genopheno.h

 
#ifndef GENOPHENO_H
#define GENOPHENO_H
 
#include <libcmaes/noboundstrategy.h>
#include <libcmaes/pwq_bound_strategy.h>
#include <libcmaes/scaling.h>
#include <vector>
 
namespace libcmaes
{
  typedef std::function<void (const double*, double*, const int&)> TransFunc;
  
  template <class TBoundStrategy=NoBoundStrategy,class TScalingStrategy=NoScalingStrategy>
    class GenoPheno
    {
      friend class CMASolutions;
      
    public:
    GenoPheno()
    :_id(true)
    {}
 
    GenoPheno(TransFunc &genof, TransFunc &phenof)
    :_genof(genof),_phenof(phenof),_id(false)
    {}
    
    GenoPheno(const double *lbounds, const double *ubounds, const int &dim)
    :_boundstrategy(lbounds,ubounds,dim),_id(true),_scalingstrategy(lbounds,ubounds,dim)
    {
      if (_scalingstrategy._id)
    _boundstrategy = TBoundStrategy(lbounds,ubounds,dim);
      else
    {
      std::vector<double> lb(dim,_scalingstrategy._intmin);
      std::vector<double> ub(dim,_scalingstrategy._intmax);
      _boundstrategy = TBoundStrategy(&lb.front(),&ub.front(),lbounds,ubounds,dim);
    }
    }
 
    GenoPheno(TransFunc &genof, TransFunc &phenof,
          const double *lbounds, const double *ubounds, const int &dim)
    :_boundstrategy(lbounds,ubounds,dim),_genof(genof),_phenof(phenof),_id(false),_scalingstrategy(lbounds,ubounds,dim)
    {
      if (_scalingstrategy._id)
    _boundstrategy = TBoundStrategy(lbounds,ubounds,dim);
      else
    {
      std::vector<double> lb(dim,_scalingstrategy._intmin);
      std::vector<double> ub(dim,_scalingstrategy._intmax);
      _boundstrategy = TBoundStrategy(&lb.front(),&ub.front(),lbounds,ubounds,dim);
    }
    }
 
    GenoPheno(const dVec &scaling,
          const dVec &shift,
          const double *lbounds=nullptr,
          const double *ubounds=nullptr)
    :_id(true)
    {
      (void)scaling;
      (void)shift;
      (void)lbounds;
      (void)ubounds;
    }
    
    ~GenoPheno() {}
 
    private:
    dMat pheno_candidates(const dMat &candidates) const
    {
      if (!_id)
    {
      dMat ncandidates = dMat(candidates.rows(),candidates.cols());
#pragma omp parallel for if (candidates.cols() >= 100)
      for (int i=0;i<candidates.cols();i++)
        {
          dVec ext = dVec(candidates.rows());
          _phenof(candidates.col(i).data(),ext.data(),candidates.rows());
          ncandidates.col(i) = ext;
        }
      return ncandidates;
    }
      return candidates;
    }
 
    dMat geno_candidates(const dMat &candidates) const
    {
      if (!_id)
    {
      dMat ncandidates = dMat(candidates.rows(),candidates.cols());
#pragma omp parallel for if (candidates.cols() >= 100)
      for (int i=0;i<candidates.cols();i++)
        {
          dVec in = dVec(candidates.rows());
          _genof(candidates.col(i).data(),in.data(),candidates.rows());
          ncandidates.col(i) = in;
        }
      return ncandidates;
    }
      return candidates;
    }
    
    public:
    dMat pheno(const dMat &candidates) const
    {
      // apply custom pheno function.
      dMat ncandidates = pheno_candidates(candidates);
 
      // apply bounds.
#pragma omp parallel for if (ncandidates.cols() >= 100)
      for (int i=0;i<ncandidates.cols();i++)
    {
      dVec ycoli;
      _boundstrategy.to_f_representation(ncandidates.col(i),ycoli);
      ncandidates.col(i) = ycoli;
    }
      
      // apply scaling.
      if (!_scalingstrategy._id)
    {
#pragma omp parallel for if (ncandidates.cols() >= 100)
      for (int i=0;i<ncandidates.cols();i++)
        {
          dVec ycoli;
          _scalingstrategy.scale_to_f(ncandidates.col(i),ycoli);
          ncandidates.col(i) = ycoli;
        }
    }
      return ncandidates;
    }
 
    dMat geno(const dMat &candidates) const
    {
      // reverse scaling.
      dMat ncandidates = candidates;
      if (!_scalingstrategy._id)
    {
#pragma omp parallel for if (ncandidates.cols() >= 100)
      for (int i=0;i<ncandidates.cols();i++)
        {
          dVec ycoli;
          _scalingstrategy.scale_to_internal(ycoli,ncandidates.col(i));
          ncandidates.col(i) = ycoli;
        }
    }
      
      // reverse bounds.
#pragma omp parallel for if (ncandidates.cols() >= 100)
      for (int i=0;i<ncandidates.cols();i++)
    {
      dVec ycoli;
      _boundstrategy.to_internal_representation(ycoli,ncandidates.col(i));
      ncandidates.col(i) = ycoli;
    }
      
      // apply custom geno function.
      ncandidates = geno_candidates(ncandidates);
      return ncandidates;
    }
    
    dVec pheno(const dVec &candidate) const
    {
      // apply custom pheno function.
      dVec ncandidate;
      if (!_id)
    {
      ncandidate = dVec(candidate.rows());
      _phenof(candidate.data(),ncandidate.data(),candidate.rows());
    }
      
