DownhillSimplexMethod.cpp

Go to the documentation of this file.

//
// File: DownhillSimplexMethod.cpp
// Authors:
//   Julien Dutheil
// Created: 2003-11-04 17:10:05
//
 
/*
  Copyright or © or Copr. CNRS, (November 17, 2004)
  
  This software is a computer program whose purpose is to provide classes
  for numerical calculus.
  
  This software is governed by the CeCILL license under French law and
  abiding by the rules of distribution of free software. You can use,
  modify and/ or redistribute the software under the terms of the CeCILL
  license as circulated by CEA, CNRS and INRIA at the following URL
  "http://www.cecill.info".
  
  As a counterpart to the access to the source code and rights to copy,
  modify and redistribute granted by the license, users are provided only
  with a limited warranty and the software's author, the holder of the
  economic rights, and the successive licensors have only limited
  liability.
  
  In this respect, the user's attention is drawn to the risks associated
  with loading, using, modifying and/or developing or reproducing the
  software by the user in light of its specific status of free software,
  that may mean that it is complicated to manipulate, and that also
  therefore means that it is reserved for developers and experienced
  professionals having in-depth computer knowledge. Users are therefore
  encouraged to load and test the software's suitability as regards their
  requirements in conditions enabling the security of their systems and/or
  data to be ensured and, more generally, to use and operate it in the
  same conditions as regards security.
  
  The fact that you are presently reading this means that you have had
  knowledge of the CeCILL license and that you accept its terms.
*/
 
 
#include "../NumTools.h"
#include "DownhillSimplexMethod.h"
 
using namespace bpp;
using namespace std;
 
/******************************************************************************/
 
double DownhillSimplexMethod::DSMStopCondition::getCurrentTolerance() const
{
  const DownhillSimplexMethod* dsm = dynamic_cast<const DownhillSimplexMethod*>(optimizer_);
  double rTol = 2.0 * NumTools::abs(dsm->y_[dsm->iHighest_] - dsm->y_[dsm->iLowest_]) /
                (NumTools::abs(dsm->y_[dsm->iHighest_]) + NumTools::abs(dsm->y_[dsm->iLowest_]));
  return rTol;
}
 
/******************************************************************************/
 
DownhillSimplexMethod::DownhillSimplexMethod(Function* function) :
  AbstractOptimizer(function), simplex_(), y_(), pSum_(), iHighest_(0), iNextHighest_(0), iLowest_(0)
{
  // Default values:
  nbEvalMax_ = 5000;
  setDefaultStopCondition_(new DSMStopCondition(this));
  setStopCondition(*getDefaultStopCondition());
}
 
/******************************************************************************/
 
void DownhillSimplexMethod::doInit(const ParameterList& params)
{
  size_t nDim = getParameters().size();
  nbEval_ = 0;
 
  // Initialize the simplex:
  simplex_.resize(nDim + 1);
  y_.resize(nDim + 1);
  double lambda = 0.2; // 20% of the parameter value.
  for (unsigned int i = 1; i < nDim + 1; i++)
  {
    // Copy the vector...
    simplex_[i] = getParameters();
    // ... and set the initial values.
    for (unsigned int j = 0; j < nDim; j++)
    {
      // Hummm... that does not work when the first point is 0!!! where does this come from???
      // simplex_[i][j].setValue(getParameters()[j].getValue() * (1. + (j == i - 1 ? lambda : 0.)));
      simplex_[i][j].setValue(getParameters()[j].getValue() + (j == i - 1 ? lambda : 0.));
    }
    // Compute the corresponding f value:
    y_[i] = getFunction()->f(simplex_[i]);
    nbEval_++;
  }
  // Last function evaluation, setting current value:
  simplex_[0] = getParameters();
  y_[0] = getFunction()->f(simplex_[0]);
  nbEval_++;
 
  pSum_ = getPSum();
}
 
/******************************************************************************/
 
double DownhillSimplexMethod::doStep()
{
  // The number of dimensions of the parameter space:
  size_t nDim = simplex_.getDimension();
  size_t mpts = nDim + 1;
 
