/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.commons.math3.fitting.leastsquares;
import java.util.Arrays;
import org.apache.commons.math3.fitting.leastsquares.LeastSquaresProblem.Evaluation;
import org.apache.commons.math3.linear.ArrayRealVector;
import org.apache.commons.math3.linear.RealMatrix;
import org.apache.commons.math3.exception.ConvergenceException;
import org.apache.commons.math3.exception.util.LocalizedFormats;
import org.apache.commons.math3.optim.ConvergenceChecker;
import org.apache.commons.math3.util.Incrementor;
import org.apache.commons.math3.util.Precision;
import org.apache.commons.math3.util.FastMath;
/**
* This class solves a least-squares problem using the Levenberg-Marquardt
* algorithm.
*
* <p>This implementation <em>should</em> work even for over-determined systems
* (i.e. systems having more point than equations). Over-determined systems
* are solved by ignoring the point which have the smallest impact according
* to their jacobian column norm. Only the rank of the matrix and some loop bounds
* are changed to implement this.</p>
*
* <p>The resolution engine is a simple translation of the MINPACK <a
* href="http://www.netlib.org/minpack/lmder.f">lmder</a> routine with minor
* changes. The changes include the over-determined resolution, the use of
* inherited convergence checker and the Q.R. decomposition which has been
* rewritten following the algorithm described in the
* P. Lascaux and R. Theodor book <i>Analyse numérique matricielle
* appliquée à l'art de l'ingénieur</i>, Masson 1986.</p>
* <p>The authors of the original fortran version are:
* <ul>
* <li>Argonne National Laboratory. MINPACK project. March 1980</li>
* <li>Burton S. Garbow</li>
* <li>Kenneth E. Hillstrom</li>
* <li>Jorge J. More</li>
* </ul>
* The redistribution policy for MINPACK is available <a
* href="http://www.netlib.org/minpack/disclaimer">here</a>, for convenience, it
* is reproduced below.</p>
*
* <table border="0" width="80%" cellpadding="10" align="center" bgcolor="#E0E0E0">
* <tr><td>
* Minpack Copyright Notice (1999) University of Chicago.
* All rights reserved
* </td></tr>
* <tr><td>
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* <ol>
* <li>Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.</li>
* <li>Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials provided
* with the distribution.</li>
* <li>The end-user documentation included with the redistribution, if any,
* must include the following acknowledgment:
* <code>This product includes software developed by the University of
* Chicago, as Operator of Argonne National Laboratory.</code>
* Alternately, this acknowledgment may appear in the software itself,
* if and wherever such third-party acknowledgments normally appear.</li>
* <li><strong>WARRANTY DISCLAIMER. THE SOFTWARE IS SUPPLIED "AS IS"
* WITHOUT WARRANTY OF ANY KIND. THE COPYRIGHT HOLDER, THE
* UNITED STATES, THE UNITED STATES DEPARTMENT OF ENERGY, AND
* THEIR EMPLOYEES: (1) DISCLAIM ANY WARRANTIES, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO ANY IMPLIED WARRANTIES
* OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE
* OR NON-INFRINGEMENT, (2) DO NOT ASSUME ANY LEGAL LIABILITY
* OR RESPONSIBILITY FOR THE ACCURACY, COMPLETENESS, OR
* USEFULNESS OF THE SOFTWARE, (3) DO NOT REPRESENT THAT USE OF
* THE SOFTWARE WOULD NOT INFRINGE PRIVATELY OWNED RIGHTS, (4)
* DO NOT WARRANT THAT THE SOFTWARE WILL FUNCTION
* UNINTERRUPTED, THAT IT IS ERROR-FREE OR THAT ANY ERRORS WILL
* BE CORRECTED.</strong></li>
* <li><strong>LIMITATION OF LIABILITY. IN NO EVENT WILL THE COPYRIGHT
* HOLDER, THE UNITED STATES, THE UNITED STATES DEPARTMENT OF
* ENERGY, OR THEIR EMPLOYEES: BE LIABLE FOR ANY INDIRECT,
* INCIDENTAL, CONSEQUENTIAL, SPECIAL OR PUNITIVE DAMAGES OF
* ANY KIND OR NATURE, INCLUDING BUT NOT LIMITED TO LOSS OF
* PROFITS OR LOSS OF DATA, FOR ANY REASON WHATSOEVER, WHETHER
* SUCH LIABILITY IS ASSERTED ON THE BASIS OF CONTRACT, TORT
* (INCLUDING NEGLIGENCE OR STRICT LIABILITY), OR OTHERWISE,
* EVEN IF ANY OF SAID PARTIES HAS BEEN WARNED OF THE
* POSSIBILITY OF SUCH LOSS OR DAMAGES.</strong></li>
* <ol></td></tr>
* </table>
*
* @since 3.3
*/
public class LevenbergMarquardtOptimizer implements LeastSquaresOptimizer {
/** Twice the "epsilon machine". */
private static final double TWO_EPS = 2 * Precision.EPSILON;
/* configuration parameters */
/** Positive input variable used in determining the initial step bound. */
private final double initialStepBoundFactor;
/** Desired relative error in the sum of squares. */
private final double costRelativeTolerance;
/** Desired relative error in the approximate solution parameters. */
private final double parRelativeTolerance;
/** Desired max cosine on the orthogonality between the function vector
* and the columns of the jacobian. */
private final double orthoTolerance;
/** Threshold for QR ranking. */
private final double qrRankingThreshold;
/** Default constructor.
