VarianceTemplate.cpp

Go to the documentation of this file.

/* *****************************************************************
    MESQUITE -- The Mesh Quality Improvement Toolkit

    Copyright 2006 Sandia National Laboratories.  Developed at the
    University of Wisconsin--Madison under SNL contract number
    624796.  The U.S. Government and the University of Wisconsin
    retain certain rights to this software.

    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
    License as published by the Free Software Foundation; either
    version 2.1 of the License, or (at your option) any later version.

    This library is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Lesser General Public License for more details.

    You should have received a copy of the GNU Lesser General Public License
    (lgpl.txt) along with this library; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

    (2006) kraftche@cae.wisc.edu

  ***************************************************************** */

/** \file VarianceTemplate.cpp
 *  \brief
 *  \author Jason Kraftcheck
 */

#include "Mesquite.hpp"
#include "VarianceTemplate.hpp"
#include "QualityMetric.hpp"
#include "MsqError.hpp"
#include "MsqHessian.hpp"
#include "PatchData.hpp"

namespace MBMesquite
{

ObjectiveFunction* VarianceTemplate::clone() const
{
    return new VarianceTemplate( *this );
}

void VarianceTemplate::clear()
{
    mCount = 0;
    mSum = mSqrSum = 0;
    saveCount      = 0;
    saveSum = saveSqrSum = 0;
}

void VarianceTemplate::accumulate( double sum,
                                   double sqr_sum,
                                   size_t count,
                                   EvalType type,
                                   double& result_sum,
                                   double& result_sqr,
                                   size_t& global_count )
{
    switch( type )
    {
        case CALCULATE:
            result_sum   = sum;
            result_sqr   = sqr_sum;
            global_count = count;
            break;

        case ACCUMULATE:
            result_sum   = mSum += sum;
            result_sqr   = mSqrSum += sqr_sum;
            global_count = mCount += count;
            break;

        case SAVE:
            saveSum      = sum;
            saveSqrSum   = sqr_sum;
            saveCount    = count;
            result_sum   = mSum;
            result_sqr   = mSqrSum;
            global_count = mCount;
            break;

        case UPDATE:
            mSum -= saveSum;
            mSqrSum -= saveSqrSum;
            mCount -= saveCount;
            result_sum = mSum += saveSum = sum;
            result_sqr = mSqrSum += saveSqrSum = sqr_sum;
            global_count = mCount += saveCount = count;
            break;

        case TEMPORARY:
            result_sum   = mSum - saveSum + sum;
            result_sqr   = mSqrSum - saveSqrSum + sqr_sum;
            global_count = mCount + count - saveCount;
            break;
    }
}

bool VarianceTemplate::evaluate( EvalType type, PatchData& pd, double& value_out, bool free, MsqError& err )
{
    QualityMetric* qm = get_quality_metric();
    if( type == ObjectiveFunction::ACCUMULATE )
        qm->get_single_pass( pd, qmHandles, free, err );
    else
        qm->get_evaluations( pd, qmHandles, free, err );
    MSQ_ERRFALSE( err );

    // calculate OF value for just the patch
    std::vector< size_t >::const_iterator i;
    double value, sum = 0.0, sqr = 0.0;
    for( i = qmHandles.begin(); i != qmHandles.end(); ++i )
    {
        bool result = qm->evaluate( pd, *i, value, err );
        if( MSQ_CHKERR( err ) || !result ) return false;

        sum += value;
        sqr += value * value;
    }

    // get overall OF value, update member data, etc.
    size_t n;
    accumulate( sum, sqr, qmHandles.size(), type, sum, sqr, n );

    // don't divide by zero
    if( n < 1 )
    {
        value_out = 0.0;
        return true;
    }

    const double rms = sqr / n;
    const double avg = sum / n;
    value_out        = qm->get_negate_flag() * ( rms - avg * avg );
    return true;
}

bool VarianceTemplate::evaluate_with_gradient( EvalType type,
                                               PatchData& pd,
                                               double& value_out,
                                               std::vector< Vector3D >& grad_out,
                                               MsqError& err )
{
    QualityMetric* qm = get_quality_metric();
    qm->get_evaluations( pd, qmHandles, OF_FREE_EVALS_ONLY, err );
    MSQ_ERRFALSE( err );

    // zero gradient data
    grad_out.clear();  // store sum of metric * gradient of metric, and later OF gradient
    grad_out.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );
    gradSum.clear();  // store sum of gradients of metrics
    gradSum.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );

