nojhan/paradiseo
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

move the distrib continue in EDO adaptive algos

Browse source 
- Call the continuator right after the distribution update,
because we may have generated an ill-conditioned distribution,
which would lead to bad solution sampling.
- Reserve mem of sampled vector.
This commit is contained in:
Johann Dreo 2020-03-17 12:04:02 +01:00
commit 64e14e09df
1 changed files with 15 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

17
edo/src/edoAlgoAdaptive.h
View file

@ -160,13 +160,26 @@ public:
// (2) Estimation of the distribution parameters
_distrib = _estimator(selected_pop);
// TODO modularization: the estimator and the continuator perform
// the same decomposition twice, see how to decompose those operators
// Call the continuator right after the distribution update,
// because we may have generated an ill-conditioned distribution,
// which would lead to bad solution sampling.
if(not _distribution_continuator(_distrib)) {
break;
}
// (3) sampling
// The sampler produces feasible solutions (@see edoSampler that
// encapsulate an edoBounder)
current_pop.clear();
current_pop.clear(); current_pop.reserve(pop.size());
for( unsigned int i = 0; i < pop.size(); ++i ) {
current_pop.push_back( _sampler(_distrib) );
}
// TODO modluraziation: the sampler may generate solution that are
// not finite. See how to stop right from there instead of
// performing useless evaluations.
// (4) Evaluate new solutions
_evaluator( pop, current_pop );
@ -174,7 +187,7 @@ public:
// (5) Replace old solutions by new ones
_replacor(pop, current_pop); // e.g. copy current_pop in pop
} while( _distribution_continuator( _distrib ) && _pop_continuator( pop ) );
} while( _pop_continuator( pop ) );
} // operator()