTrips.keySet()) {
        countsPerAnmTrip.put(amTrip, 0);
    }//from ww  w.  j av a  2s  .  c  o  m
    List<Id<Trip>> amTripsKeySet = new ArrayList<>(amTrips.keySet());

    for (Id<Trip> msTrip : msTrips.keySet()) {
        progressCounter++;
        Long[] msTripZones = msTrips.get(msTrip);

        Id<Trip> bestMatchingAmTrip = null;
        int bestMatchScore = Integer.MIN_VALUE;

        // Shuffle key set:
        Collections.shuffle(amTripsKeySet);
        for (Id<Trip> amTrip : amTripsKeySet) {
            Long[] amTripZones = amTrips.get(amTrip);

            // Linear regression between the to trips:
            SimpleRegression simpleRegression = new SimpleRegression();
            for (int i = 0; i < msTripZones.length; i++) {
                boolean foundNone = true;
                for (int j = 0; j < amTripZones.length; j++) {
                    if (msTripZones[i].equals(amTripZones[j])) {
                        simpleRegression.addData(i, j);
                        foundNone = false;
                    }
                }
                if (foundNone) {
                    int yPos = -(msTripZones.length - i) - NEG_OFFSET_IF_NOT_FOUNG;
                    simpleRegression.addData(i, yPos);
                }
            }

            // Scoring:
            int matchScore = 0;

            // Criterion 1.1: Difference in length of trips not greater than 10%.
            if (((double) Math.abs(msTripZones.length - amTripZones.length))
                    / ((double) msTripZones.length) <= 0.1) {
                matchScore += TEN_PRCT_SCORE;
            }
            // Criterion 1.2: The smaller the difference in length, the better.
            matchScore -= (Math.abs(msTripZones.length - amTripZones.length) * LENGTH_SCORE);

            // Criterion 2: The closer the intercept to zero, the better.
            matchScore -= (int) (Math.abs(simpleRegression.getIntercept()) * INTERCEPT_SCORE);

            // Criterion 3: The closer the slope to one, the better.
            matchScore -= (int) (Math.abs(1 - simpleRegression.getSlope()) * SLOPE_SCORE);

            // Criterion 4: The smaller the mean square error of the regression, the better.
            matchScore -= (int) (Math.abs(simpleRegression.getMeanSquareError()) * MSE_SCORE);

            if (matchScore > bestMatchScore) {
                bestMatchScore = matchScore;
                bestMatchingAmTrip = amTrip;
            }
        }

        countsPerAnmTrip.put(bestMatchingAmTrip, (countsPerAnmTrip.get(bestMatchingAmTrip) + 1));

        if (bestMatchScore >= 0.9 * TEN_PRCT_SCORE) {
            matchesWithHighScores++;
            if (bestMatchScore >= 0.99 * TEN_PRCT_SCORE) {
                matchesWithVeryHighScores++;
            }
        }

        // log progress:
        if (progressCounter >= progressChecker) {
            log.info(progressChecker + " trips matched.");
            progressChecker *= 2;
        }
    }

    log.info("Of total " + msTrips.size() + " trips, " + matchesWithHighScores
            + " were matched with a high score above " + 0.9 * TEN_PRCT_SCORE + " points.");
    log.info("Of total " + msTrips.size() + " trips, " + matchesWithVeryHighScores
            + " were matched with a very high score above " + 0.99 * TEN_PRCT_SCORE + " points.");

    return countsPerAnmTrip;
}

---