    // calculate OF value and gradient for just the patch
    Matrix3D op;
    std::vector< size_t >::const_iterator i;
    double value, sum = 0.0, sqr = 0.0;
    for( i = qmHandles.begin(); i != qmHandles.end(); ++i )
    {
        bool result = qm->evaluate_with_gradient( pd, *i, value, mIndices, mGradient, err );
        if( MSQ_CHKERR( err ) || !result ) return false;
        if( fabs( value ) < DBL_EPSILON ) continue;

        sum += value;
        sqr += value * value;

        for( size_t j = 0; j < mIndices.size(); ++j )
        {
            const size_t r = mIndices[j];
            gradSum[r] += mGradient[j];
            mGradient[j] *= value;
            grad_out[r] += mGradient[j];
        }
    }

    // update accumulated values (if doing BCD)
    size_t n;
    accumulate( sum, sqr, qmHandles.size(), type, sum, sqr, n );

    // don't divide by zero
    if( n < 1 )
    {
        value_out = 0.0;
        grad_out.clear();
        grad_out.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );
        return true;
    }

    // calculate OF value
    const double rms = sqr / n;
    const double avg = sum / n;
    value_out        = qm->get_negate_flag() * ( rms - avg * avg );

    // Finish calculation of gradient and Hessian
    const double f = qm->get_negate_flag() * 2.0 / n;
    for( size_t k = 0; k < pd.num_free_vertices(); ++k )
    {
        gradSum[k] *= avg;
        grad_out[k] -= gradSum[k];
        grad_out[k] *= f;
    }

    return true;
}

bool VarianceTemplate::evaluate_with_Hessian_diagonal( EvalType type,
                                                       PatchData& pd,
                                                       double& value_out,
                                                       std::vector< Vector3D >& grad_out,
                                                       std::vector< SymMatrix3D >& hess_diag_out,
                                                       MsqError& err )
{
    QualityMetric* qm = get_quality_metric();
    qm->get_evaluations( pd, qmHandles, OF_FREE_EVALS_ONLY, err );
    MSQ_ERRFALSE( err );

    // zero gradient and Hessian data
    grad_out.clear();  // store sum of metric * gradient of metric, and later OF gradient
    grad_out.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );
    gradSum.clear();  // store sum of gradients of metrics
    gradSum.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );
    hessSum.clear();  // store sum of Hessians of metrics
    hessSum.resize( pd.num_free_vertices(), SymMatrix3D( 0.0 ) );
    hess_diag_out.clear();  // store sum of metric * outer_product(metric gradient), and later OF Hessian
    hess_diag_out.resize( pd.num_free_vertices(), SymMatrix3D( 0.0 ) );

    // calculate OF value and gradient for just the patch
    Matrix3D op;
    std::vector< size_t >::const_iterator i;
    double value, sum = 0.0, sqr = 0.0;
    for( i = qmHandles.begin(); i != qmHandles.end(); ++i )
    {
        bool result = qm->evaluate_with_Hessian_diagonal( pd, *i, value, mIndices, mGradient, mHessDiag, err );
        if( MSQ_CHKERR( err ) || !result ) return false;
        if( fabs( value ) < DBL_EPSILON ) continue;

        sum += value;
        sqr += value * value;

        for( size_t j = 0; j < mIndices.size(); ++j )
        {
            const size_t r = mIndices[j];

            hessSum[r] += mHessDiag[j];
            hess_diag_out[r] += outer( mGradient[j] );
            mHessDiag[j] *= value;
            hess_diag_out[r] += mHessDiag[j];

            gradSum[r] += mGradient[j];
            mGradient[j] *= value;
            grad_out[r] += mGradient[j];
        }
    }

    // update accumulated values (if doing BCD)
    size_t n;
    accumulate( sum, sqr, qmHandles.size(), type, sum, sqr, n );

    // don't divide by zero
    if( n < 1 )
    {
        value_out = 0.0;
        grad_out.clear();
        grad_out.resize( pd.num_free_vertices(), Vector3D( 0.0, 0.0, 0.0 ) );
        hess_diag_out.clear();
        hess_diag_out.resize( pd.num_free_vertices(), SymMatrix3D( 0.0 ) );
        return true;
    }

    // calculate OF value
    const double rms = sqr / n;
    const double avg = sum / n;
    value_out        = qm->get_negate_flag() * ( rms - avg * avg );

    // Finish calculation of gradient and Hessian
    const double inv_n = 1.0 / n;
    const double f     = qm->get_negate_flag() * 2.0 / n;
    for( size_t k = 0; k < pd.num_free_vertices(); ++k )
    {
        hessSum[k] *= avg;
        hess_diag_out[k] -= hessSum[k];
        hess_diag_out[k] -= inv_n * outer( gradSum[k] );
        hess_diag_out[k] *= f;

        gradSum[k] *= avg;
        grad_out[k] -= gradSum[k];
        grad_out[k] *= f;
    }

    return true;
}

}  // namespace MBMesquite