* <p>
* The default values for the algorithm settings are:
* <ul>
* <li>Initial step bound factor: 100</li>
* <li>Cost relative tolerance: 1e-10</li>
* <li>Parameters relative tolerance: 1e-10</li>
* <li>Orthogonality tolerance: 1e-10</li>
* <li>QR ranking threshold: {@link Precision#SAFE_MIN}</li>
* </ul>
**/
public LevenbergMarquardtOptimizer() {
this(100, 1e-10, 1e-10, 1e-10, Precision.SAFE_MIN);
}
/**
* Construct an instance with all parameters specified.
*
* @param initialStepBoundFactor initial step bound factor
* @param costRelativeTolerance cost relative tolerance
* @param parRelativeTolerance parameters relative tolerance
* @param orthoTolerance orthogonality tolerance
* @param qrRankingThreshold threshold in the QR decomposition. Columns with a 2
* norm less than this threshold are considered to be
* all 0s.
*/
public LevenbergMarquardtOptimizer(
final double initialStepBoundFactor,
final double costRelativeTolerance,
final double parRelativeTolerance,
final double orthoTolerance,
final double qrRankingThreshold) {
this.initialStepBoundFactor = initialStepBoundFactor;
this.costRelativeTolerance = costRelativeTolerance;
this.parRelativeTolerance = parRelativeTolerance;
this.orthoTolerance = orthoTolerance;
this.qrRankingThreshold = qrRankingThreshold;
}
/**
* @param newInitialStepBoundFactor Positive input variable used in
* determining the initial step bound. This bound is set to the
* product of initialStepBoundFactor and the euclidean norm of
* {@code diag * x} if non-zero, or else to {@code newInitialStepBoundFactor}
* itself. In most cases factor should lie in the interval
* {@code (0.1, 100.0)}. {@code 100} is a generally recommended value.
* of the matrix is reduced.
* @return a new instance.
*/
public LevenbergMarquardtOptimizer withInitialStepBoundFactor(double newInitialStepBoundFactor) {
return new LevenbergMarquardtOptimizer(
newInitialStepBoundFactor,
costRelativeTolerance,
parRelativeTolerance,
orthoTolerance,
qrRankingThreshold);
}
/**
* @param newCostRelativeTolerance Desired relative error in the sum of squares.
* @return a new instance.
*/
public LevenbergMarquardtOptimizer withCostRelativeTolerance(double newCostRelativeTolerance) {
return new LevenbergMarquardtOptimizer(
initialStepBoundFactor,
newCostRelativeTolerance,
parRelativeTolerance,
orthoTolerance,
qrRankingThreshold);
}
/**
* @param newParRelativeTolerance Desired relative error in the approximate solution
* parameters.
* @return a new instance.
*/
public LevenbergMarquardtOptimizer withParameterRelativeTolerance(double newParRelativeTolerance) {
return new LevenbergMarquardtOptimizer(
initialStepBoundFactor,
costRelativeTolerance,
newParRelativeTolerance,
orthoTolerance,
qrRankingThreshold);
}
/**
* Modifies the given parameter.
*
* @param newOrthoTolerance Desired max cosine on the orthogonality between
* the function vector and the columns of the Jacobian.
* @return a new instance.
