paradiseo/eo/src/eoEvalFoundryFastGA.h

224 lines
8.5 KiB
C++

/*
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;
version 2 of the License.
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
© 2020 Thales group
Authors:
Johann Dreo <johann.dreo@thalesgroup.com>
*/
#ifndef _eoEvalFoundryFastGA_H_
#define _eoEvalFoundryFastGA_H_
#include "eoEvalFunc.h"
#include "eoAlgoFoundryFastGA.h"
#include "eoInit.h"
#include "eoPopEvalFunc.h"
/** Evaluate an algorithm assembled by an eoAlgoFoundryFastGA, encoded as a numeric vector.
*
* Allows to plug another search algorithm on top of an eoAlgoFoundryFastGA,
* so as to find the best configuration.
*
* The first template EOT is the encoding of the high-level algorithm selection problem,
* the second template SUB is the encoding of the low-level generic problem.
*
* @ingroup Evaluation
* @ingroup Foundry
*/
template<class EOT, class SUB>
class eoEvalFoundryFastGA : public eoEvalFunc<EOT>
{
public:
/** Takes the necessary parameters to perform a search on the sub-problem.
*
* @param foundry The set of algorithms among which to select.
* @param subpb_init An initilizer for sub-problem encoding.
* @param offspring_size Population size for the sub-problem solver.
* @param subpb_eval The sub-problem itself.
* @param penalization If any solution to the high-level algorithm selection problem is out of bounds, set it to this value.
*/
eoEvalFoundryFastGA(
eoAlgoFoundryFastGA<SUB>& foundry,
const size_t pop_size,
eoInit<SUB>& subpb_init,
eoPopEvalFunc<SUB>& subpb_eval,
const typename SUB::Fitness penalization,
const bool normalized = false
) :
_pop_size(pop_size),
_subpb_init(subpb_init),
_subpb_eval(subpb_eval),
_foundry(foundry),
_penalization(penalization),
_normalized(normalized),
i_crat(foundry.crossover_rates.index()),
i_crsl(foundry.crossover_selectors.index()),
i_cros(foundry.crossovers.index()),
i_afcr(foundry.aftercross_selectors.index()),
i_mrat(foundry.mutation_rates.index()),
i_musl(foundry.mutation_selectors.index()),
i_muta(foundry.mutations.index()),
i_repl(foundry.replacements.index()),
i_cont(foundry.continuators.index()),
i_offs(foundry.offspring_sizes.index())
{ }
protected:
const size_t i_crat;
const size_t i_crsl;
const size_t i_cros;
const size_t i_afcr;
const size_t i_mrat;
const size_t i_musl;
const size_t i_muta;
const size_t i_repl;
const size_t i_cont;
const size_t i_offs;
public:
/** Decode the high-level problem encoding as an array of indices.
*
* @note: If the EOT is an eoInt, this will be optimized out.
*
* May be useful for getting a solution back into an eoAlgoFoundryFastGA.
