teichmann@5863: /* Copyright (C) 2011, 2012, 2013 by Bundesanstalt für Gewässerkunde
teichmann@5863:  * Software engineering by Intevation GmbH
teichmann@5863:  *
teichmann@5994:  * This file is Free Software under the GNU AGPL (>=v3)
teichmann@5863:  * and comes with ABSOLUTELY NO WARRANTY! Check out the
teichmann@5994:  * documentation coming with Dive4Elements River for details.
teichmann@5863:  */
teichmann@5863: 
teichmann@5831: package org.dive4elements.river.artifacts.model.sq;
sascha@3188: 
teichmann@5831: import org.dive4elements.river.artifacts.math.fitting.Function;
teichmann@6761: import org.dive4elements.river.artifacts.math.fitting.Linear;
sascha@3188: 
sascha@3188: import java.util.ArrayList;
sascha@3188: import java.util.List;
sascha@3188: 
sascha@3188: import org.apache.commons.math.MathException;
sascha@3188: 
sascha@3188: import org.apache.commons.math.optimization.fitting.CurveFitter;
sascha@3188: 
teichmann@6761: import org.apache.commons.math.optimization.general.AbstractLeastSquaresOptimizer;
teichmann@6761: import org.apache.commons.math.optimization.general.GaussNewtonOptimizer;
sascha@3188: import org.apache.commons.math.optimization.general.LevenbergMarquardtOptimizer;
sascha@3188: 
sascha@3188: import org.apache.log4j.Logger;
sascha@3188: 
sascha@3188: public class Fitting
sascha@3552: implements   Outlier.Callback
sascha@3188: {
teichmann@6761:     // XXX: Hack to force linear fitting!
teichmann@6761:     private static final boolean USE_NON_LINEAR_FITTING =
teichmann@6761:         Boolean.getBoolean("minfo.sq.fitting.nonlinear");
teichmann@6761: 
sascha@3188:     private static Logger log = Logger.getLogger(Fitting.class);
sascha@3188: 
sascha@3552:     public interface Callback {
sascha@3552: 
sascha@3552:         void afterIteration(
sascha@3552:             double [] parameters,
sascha@3552:             SQ []     measurements,
sascha@3552:             SQ []     outliers,
sascha@3552:             double    standardDeviation,
sascha@3552:             double    chiSqr);
sascha@3552:     } // interfacte
sascha@3552: 
sascha@3188:     protected Function function;
sascha@3188: 
sascha@3552:     protected double [] coeffs;
sascha@3188: 
teichmann@5831:     protected org.dive4elements.river.artifacts.math.Function instance;
sascha@3552: 
sascha@3552:     protected double stdDevFactor;
sascha@3188:     protected double chiSqr;
sascha@3188: 
sascha@3552:     protected Callback callback;
sascha@3188: 
sascha@3188:     public Fitting() {
sascha@3188:     }
sascha@3188: 
sascha@3188:     public Fitting(Function function, double stdDevFactor) {
sascha@3552:         this();
sascha@3188:         this.function     = function;
sascha@3188:         this.stdDevFactor = stdDevFactor;
sascha@3188:     }
sascha@3188: 
sascha@3188:     public Function getFunction() {
sascha@3188:         return function;
sascha@3188:     }
sascha@3188: 
sascha@3188:     public void setFunction(Function function) {
sascha@3188:         this.function = function;
sascha@3188:     }
sascha@3188: 
sascha@3188:     public double getStdDevFactor() {
sascha@3188:         return stdDevFactor;
sascha@3188:     }
sascha@3188: 
sascha@3188:     public void setStdDevFactor(double stdDevFactor) {
sascha@3188:         this.stdDevFactor = stdDevFactor;
sascha@3188:     }
sascha@3188: 
sascha@3552:     @Override
sascha@3566:     public void initialize(List<SQ> sqs) throws MathException {
sascha@3552: 
teichmann@6761:         AbstractLeastSquaresOptimizer optimizer = getOptimizer();
sascha@3552: 
teichmann@6761:         CurveFitter cf = new CurveFitter(optimizer);
teichmann@6761:         double [] values = new double[2];
sascha@3566:         for (SQ sq: sqs) {
teichmann@6761:             values[0] = sq.getQ();
teichmann@6761:             values[1] = sq.getS();
teichmann@6761:             transformInputValues(values);
teichmann@6761:             cf.addObservedPoint(values[0], values[1]);
sascha@3552:         }
sascha@3552: 
sascha@3552:         coeffs = cf.fit(