  iLowest_ = 0;
  // First we must determine which point is the highest (worst),
  // next-highest, and lowest (best), by looping over the points
  // in the simplex.
  if (y_[0] > y_[1])
  {
    iHighest_ = 0;
    iNextHighest_ = 1;
  }
  else
  {
    iHighest_ = 1;
    iNextHighest_ = 0;
  }
 
  for (unsigned int i = 0; i < mpts; i++)
  {
    if (y_[i] <= y_[iLowest_])
      iLowest_ = i;
    if (y_[i] > y_[iHighest_])
    {
      iNextHighest_ = iHighest_;
      iHighest_ = i;
    }
    else if (y_[i] > y_[iNextHighest_] && i != iHighest_)
      iNextHighest_ = i;
  }
 
  // Set current best point:
  getParameters_() = simplex_[iLowest_];
 
  // Begin a new iteration.
  // First extrapolate by a factor -1 through the face of the simplex
  // across from high point, i.e., reflect the simplex from the high point.</p>
 
  double yTry = tryExtrapolation(-1.0);
  if (yTry <= y_[iLowest_])
  {
    // Expansion.
    yTry = tryExtrapolation(2.0);
  }
  else if (yTry >= y_[iNextHighest_])
  {
    // Contraction.
    double ySave = y_[iHighest_];
    yTry = tryExtrapolation(0.5);
    if (yTry >= ySave)
    {
      for (size_t i = 0; i < mpts; i++)
      {
        if (i != iLowest_)
        {
          for (size_t j = 0; j < nDim; j++)
          {
            pSum_[j].setValue(0.5 * (simplex_[i][j].getValue() + simplex_[iLowest_][j].getValue()));
            simplex_[i][j].setValue(pSum_[j].getValue());
          }
          y_[i] = getFunction()->f(pSum_);
          nbEval_++;
        }
      }
      nbEval_ += static_cast<unsigned int>(nDim);
      pSum_ = getPSum();
    }
  }
 
  return y_[iLowest_];
}
 
/******************************************************************************/
 
double DownhillSimplexMethod::optimize()
{
  AbstractOptimizer::optimize();
 
  // set best shot:
  return getFunction()->f(simplex_[iLowest_]);
}
 
/******************************************************************************/
 
ParameterList DownhillSimplexMethod::getPSum()
{
  size_t ndim = simplex_.getDimension();
  size_t mpts = ndim + 1;
 
  // Get a copy...
  ParameterList pSum = getParameters();
  // ... and initializes it.
  for (size_t j = 0; j < ndim; j++)
  {
    double sum = 0.;
    for (size_t i = 0; i < mpts; i++)
    {
      sum += simplex_[i][j].getValue();
    }
    pSum[j].setValue(sum);
  }
  return pSum;
}
 
/******************************************************************************/
 
double DownhillSimplexMethod::tryExtrapolation(double fac)
{
  size_t ndim = simplex_.getDimension();
  double fac1, fac2, yTry;
 
  fac1 = (1.0 - fac) / static_cast<double>(ndim);
  fac2 = fac1 - fac;
 
  // Get a copy...
  ParameterList pTry = getParameters();
  // and initialize it:
  for (size_t j = 0; j < ndim; j++)
  {
    pTry[j].setValue(pSum_[j].getValue() * fac1 - simplex_[iHighest_][j].getValue() * fac2);
  }
  // Now compute the function for this new set of parameters:
  yTry = getFunction()->f(pTry);
  nbEval_++;
 
  // Then test this new point:
  if (yTry < y_[iHighest_])
  {
    y_[iHighest_] = yTry;
    for (size_t j = 0; j < ndim; j++)
    {
      pSum_[j].setValue(pSum_[j].getValue() + pTry[j].getValue() - simplex_[iHighest_][j].getValue());
      simplex_[iHighest_][j].setValue(pTry[j].getValue());
    }
  }
  return yTry;
}
 
/******************************************************************************/