* @code
* foundry = eval.decode(pop.best_element());
* std::cout << foundry.name() << std::endl;
* auto& cont = foundry.continuator(); // Get the configured operator
* @encode
*/
std::vector<size_t> decode( const EOT& sol ) const
{
size_t crat;
size_t crsl;
size_t cros;
size_t afcr;
size_t mrat;
size_t musl;
size_t muta;
size_t repl;
size_t cont;
size_t offs;
if(_normalized) {
crat = static_cast<size_t>(std::ceil( sol[i_crat] * _foundry.crossover_rates.size() ));
crsl = static_cast<size_t>(std::ceil( sol[i_crsl] * _foundry.crossover_selectors.size() ));
cros = static_cast<size_t>(std::ceil( sol[i_cros] * _foundry.crossovers.size() ));
afcr = static_cast<size_t>(std::ceil( sol[i_afcr] * _foundry.aftercross_selectors.size() ));
mrat = static_cast<size_t>(std::ceil( sol[i_mrat] * _foundry.mutation_rates.size() ));
musl = static_cast<size_t>(std::ceil( sol[i_musl] * _foundry.mutation_selectors.size() ));
muta = static_cast<size_t>(std::ceil( sol[i_muta] * _foundry.mutations.size() ));
repl = static_cast<size_t>(std::ceil( sol[i_repl] * _foundry.replacements.size() ));
cont = static_cast<size_t>(std::ceil( sol[i_cont] * _foundry.continuators.size() ));
offs = static_cast<size_t>(std::ceil( sol[i_offs] * _foundry.offspring_sizes.size() ));
} else {
crat = static_cast<size_t>(std::ceil( sol[i_crat] ));
crsl = static_cast<size_t>(std::ceil( sol[i_crsl] ));
cros = static_cast<size_t>(std::ceil( sol[i_cros] ));
afcr = static_cast<size_t>(std::ceil( sol[i_afcr] ));
mrat = static_cast<size_t>(std::ceil( sol[i_mrat] ));
musl = static_cast<size_t>(std::ceil( sol[i_musl] ));
muta = static_cast<size_t>(std::ceil( sol[i_muta] ));
repl = static_cast<size_t>(std::ceil( sol[i_repl] ));
cont = static_cast<size_t>(std::ceil( sol[i_cont] ));
offs = static_cast<size_t>(std::ceil( sol[i_offs] ));
}
return {crat, crsl, cros, afcr, mrat, musl, muta, repl, cont, offs};
}
/** Perform a sub-problem search with the configuration encoded in the given solution
* and set its (high-level) fitness to the best (low-level) fitness found.
*
* You may want to overload this to perform multiple runs or solve multiple sub-problems.
*/
virtual void operator()(EOT& sol)
{
if(not sol.invalid()) {
return;
}
auto config = decode(sol);
double crat = config[i_crat];
double crsl = config[i_crsl];
double cros = config[i_cros];
double afcr = config[i_afcr];
double mrat = config[i_mrat];
double musl = config[i_musl];
double muta = config[i_muta];
double repl = config[i_repl];
double cont = config[i_cont];
double offs = config[i_offs];
if(
0 <= crat and crat < _foundry.crossover_rates.size()
and 0 <= crsl and crsl < _foundry.crossover_selectors.size()
and 0 <= cros and cros < _foundry.crossovers.size()
and 0 <= afcr and afcr < _foundry.aftercross_selectors.size()
and 0 <= mrat and mrat < _foundry.mutation_rates.size()
and 0 <= musl and musl < _foundry.mutation_selectors.size()
and 0 <= muta and muta < _foundry.mutations.size()
and 0 <= repl and repl < _foundry.replacements.size()
and 0 <= cont and cont < _foundry.continuators.size()
and 0 <= offs and offs < _foundry.offspring_sizes.size()
) {
_foundry.select(config);
// Reset pop
eoPop<SUB> pop;
pop.append( _pop_size, _subpb_init);
_subpb_eval(pop,pop);
// Actually perform a search
_foundry(pop);
sol.fitness( pop.best_element().fitness() );
} else {
eo::log << eo::warnings << "WARNING: encoded algo is out of bounds, penalize to: " << _penalization << std::endl;
sol.fitness( _penalization ); // penalization
}
}
protected:
const size_t _pop_size;
eoInit<SUB>& _subpb_init;
eoPopEvalFunc<SUB>& _subpb_eval;
eoAlgoFoundryFastGA<SUB>& _foundry;
const typename EOT::Fitness _penalization;
const bool _normalized;
};
/** Helper function to instanciate an eoEvalFoundryFastGA without having to indicate the template for the sub-problem encoding.
*
* The template is deduced from the constructor's parameters.
* Not sure it's more concise than a classical instanciation…
*/
template<class EOT, class SUB>
eoEvalFoundryFastGA<EOT,SUB>&
make_eoEvalFoundryFastGA(
eoInit<SUB>& subpb_init,
eoPopEvalFunc<SUB>& subpb_eval,
eoAlgoFoundryFastGA<SUB>& foundry,
const typename SUB::Fitness penalization,
const bool normalized = false )
{
return *(new eoEvalFoundryFastGA<EOT,SUB>(subpb_init, subpb_eval, foundry, penalization, normalized));
}
#endif // _eoEvalFoundryFastGA_H_