make_op.h

这是linux下的进化计算的源代码。 === === === === === === === === === === === ===== check latest news at http:

// -*- mode: c++; c-indent-level: 4; c++-member-init-indent: 8; comment-column: 35; -*-

//-----------------------------------------------------------------------------
// make_op.h - the real-valued version
// (c) Maarten Keijzer, Marc Schoenauer and GeNeura Team, 2001
/*
    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 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 along with this library; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

    Contact: [e-mail], http://geneura.ugr.es
             [e-mail]
             [e-mail]
 */
//-----------------------------------------------------------------------------

#ifndef _make_op_h
#define _make_op_h

// the operators
#include <eoOp.h>
#include <eoGenOp.h>
#include <eoCloneOps.h>
#include <eoOpContainer.h>
// combinations of simple eoOps (eoMonOp and eoQuadOp)
#include <eoProportionalCombinedOp.h>

// the specialized Real stuff
#include <es/eoReal.h>
#include <es/eoRealInitBounded.h>
#include <es/eoRealOp.h>
#include <es/eoNormalMutation.h>
  // also need the parser and param includes
#include <utils/eoParser.h>
#include <utils/eoState.h>


/*
 * This function builds the operators that will be applied to the eoReal
 *
 * It uses a parser (to get user parameters) and a state (to store the memory)
 * the last argument is an individual, needed for 2 reasons
 *     it disambiguates the call after instanciations
 *     some operator might need some private information about the indis
 *
 * This is why the template is the complete EOT even though only the fitness
 * is actually templatized here: the following only applies to bitstrings
 *
 * Note : the last parameter is an eoInit: if some operator needs some info
 *        about the gneotypes, the init has it all (e.g. bounds, ...)
 *        Simply do
 *        EOT myEO;
 *        _init(myEO);
 *        and myEO is then an ACTUAL object
*/
template <class EOT>
eoGenOp<EOT> & do_make_op(eoParameterLoader& _parser, eoState& _state, eoInit<EOT>& _init)
{
  // First, decide whether the objective variables are bounded
  eoValueParam<eoParamParamType>& boundsParam
      = _parser.getORcreateParam(eoParamParamType("(0,1)"), "objectBounds",
                                 "Bounds for variables (unbounded if absent)",
                                 'B', "Genetic Operators");

  // get initisalizer size == std::vector size
  //  eoRealInitBounded<EOT> * realInit = (eoRealInitBounded<EOT>*)(&_init);
  //  unsigned vecSize = realInit->theBounds().size();

  // get std::vector size: safer???
  EOT eoTmp;
  _init(eoTmp);
  unsigned vecSize = eoTmp.size();

  // the bounds pointer
  eoRealVectorBounds * ptBounds;
  if (_parser.isItThere(boundsParam))	// otherwise, no bounds
    {
      /////Warning: this code should probably be replaced by creating
      /////    some eoValueParam<eoRealVectorBounds> with specific implementation
      ////     in eoParser.cpp. At the moemnt, it is there (cf also make_genotype
      eoParamParamType & ppBounds = boundsParam.value(); // std::pair<std::string,std::vector<std::string> >
      // transform into a std::vector<double>
      std::vector<double> v;
      std::vector<std::string>::iterator it;
      for (it=ppBounds.second.begin(); it<ppBounds.second.end(); it++)
	{
	  istrstream is(it->c_str());
	  double r;
	  is >> r;
	  v.push_back(r);
	}
      // now create the eoRealVectorBounds object
      if (v.size() == 2) // a min and a max for all variables
	ptBounds = new eoRealVectorBounds(vecSize, v[0], v[1]);
      else				   // no time now
	throw std::runtime_error("Sorry, only unique bounds for all variables implemented at the moment. Come back later");
      // we need to give ownership of this pointer to somebody
      /////////// end of temporary code
    }
  else			   // no param for bounds was given
    ptBounds = new eoRealVectorNoBounds(vecSize); // DON'T USE eoDummyVectorNoBounds
				   // as it does not have any dimension