*/
public LevenbergMarquardtOptimizer withOrthoTolerance(double newOrthoTolerance) {
return new LevenbergMarquardtOptimizer(
initialStepBoundFactor,
costRelativeTolerance,
parRelativeTolerance,
newOrthoTolerance,
qrRankingThreshold);
}
/**
* @param newQRRankingThreshold Desired threshold for QR ranking.
* If the squared norm of a column vector is smaller or equal to this
* threshold during QR decomposition, it is considered to be a zero vector
* and hence the rank of the matrix is reduced.
* @return a new instance.
*/
public LevenbergMarquardtOptimizer withRankingThreshold(double newQRRankingThreshold) {
return new LevenbergMarquardtOptimizer(
initialStepBoundFactor,
costRelativeTolerance,
parRelativeTolerance,
orthoTolerance,
newQRRankingThreshold);
}
/**
* Gets the value of a tuning parameter.
* @see #withInitialStepBoundFactor(double)
*
* @return the parameter's value.
*/
public double getInitialStepBoundFactor() {
return initialStepBoundFactor;
}
/**
* Gets the value of a tuning parameter.
* @see #withCostRelativeTolerance(double)
*
* @return the parameter's value.
*/
public double getCostRelativeTolerance() {
return costRelativeTolerance;
}
/**
* Gets the value of a tuning parameter.
* @see #withParameterRelativeTolerance(double)
*
* @return the parameter's value.
*/
public double getParameterRelativeTolerance() {
return parRelativeTolerance;
}
/**
* Gets the value of a tuning parameter.
* @see #withOrthoTolerance(double)
*
* @return the parameter's value.
*/
public double getOrthoTolerance() {
return orthoTolerance;
}
/**
* Gets the value of a tuning parameter.
* @see #withRankingThreshold(double)
*
* @return the parameter's value.
*/
public double getRankingThreshold() {
return qrRankingThreshold;
}
/** {@inheritDoc} */
public Optimum optimize(final LeastSquaresProblem problem) {
// Pull in relevant data from the problem as locals.
final int nR = problem.getObservationSize(); // Number of observed data.
final int nC = problem.getParameterSize(); // Number of parameters.
// Counters.
final Incrementor iterationCounter = problem.getIterationCounter();
final Incrementor evaluationCounter = problem.getEvaluationCounter();
// Convergence criterion.
final ConvergenceChecker<Evaluation> checker = problem.getConvergenceChecker();
// arrays shared with the other private methods
final int solvedCols = FastMath.min(nR, nC);
/* Parameters evolution direction associated with lmPar. */
double[] lmDir = new double[nC];
/* Levenberg-Marquardt parameter. */
double lmPar = 0;
// local point
double delta = 0;
double xNorm = 0;
double[] diag = new double[nC];
double[] oldX = new double[nC];
double[] oldRes = new double[nR];
double[] qtf = new double[nR];
double[] work1 = new double[nC];
double[] work2 = new double[nC];
double[] work3 = new double[nC];
// Evaluate the function at the starting point and calculate its norm.
evaluationCounter.incrementCount();
//value will be reassigned in the loop
Evaluation current = problem.evaluate(problem.getStart());
double[] currentResiduals = current.getResiduals().toArray();
double currentCost = current.getCost();
double[] currentPoint = current.getPoint().toArray();
// Outer loop.
boolean firstIteration = true;
while (true) {
iterationCounter.incrementCount();
final Evaluation previous = current;
// QR decomposition of the jacobian matrix
final InternalData internalData
= qrDecomposition(current.getJacobian(), solvedCols);
final double[][] weightedJacobian = internalData.weightedJacobian;
final int[] permutation = internalData.permutation;
final double[] diagR = internalData.diagR;
final double[] jacNorm = internalData.jacNorm;
//residuals already have weights applied
double[] weightedResidual = currentResiduals;
for (int i = 0; i < nR; i++) {
qtf[i] = weightedResidual[i];
}
// compute Qt.res
qTy(qtf, internalData);
// now we don't need Q anymore,
// so let jacobian contain the R matrix with its diagonal elements
for (int k = 0; k < solvedCols; ++k) {
int pk = permutation[k];
weightedJacobian[k][pk] = diagR[pk];
}
if (firstIteration) {
// scale the point according to the norms of the columns
// of the initial jacobian
xNorm = 0;
for (int k = 0; k < nC; ++k) {
double dk = jacNorm[k];
if (dk == 0) {
dk = 1.0;
}
double xk = dk * currentPoint[k];
xNorm += xk * xk;
diag[k] = dk;
}
xNorm = FastMath.sqrt(xNorm);
// initialize the step bound delta
delta = (xNorm == 0) ? initialStepBoundFactor : (initialStepBoundFactor * xNorm);
}
// check orthogonality between function vector and jacobian columns
double maxCosine = 0;
if (currentCost != 0) {
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double s = jacNorm[pj];
if (s != 0) {
double sum = 0;
for (int i = 0; i <= j; ++i) {
sum += weightedJacobian[i][pj] * qtf[i];
}
maxCosine = FastMath.max(maxCosine, FastMath.abs(sum) / (s * currentCost));
}
}
}
if (maxCosine <= orthoTolerance) {
// Convergence has been reached.