      // apply bounds.
      dVec phen = dVec::Zero(candidate.rows());
      if (_id)
    _boundstrategy.to_f_representation(candidate,phen);
      else _boundstrategy.to_f_representation(ncandidate,phen);
      
      // apply scaling.
      if (!_scalingstrategy._id)
    {
      dVec sphen = dVec::Zero(phen.rows());
      _scalingstrategy.scale_to_f(phen,sphen);
      phen = sphen;
    }
      return phen;
    }
    
    dVec geno(const dVec &candidate) const
    {
      dVec ccandidate = candidate;
      dVec gen = dVec::Zero(candidate.rows());
      
      // reverse scaling.
      if (!_scalingstrategy._id)
    {
      _scalingstrategy.scale_to_internal(gen,candidate);
      ccandidate = gen;
    }
      
      // reverse bounds.
      _boundstrategy.to_internal_representation(gen,ccandidate);
      
      // apply custom geno function.
      if (!_id)
    {
      dVec ncandidate(gen.rows());
      _genof(gen.data(),ncandidate.data(),gen.rows());
      return ncandidate;
    }
      else return gen;
    }
 
    TBoundStrategy get_boundstrategy() const { return _boundstrategy; }
      
    TBoundStrategy& get_boundstrategy_ref() { return _boundstrategy; }
      
    TScalingStrategy get_scalingstrategy() const { return _scalingstrategy; }
 
    void remove_dimensions(const std::vector<int> &k)
      {
    if (!_scalingstrategy.is_id())
      _scalingstrategy.remove_dimensions(k);
        if (!_boundstrategy.is_id())
      _boundstrategy.remove_dimensions(k);
      }
 
    private:
    TBoundStrategy _boundstrategy;
    TransFunc _genof;
    TransFunc _phenof;
    bool _id = false; 
    TScalingStrategy _scalingstrategy;
  };
 
  // specialization when no bound strategy nor scaling applies.
  template<> inline dMat GenoPheno<NoBoundStrategy,NoScalingStrategy>::pheno(const dMat &candidates) const
    {
      if (_id)
    return candidates;
      else return pheno_candidates(candidates);
    }
  template<> inline dVec GenoPheno<NoBoundStrategy,NoScalingStrategy>::pheno(const dVec &candidate) const
    {
      if (_id)
    return candidate;
      else
    {
      dVec ncandidate(candidate.rows());
      _phenof(candidate.data(),ncandidate.data(),candidate.rows());
      return ncandidate;
    }
    }
  template<> inline dVec GenoPheno<NoBoundStrategy,NoScalingStrategy>::geno(const dVec &candidate) const
    {
      if (_id)
    return candidate;
      else
    {
      dVec ncandidate(candidate.rows());
      _genof(candidate.data(),ncandidate.data(),candidate.rows());
      return ncandidate;
    }
    }
 
  template<> inline dVec GenoPheno<NoBoundStrategy,linScalingStrategy>::pheno(const dVec &candidate) const
    {
      dVec ncandidate(candidate.rows());
      if (!_id)
    _phenof(candidate.data(),ncandidate.data(),candidate.rows());
      
      dVec sphen;
      if (!_id)
    _scalingstrategy.scale_to_f(ncandidate,sphen);
      else _scalingstrategy.scale_to_f(candidate,sphen);
      return sphen;
    }
  template<> inline dVec GenoPheno<NoBoundStrategy,linScalingStrategy>::geno(const dVec &candidate) const
    {
      dVec scand = dVec::Zero(candidate.rows());
      _scalingstrategy.scale_to_internal(scand,candidate);
      if (_id)
    return scand;
      else
    {
      dVec ncandidate(candidate.rows());
      _genof(scand.data(),scand.data(),candidate.rows());
      return ncandidate;
    }
    }
  template<> inline dMat GenoPheno<NoBoundStrategy,linScalingStrategy>::pheno(const dMat &candidates) const
    {
      dMat ncandidates;
      if (!_id)
    ncandidates = pheno_candidates(candidates);
      else ncandidates = candidates;
      
      // apply scaling.
#pragma omp parallel for if (ncandidates.cols() >= 100)
      for (int i=0;i<ncandidates.cols();i++)
    {
      dVec ycoli;
      _scalingstrategy.scale_to_f(ncandidates.col(i),ycoli);
      ncandidates.col(i) = ycoli;
    }
      return ncandidates;
    }
  
  template<> inline GenoPheno<NoBoundStrategy,linScalingStrategy>::GenoPheno(const dVec &scaling,
                                         const dVec &shift,
                                         const double *lbounds,
                                         const double *ubounds)
    :_id(true)
    {
      (void)lbounds;
      (void)ubounds;
      _scalingstrategy = linScalingStrategy(scaling,shift);
    }
  
  template<> inline GenoPheno<pwqBoundStrategy,linScalingStrategy>::GenoPheno(const dVec &scaling,
                                          const dVec &shift,
                                          const double *lbounds,
                                          const double *ubounds)
    :_id(true)
    {
      _scalingstrategy = linScalingStrategy(scaling,shift);
      if (lbounds == nullptr || ubounds == nullptr)
    return;
      dVec vlbounds = Eigen::Map<dVec>(const_cast<double*>(lbounds),scaling.size());
      dVec vubounds = Eigen::Map<dVec>(const_cast<double*>(ubounds),scaling.size());
      dVec nlbounds, nubounds;
      _scalingstrategy.scale_to_internal(nlbounds,vlbounds);
      _scalingstrategy.scale_to_internal(nubounds,vubounds);
      _boundstrategy = pwqBoundStrategy(nlbounds.data(),nubounds.data(),scaling.size());
    }
}
 
#endif