  // this is a temporary version(!),
  // while Maarten codes the full tree-structured general operator input
  // BTW we must leave that simple version available somehow, as it is the one
  // that 90% people use!
  eoValueParam<std::string>& operatorParam
      =  _parser.getORcreateParam(std::string("SGA"), "operator",
                                  "Description of the operator (SGA only now)",
                                  'o', "Genetic Operators");

  if (operatorParam.value() != std::string("SGA"))
    throw std::runtime_error("Sorry, only SGA-like operator available right now\n");

    // now we read Pcross and Pmut,
    // the relative weights for all crossovers -> proportional choice
    // the relative weights for all mutations -> proportional choice
    // and create the eoGenOp that is exactly
    // crossover with pcross + mutation with pmut

  eoValueParam<double>& pCrossParam
      = _parser.getORcreateParam(0.6, "pCross", "Probability of Crossover",
                                 'C', "Genetic Operators" );
  // minimum check
  if ( (pCrossParam.value() < 0) || (pCrossParam.value() > 1) )
    throw std::runtime_error("Invalid pCross");

  eoValueParam<double>& pMutParam
      = _parser.getORcreateParam(0.1, "pMut", "Probability of Mutation",
                                 'M', "Genetic Operators" );
  // minimum check
  if ( (pMutParam.value() < 0) || (pMutParam.value() > 1) )
    throw std::runtime_error("Invalid pMut");

    // the crossovers
    /////////////////
    // the parameters
  eoValueParam<double>& segmentRateParam
      = _parser.getORcreateParam(double(1.0), "segmentRate",
                                 "Relative rate for segment crossover",
                                 's', "Genetic Operators" );
  // minimum check
  if ( (segmentRateParam.value() < 0) )
    throw std::runtime_error("Invalid segmentRate");

  eoValueParam<double>& arithmeticRateParam
      = _parser.getORcreateParam(double(2.0), "arithmeticRate",
                                 "Relative rate for arithmetic crossover",
                                 'A', "Genetic Operators" );
  // minimum check
  if ( (arithmeticRateParam.value() < 0) )
    throw std::runtime_error("Invalid arithmeticRate");

    // minimum check
  bool bCross = true;
  if (segmentRateParam.value()+arithmeticRateParam.value()==0)
    {
      std::cerr << "Warning: no crossover" << std::endl;
      bCross = false;
    }

  // Create the CombinedQuadOp
  eoPropCombinedQuadOp<EOT> *ptCombinedQuadOp = NULL;
  eoQuadOp<EOT> *ptQuad = NULL;

  if (bCross)
    {
      // segment crossover for bitstring - pass it the bounds
      ptQuad = new eoSegmentCrossover<EOT>(*ptBounds);
      _state.storeFunctor(ptQuad);
      ptCombinedQuadOp = new eoPropCombinedQuadOp<EOT>(*ptQuad, segmentRateParam.value());

	// arithmetic crossover
      ptQuad = new eoArithmeticCrossover<EOT>(*ptBounds);
      _state.storeFunctor(ptQuad);
      ptCombinedQuadOp->add(*ptQuad, arithmeticRateParam.value());

      // don't forget to store the CombinedQuadOp
      _state.storeFunctor(ptCombinedQuadOp);
    }