return new OptimumImpl(
current,
evaluationCounter.getCount(),
iterationCounter.getCount());
}
// rescale if necessary
for (int j = 0; j < nC; ++j) {
diag[j] = FastMath.max(diag[j], jacNorm[j]);
}
// Inner loop.
for (double ratio = 0; ratio < 1.0e-4;) {
// save the state
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
oldX[pj] = currentPoint[pj];
}
final double previousCost = currentCost;
double[] tmpVec = weightedResidual;
weightedResidual = oldRes;
oldRes = tmpVec;
// determine the Levenberg-Marquardt parameter
lmPar = determineLMParameter(qtf, delta, diag,
internalData, solvedCols,
work1, work2, work3, lmDir, lmPar);
// compute the new point and the norm of the evolution direction
double lmNorm = 0;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
lmDir[pj] = -lmDir[pj];
currentPoint[pj] = oldX[pj] + lmDir[pj];
double s = diag[pj] * lmDir[pj];
lmNorm += s * s;
}
lmNorm = FastMath.sqrt(lmNorm);
// on the first iteration, adjust the initial step bound.
if (firstIteration) {
delta = FastMath.min(delta, lmNorm);
}
// Evaluate the function at x + p and calculate its norm.
evaluationCounter.incrementCount();
current = problem.evaluate(new ArrayRealVector(currentPoint));
currentResiduals = current.getResiduals().toArray();
currentCost = current.getCost();
currentPoint = current.getPoint().toArray();
// compute the scaled actual reduction
double actRed = -1.0;
if (0.1 * currentCost < previousCost) {
double r = currentCost / previousCost;
actRed = 1.0 - r * r;
}
// compute the scaled predicted reduction
// and the scaled directional derivative
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double dirJ = lmDir[pj];
work1[j] = 0;
for (int i = 0; i <= j; ++i) {
work1[i] += weightedJacobian[i][pj] * dirJ;
}
}
double coeff1 = 0;
for (int j = 0; j < solvedCols; ++j) {
coeff1 += work1[j] * work1[j];
}
double pc2 = previousCost * previousCost;
coeff1 /= pc2;
double coeff2 = lmPar * lmNorm * lmNorm / pc2;
double preRed = coeff1 + 2 * coeff2;
double dirDer = -(coeff1 + coeff2);
// ratio of the actual to the predicted reduction
ratio = (preRed == 0) ? 0 : (actRed / preRed);
// update the step bound
if (ratio <= 0.25) {
double tmp =
(actRed < 0) ? (0.5 * dirDer / (dirDer + 0.5 * actRed)) : 0.5;
if ((0.1 * currentCost >= previousCost) || (tmp < 0.1)) {
tmp = 0.1;
}
delta = tmp * FastMath.min(delta, 10.0 * lmNorm);
lmPar /= tmp;
} else if ((lmPar == 0) || (ratio >= 0.75)) {
delta = 2 * lmNorm;
lmPar *= 0.5;
}
// test for successful iteration.
if (ratio >= 1.0e-4) {
// successful iteration, update the norm
firstIteration = false;
xNorm = 0;
for (int k = 0; k < nC; ++k) {
double xK = diag[k] * currentPoint[k];
xNorm += xK * xK;
}
xNorm = FastMath.sqrt(xNorm);
// tests for convergence.
if (checker != null && checker.converged(iterationCounter.getCount(), previous, current)) {
return new OptimumImpl(current, evaluationCounter.getCount(), iterationCounter.getCount());
}
} else {
// failed iteration, reset the previous values
currentCost = previousCost;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
currentPoint[pj] = oldX[pj];
}
tmpVec = weightedResidual;
weightedResidual = oldRes;
oldRes = tmpVec;
// Reset "current" to previous values.