sascha@3552:             function, function.getInitialGuess());
sascha@3552: 
teichmann@6761:         transformCoeffsBack(coeffs);
teichmann@6761: 
sascha@3552:         instance = function.instantiate(coeffs);
sascha@3552: 
teichmann@6761:         chiSqr = optimizer.getChiSquare();
teichmann@6761:     }
teichmann@6761: 
teichmann@6761:     protected Function getFunction(Function function) {
teichmann@6761:         return USE_NON_LINEAR_FITTING
teichmann@6761:             ? function
teichmann@6761:             : Linear.INSTANCE;
teichmann@6761:     }
teichmann@6761: 
teichmann@6761:     protected void transformInputValues(double [] values) {
teichmann@6761:         if (!USE_NON_LINEAR_FITTING) {
teichmann@6761:             for (int i = 0; i < values.length; ++i) {
teichmann@6761:                 values[i] = Math.log(values[i]);
teichmann@6761:             }
teichmann@6761:         }
teichmann@6761:     }
teichmann@6761: 
teichmann@6761:     protected AbstractLeastSquaresOptimizer getOptimizer() {
teichmann@6761:         return USE_NON_LINEAR_FITTING
teichmann@6761:             ? new LevenbergMarquardtOptimizer()
teichmann@6761:             : new GaussNewtonOptimizer(false);
teichmann@6761:     }
teichmann@6761: 
teichmann@6761:     protected void transformCoeffsBack(double [] coeffs) {
teichmann@6761:         if (!USE_NON_LINEAR_FITTING && coeffs.length > 0) {
teichmann@6761:             coeffs[0] = Math.exp(coeffs[0]);
teichmann@6761:         }
sascha@3188:     }
sascha@3188: 
sascha@3552:     @Override
sascha@3552:     public double eval(SQ sq) {
sascha@3552:         double s = instance.value(sq.q);
sascha@3552:         return sq.s - s;
sascha@3552:     }
sascha@3552: 
sascha@3552:     @Override
sascha@3566:     public void iterationFinished(
sascha@3566:         double   standardDeviation,
sascha@3566:         SQ       outlier,
sascha@3566:         List<SQ> remainings
sascha@3566:     ) {
sascha@3552:         if (log.isDebugEnabled()) {
sascha@3552:             log.debug("iterationFinished ----");
sascha@3552:             log.debug(" num remainings: " + remainings.size());
sascha@3566:             log.debug(" has outlier: " + outlier != null);
sascha@3552:             log.debug(" standardDeviation: " + standardDeviation);
sascha@3552:             log.debug(" Chi^2: " + chiSqr);
sascha@3552:             log.debug("---- iterationFinished");
sascha@3552:         }
sascha@3552:         callback.afterIteration(
sascha@3552:             coeffs,
sascha@3552:             remainings.toArray(new SQ[remainings.size()]),
sascha@3566:             outlier != null ? new SQ [] { outlier} : new SQ [] {},
sascha@3552:             standardDeviation,
sascha@3552:             chiSqr);
sascha@3188:     }
sascha@3188: 
sascha@3188:     protected static final List<SQ> onlyValid(List<SQ> sqs) {
sascha@3188: 
sascha@3188:         List<SQ> good = new ArrayList<SQ>(sqs.size());
sascha@3188: 
sascha@3188:         for (SQ sq: sqs) {
sascha@3188:             if (sq.isValid()) {
sascha@3188:                 good.add(sq);
sascha@3188:             }
sascha@3188:         }
sascha@3188: 
sascha@3188:         return good;
sascha@3188:     }
sascha@3188: 
rrenkert@5396:     public boolean fit(List<SQ> sqs, String method, Callback callback) {
sascha@3188: 
sascha@3188:         sqs = onlyValid(sqs);
sascha@3188: 
sascha@3188:         if (sqs.size() < 2) {
sascha@3188:             log.warn("Too less points for fitting.");
sascha@3188:             return false;
sascha@3188:         }
sascha@3188: 
sascha@3552:         this.callback = callback;
sascha@3188: 
sascha@3188:         try {
rrenkert@5396:             Outlier.detectOutliers(this, sqs, stdDevFactor, method);
sascha@3188:         }
sascha@3188:         catch (MathException me) {
sascha@3188:             log.warn(me);
sascha@3188:             return false;
sascha@3188:         }
sascha@3188: 
sascha@3188:         return true;
sascha@3188:     }
sascha@3188: }
sascha@3188: // vim:set ts=4 sw=4 si et sta sts=4 fenc=utf8 :