  // the mutations
  /////////////////
  // the parameters
  eoValueParam<double> & epsilonParam
      = _parser.getORcreateParam(0.01, "epsilon", "Half-size of interval for Uniform Mutation",
                                 'e', "Genetic Operators" );
  // minimum check
  if ( (epsilonParam.value() < 0) )
    throw std::runtime_error("Invalid epsilon");

  eoValueParam<double> & uniformMutRateParam
      = _parser.getORcreateParam(1.0, "uniformMutRate",
                                 "Relative rate for uniform mutation", 'u', "Genetic Operators" );
  // minimum check
  if ( (uniformMutRateParam.value() < 0) )
    throw std::runtime_error("Invalid uniformMutRate");

  eoValueParam<double> & detMutRateParam
      = _parser.getORcreateParam(1.0, "detMutRate",
                                 "Relative rate for deterministic uniform mutation",
                                 'd', "Genetic Operators" );
  // minimum check
  if ( (detMutRateParam.value() < 0) )
    throw std::runtime_error("Invalid detMutRate");

  eoValueParam<double> & normalMutRateParam
      = _parser.getORcreateParam(1.0, "normalMutRate",
                                 "Relative rate for Gaussian mutation",
                                 'd', "Genetic Operators" );
  // minimum check
  if ( (normalMutRateParam.value() < 0) )
    throw std::runtime_error("Invalid normalMutRate");
  // and the sigma
  eoValueParam<double> & sigmaParam
      = _parser.getORcreateParam(1.0, "sigma",
                                 "Sigma (fixed) for Gaussian mutation",
                                 'S', "Genetic Operators" );
  // minimum check
  if ( (sigmaParam.value() < 0) )
    throw std::runtime_error("Invalid sigma");

    // minimum check
  bool bMut = true;
  if (uniformMutRateParam.value()+detMutRateParam.value()+normalMutRateParam.value()==0)
    {
      std::cerr << "Warning: no mutation" << std::endl;
      bMut = false;
    }
  if (!bCross && !bMut)
    throw std::runtime_error("No operator called in SGA operator definition!!!");

    // Create the CombinedMonOp
  eoPropCombinedMonOp<EOT> *ptCombinedMonOp = NULL;
  eoMonOp<EOT> *ptMon = NULL;

  if (bMut)
    {
      // uniform mutation on all components:
      // offspring(i) uniformly chosen in [parent(i)-epsilon, parent(i)+epsilon]
      ptMon = new eoUniformMutation<EOT>(*ptBounds, epsilonParam.value());
      _state.storeFunctor(ptMon);
      // create the CombinedMonOp
      ptCombinedMonOp = new eoPropCombinedMonOp<EOT>(*ptMon, uniformMutRateParam.value());

	// mutate exactly 1 component (uniformly) per individual
      ptMon = new eoDetUniformMutation<EOT>(*ptBounds, epsilonParam.value());
      _state.storeFunctor(ptMon);
      ptCombinedMonOp->add(*ptMon, detMutRateParam.value());

      // mutate all component using Gaussian mutation
      ptMon = new eoNormalMutation<EOT>(*ptBounds, sigmaParam.value());
      _state.storeFunctor(ptMon);
      ptCombinedMonOp->add(*ptMon, normalMutRateParam.value());
      _state.storeFunctor(ptCombinedMonOp);
    }

  // now build the eoGenOp:
  // to simulate SGA (crossover with proba pCross + mutation with proba pMut
  // we must construct
  //     a sequential combination of
  //          with proba 1, a proportional combination of
  //                        a QuadCopy and our crossover
  //          with proba pMut, our mutation

  // the crossover - with probability pCross
  eoProportionalOp<EOT> * cross = new eoProportionalOp<EOT> ;
  _state.storeFunctor(cross);
  ptQuad = new eoQuadCloneOp<EOT>;
  _state.storeFunctor(ptQuad);
  cross->add(*ptCombinedQuadOp, pCrossParam.value()); // user crossover
  cross->add(*ptQuad, 1-pCrossParam.value()); // clone operator

  // now the sequential
  eoSequentialOp<EOT> *op = new eoSequentialOp<EOT>;
  _state.storeFunctor(op);
  op->add(*cross, 1.0);	 // always crossover (but clone with prob 1-pCross
  op->add(*ptCombinedMonOp, pMutParam.value());

  // that's it!
  return *op;
}
#endif