current = previous;
}
// Default convergence criteria.
if ((FastMath.abs(actRed) <= costRelativeTolerance &&
preRed <= costRelativeTolerance &&
ratio <= 2.0) ||
delta <= parRelativeTolerance * xNorm) {
return new OptimumImpl(current, evaluationCounter.getCount(), iterationCounter.getCount());
}
// tests for termination and stringent tolerances
if (FastMath.abs(actRed) <= TWO_EPS &&
preRed <= TWO_EPS &&
ratio <= 2.0) {
throw new ConvergenceException(LocalizedFormats.TOO_SMALL_COST_RELATIVE_TOLERANCE,
costRelativeTolerance);
} else if (delta <= TWO_EPS * xNorm) {
throw new ConvergenceException(LocalizedFormats.TOO_SMALL_PARAMETERS_RELATIVE_TOLERANCE,
parRelativeTolerance);
} else if (maxCosine <= TWO_EPS) {
throw new ConvergenceException(LocalizedFormats.TOO_SMALL_ORTHOGONALITY_TOLERANCE,
orthoTolerance);
}
}
}
}
/**
* Holds internal data.
* This structure was created so that all optimizer fields can be "final".
* Code should be further refactored in order to not pass around arguments
* that will modified in-place (cf. "work" arrays).
*/
private static class InternalData {
/** Weighted Jacobian. */
private final double[][] weightedJacobian;
/** Columns permutation array. */
private final int[] permutation;
/** Rank of the Jacobian matrix. */
private final int rank;
/** Diagonal elements of the R matrix in the QR decomposition. */
private final double[] diagR;
/** Norms of the columns of the jacobian matrix. */
private final double[] jacNorm;
/** Coefficients of the Householder transforms vectors. */
private final double[] beta;
/**
* @param weightedJacobian Weighted Jacobian.
* @param permutation Columns permutation array.
* @param rank Rank of the Jacobian matrix.
* @param diagR Diagonal elements of the R matrix in the QR decomposition.
* @param jacNorm Norms of the columns of the jacobian matrix.
* @param beta Coefficients of the Householder transforms vectors.
*/
InternalData(double[][] weightedJacobian,
int[] permutation,
int rank,
double[] diagR,
double[] jacNorm,
double[] beta) {
this.weightedJacobian = weightedJacobian;
this.permutation = permutation;
this.rank = rank;
this.diagR = diagR;
this.jacNorm = jacNorm;
this.beta = beta;
}
}
/**
* Determines the Levenberg-Marquardt parameter.
*
* <p>This implementation is a translation in Java of the MINPACK
* <a href="http://www.netlib.org/minpack/lmpar.f">lmpar</a>
* routine.</p>
* <p>This method sets the lmPar and lmDir attributes.</p>
* <p>The authors of the original fortran function are:</p>
* <ul>
* <li>Argonne National Laboratory. MINPACK project. March 1980</li>
* <li>Burton S. Garbow</li>
* <li>Kenneth E. Hillstrom</li>
* <li>Jorge J. More</li>
* </ul>
* <p>Luc Maisonobe did the Java translation.</p>
*
* @param qy Array containing qTy.
* @param delta Upper bound on the euclidean norm of diagR * lmDir.
* @param diag Diagonal matrix.
* @param internalData Data (modified in-place in this method).
* @param solvedCols Number of solved point.
* @param work1 work array
* @param work2 work array
* @param work3 work array
* @param lmDir the "returned" LM direction will be stored in this array.
* @param lmPar the value of the LM parameter from the previous iteration.
* @return the new LM parameter
*/
private double determineLMParameter(double[] qy, double delta, double[] diag,
InternalData internalData, int solvedCols,
double[] work1, double[] work2, double[] work3,
double[] lmDir, double lmPar) {
final double[][] weightedJacobian = internalData.weightedJacobian;
final int[] permutation = internalData.permutation;
final int rank = internalData.rank;
final double[] diagR = internalData.diagR;
final int nC = weightedJacobian[0].length;
// compute and store in x the gauss-newton direction, if the
// jacobian is rank-deficient, obtain a least squares solution
for (int j = 0; j < rank; ++j) {
lmDir[permutation[j]] = qy[j];
}
for (int j = rank; j < nC; ++j) {
lmDir[permutation[j]] = 0;
}
for (int k = rank - 1; k >= 0; --k) {
int pk = permutation[k];
double ypk = lmDir[pk] / diagR[pk];
for (int i = 0; i < k; ++i) {
lmDir[permutation[i]] -= ypk * weightedJacobian[i][pk];
}
lmDir[pk] = ypk;
}
// evaluate the function at the origin, and test
// for acceptance of the Gauss-Newton direction
double dxNorm = 0;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double s = diag[pj] * lmDir[pj];
work1[pj] = s;
dxNorm += s * s;
}
dxNorm = FastMath.sqrt(dxNorm);
double fp = dxNorm - delta;
if (fp <= 0.1 * delta) {
lmPar = 0;
return lmPar;
}
// if the jacobian is not rank deficient, the Newton step provides
// a lower bound, parl, for the zero of the function,
// otherwise set this bound to zero
double sum2;
double parl = 0;
if (rank == solvedCols) {
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
work1[pj] *= diag[pj] / dxNorm;
}
sum2 = 0;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double sum = 0;
for (int i = 0; i < j; ++i) {
sum += weightedJacobian[i][pj] * work1[permutation[i]];
}
double s = (work1[pj] - sum) / diagR[pj];
work1[pj] = s;
sum2 += s * s;
}
parl = fp / (delta * sum2);
}
// calculate an upper bound, paru, for the zero of the function
sum2 = 0;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double sum = 0;
for (int i = 0; i <= j; ++i) {
sum += weightedJacobian[i][pj] * qy[i];
}
sum /= diag[pj];
sum2 += sum * sum;
}
double gNorm = FastMath.sqrt(sum2);
double paru = gNorm / delta;
if (paru == 0) {
paru = Precision.SAFE_MIN / FastMath.min(delta, 0.1);
}
// if the input par lies outside of the interval (parl,paru),
// set par to the closer endpoint
lmPar = FastMath.min(paru, FastMath.max(lmPar, parl));
if (lmPar == 0) {
lmPar = gNorm / dxNorm;
}
for (int countdown = 10; countdown >= 0; --countdown) {
// evaluate the function at the current value of lmPar
if (lmPar == 0) {
lmPar = FastMath.max(Precision.SAFE_MIN, 0.001 * paru);
}
double sPar = FastMath.sqrt(lmPar);
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
work1[pj] = sPar * diag[pj];
}
determineLMDirection(qy, work1, work2, internalData, solvedCols, work3, lmDir);
dxNorm = 0;
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
double s = diag[pj] * lmDir[pj];
work3[pj] = s;
dxNorm += s * s;
}
dxNorm = FastMath.sqrt(dxNorm);
double previousFP = fp;
fp = dxNorm - delta;
// if the function is small enough, accept the current value
// of lmPar, also test for the exceptional cases where parl is zero
if (FastMath.abs(fp) <= 0.1 * delta ||
(parl == 0 &&
fp <= previousFP &&
previousFP < 0)) {
return lmPar;
}
// compute the Newton correction
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
work1[pj] = work3[pj] * diag[pj] / dxNorm;
}
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
work1[pj] /= work2[j];
double tmp = work1[pj];
for (int i = j + 1; i < solvedCols; ++i) {
work1[permutation[i]] -= weightedJacobian[i][pj] * tmp;
}
}
sum2 = 0;
for (int j = 0; j < solvedCols; ++j) {
double s = work1[permutation[j]];
sum2 += s * s;
}
double correction = fp / (delta * sum2);
// depending on the sign of the function, update parl or paru.
if (fp > 0) {
parl = FastMath.max(parl, lmPar);
} else if (fp < 0) {
paru = FastMath.min(paru, lmPar);
}
// compute an improved estimate for lmPar
lmPar = FastMath.max(parl, lmPar + correction);
}
return lmPar;
}
/**
* Solve a*x = b and d*x = 0 in the least squares sense.
* <p>This implementation is a translation in Java of the MINPACK
* <a href="http://www.netlib.org/minpack/qrsolv.f">qrsolv</a>
* routine.</p>
* <p>This method sets the lmDir and lmDiag attributes.</p>
* <p>The authors of the original fortran function are:</p>
* <ul>
* <li>Argonne National Laboratory. MINPACK project. March 1980</li>
* <li>Burton S. Garbow</li>
* <li>Kenneth E. Hillstrom</li>
* <li>Jorge J. More</li>
* </ul>
* <p>Luc Maisonobe did the Java translation.</p>
*
* @param qy array containing qTy
* @param diag diagonal matrix
* @param lmDiag diagonal elements associated with lmDir
* @param internalData Data (modified in-place in this method).
* @param solvedCols Number of sloved point.
* @param work work array
* @param lmDir the "returned" LM direction is stored in this array
*/
private void determineLMDirection(double[] qy, double[] diag,
double[] lmDiag,
InternalData internalData,
int solvedCols,
double[] work,
double[] lmDir) {
final int[] permutation = internalData.permutation;
final double[][] weightedJacobian = internalData.weightedJacobian;
final double[] diagR = internalData.diagR;
// copy R and Qty to preserve input and initialize s
// in particular, save the diagonal elements of R in lmDir
for (int j = 0; j < solvedCols; ++j) {
int pj = permutation[j];
for (int i = j + 1; i < solvedCols; ++i) {
weightedJacobian[i][pj] = weightedJacobian[j][permutation[i]];
}
lmDir[j] = diagR[pj];
work[j] = qy[j];
}
// eliminate the diagonal matrix d using a Givens rotation
for (int j = 0; j < solvedCols; ++j) {
// prepare the row of d to be eliminated, locating the
// diagonal element using p from the Q.R. factorization
int pj = permutation[j];
double dpj = diag[pj];
if (dpj != 0) {
Arrays.fill(lmDiag, j + 1, lmDiag.length, 0);
}
lmDiag[j] = dpj;
// the transformations to eliminate the row of d
// modify only a single element of Qty
// beyond the first n, which is initially zero.
double qtbpj = 0;
for (int k = j; k < solvedCols; ++k) {
int pk = permutation[k];
// determine a Givens rotation which eliminates the
// appropriate element in the current row of d
if (lmDiag[k] != 0) {
final double sin;
final double cos;
double rkk = weightedJacobian[k][pk];
if (FastMath.abs(rkk) < FastMath.abs(lmDiag[k])) {
final double cotan = rkk / lmDiag[k];
sin = 1.0 / FastMath.sqrt(1.0 + cotan * cotan);
cos = sin * cotan;
} else {
final double tan = lmDiag[k] / rkk;
cos = 1.0 / FastMath.sqrt(1.0 + tan * tan);
sin = cos * tan;
}
// compute the modified diagonal element of R and
// the modified element of (Qty,0)
weightedJacobian[k][pk] = cos * rkk + sin * lmDiag[k];
final double temp = cos * work[k] + sin * qtbpj;
qtbpj = -sin * work[k] + cos * qtbpj;
work[k] = temp;
// accumulate the tranformation in the row of s
for (int i = k + 1; i < solvedCols; ++i) {
double rik = weightedJacobian[i][pk];
final double temp2 = cos * rik + sin * lmDiag[i];
lmDiag[i] = -sin * rik + cos * lmDiag[i];
weightedJacobian[i][pk] = temp2;
}
}
}
// store the diagonal element of s and restore
// the corresponding diagonal element of R
lmDiag[j] = weightedJacobian[j][permutation[j]];
weightedJacobian[j][permutation[j]] = lmDir[j];
}
// solve the triangular system for z, if the system is
// singular, then obtain a least squares solution
int nSing = solvedCols;
for (int j = 0; j < solvedCols; ++j) {
if ((lmDiag[j] == 0) && (nSing == solvedCols)) {
nSing = j;
}
if (nSing < solvedCols) {
work[j] = 0;
}
}
if (nSing > 0) {
for (int j = nSing - 1; j >= 0; --j) {
int pj = permutation[j];
double sum = 0;
for (int i = j + 1; i < nSing; ++i) {
sum += weightedJacobian[i][pj] * work[i];
}
work[j] = (work[j] - sum) / lmDiag[j];
}
}
// permute the components of z back to components of lmDir
for (int j = 0; j < lmDir.length; ++j) {
lmDir[permutation[j]] = work[j];
}
}
/**
* Decompose a matrix A as A.P = Q.R using Householder transforms.
* <p>As suggested in the P. Lascaux and R. Theodor book
* <i>Analyse numérique matricielle appliquée à
* l'art de l'ingénieur</i> (Masson, 1986), instead of representing
* the Householder transforms with u<sub>k</sub> unit vectors such that:
* <pre>
* H<sub>k</sub> = I - 2u<sub>k</sub>.u<sub>k</sub><sup>t</sup>
* </pre>
* we use <sub>k</sub> non-unit vectors such that:
* <pre>
* H<sub>k</sub> = I - beta<sub>k</sub>v<sub>k</sub>.v<sub>k</sub><sup>t</sup>
* </pre>
* where v<sub>k</sub> = a<sub>k</sub> - alpha<sub>k</sub> e<sub>k</sub>.
* The beta<sub>k</sub> coefficients are provided upon exit as recomputing
* them from the v<sub>k</sub> vectors would be costly.</p>
* <p>This decomposition handles rank deficient cases since the tranformations
* are performed in non-increasing columns norms order thanks to columns
* pivoting. The diagonal elements of the R matrix are therefore also in
* non-increasing absolute values order.</p>
*
* @param jacobian Weighted Jacobian matrix at the current point.
* @param solvedCols Number of solved point.
* @return data used in other methods of this class.
* @throws ConvergenceException if the decomposition cannot be performed.
*/
private InternalData qrDecomposition(RealMatrix jacobian,
int solvedCols) throws ConvergenceException {
// Code in this class assumes that the weighted Jacobian is -(W^(1/2) J),
// hence the multiplication by -1.
final double[][] weightedJacobian = jacobian.scalarMultiply(-1).getData();
final int nR = weightedJacobian.length;
final int nC = weightedJacobian[0].length;
final int[] permutation = new int[nC];
final double[] diagR = new double[nC];
final double[] jacNorm = new double[nC];
final double[] beta = new double[nC];
// initializations
for (int k = 0; k < nC; ++k) {
permutation[k] = k;
double norm2 = 0;
for (int i = 0; i < nR; ++i) {
double akk = weightedJacobian[i][k];
norm2 += akk * akk;
}
jacNorm[k] = FastMath.sqrt(norm2);
}
// transform the matrix column after column
for (int k = 0; k < nC; ++k) {
// select the column with the greatest norm on active components
int nextColumn = -1;
double ak2 = Double.NEGATIVE_INFINITY;
for (int i = k; i < nC; ++i) {
double norm2 = 0;
for (int j = k; j < nR; ++j) {
double aki = weightedJacobian[j][permutation[i]];
norm2 += aki * aki;
}
if (Double.isInfinite(norm2) || Double.isNaN(norm2)) {
throw new ConvergenceException(LocalizedFormats.UNABLE_TO_PERFORM_QR_DECOMPOSITION_ON_JACOBIAN,
nR, nC);
}
if (norm2 > ak2) {
nextColumn = i;
ak2 = norm2;
}
}
if (ak2 <= qrRankingThreshold) {
return new InternalData(weightedJacobian, permutation, k, diagR, jacNorm, beta);
}
int pk = permutation[nextColumn];
permutation[nextColumn] = permutation[k];
permutation[k] = pk;
// choose alpha such that Hk.u = alpha ek
double akk = weightedJacobian[k][pk];
double alpha = (akk > 0) ? -FastMath.sqrt(ak2) : FastMath.sqrt(ak2);
double betak = 1.0 / (ak2 - akk * alpha);
beta[pk] = betak;
// transform the current column
diagR[pk] = alpha;
weightedJacobian[k][pk] -= alpha;
// transform the remaining columns
for (int dk = nC - 1 - k; dk > 0; --dk) {
double gamma = 0;
for (int j = k; j < nR; ++j) {
gamma += weightedJacobian[j][pk] * weightedJacobian[j][permutation[k + dk]];
}
gamma *= betak;
for (int j = k; j < nR; ++j) {
weightedJacobian[j][permutation[k + dk]] -= gamma * weightedJacobian[j][pk];
}
}
}
return new InternalData(weightedJacobian, permutation, solvedCols, diagR, jacNorm, beta);
}
/**
* Compute the product Qt.y for some Q.R. decomposition.
*
* @param y vector to multiply (will be overwritten with the result)
* @param internalData Data.
*/
private void qTy(double[] y,
InternalData internalData) {
final double[][] weightedJacobian = internalData.weightedJacobian;
final int[] permutation = internalData.permutation;
final double[] beta = internalData.beta;
final int nR = weightedJacobian.length;
final int nC = weightedJacobian[0].length;
for (int k = 0; k < nC; ++k) {
int pk = permutation[k];
double gamma = 0;
for (int i = k; i < nR; ++i) {
gamma += weightedJacobian[i][pk] * y[i];
}
gamma *= beta[pk];
for (int i = k; i < nR; ++i) {
y[i] -= gamma * weightedJacobian[i][pk];
}
}
}
}