void Sample2_Shared()
{
// (A) create a new CSample instance with one reference
boost::shared_ptr<Sample> mySample(new Sample);
std::cout << "The Sample now has references:" << mySample.use_count() << "\n"; // should be 1
// (B) assign a second pointer to it:
boost::shared_ptr<Sample> mySample2 = mySample; // should be 2 refs by now
std::cout << "The Sample now has references:" << mySample.use_count() << "\n";
// (C) set the first pointer to NULL
mySample.reset();
std::cout << "The Sample now has references: " << mySample2.use_count() << "\n"; // 1
// the object allocated in (1) is deleted automatically
// when mySample2 goes out of scope
}
void cma_es(Swarm &sw){
if(logCMAES){
printf("\n--------------------------------------------------------------\n");
printf("Repository (%d)\n", sw.repository.getActualSize());
for(int i=0;i<sw.repository.getActualSize();i++){
printVector(sw.repository.getSolution(i).decisionVector, decisionNumber);
printf("\n");
}
}
double sigmaPrev=sw.sigma;
myLearn(sw);
// if(fabs(sigmaPrev-sw.sigma) > 10000){
// fprintf(stderr, "\nWARNING!, sigma changed too much: %f -> %f. resetting...\n",sigmaPrev, sw.sigma); //shows warning message
// sw.init=true; //set the init flag, so all the CMA-ES variables are reset
// myLearn(sw); //learn again using the default values as base
// }
if(sw.sigma != sw.sigma || sw.sigma >= MAXDOUBLE){ //check for invalid numbers NaN or inf
fprintf(stderr, "WARNING!, sigma: %f. resetting...\n", sw.sigma); //shows warning message
// exit(1);
sw.init=true; //set the init flag, so all the CMA-ES variables are reset
myLearn(sw); //learn again using the default values as base
}
//four reset criteria as in the paper "injecting cma-es into moea/d"
//NoEffectCoord
for (int iKoo = 0; iKoo < decisionNumber; ++iKoo){
if (sw.mean[iKoo] == sw.mean[iKoo] + 0.2*sw.sigma*sqrt(sw.C[iKoo][iKoo])){
fprintf(stderr, "NoEffectCoordinate: standard deviation 0.2*%7.2e in coordinate %d without effect\n", sw.sigma*sqrt(sw.C[iKoo][iKoo]), iKoo); //shows warning message
sw.init=true; //set the init flag, so all the CMA-ES variables are reset
myLearn(sw); //learn again using the default values as base
break;
}
}
//NoEffectAxis
int iKoo;
for (int iAchse = 0; iAchse < decisionNumber; ++iAchse){
double fac = 0.1 * sw.sigma * sw.D[iAchse];
for (iKoo = 0; iKoo < decisionNumber; ++iKoo){
if (sw.mean[iKoo] != sw.mean[iKoo] + fac * sw.B[iKoo][iAchse])
break;
}
if (iKoo == decisionNumber){
/* t->sigma *= exp(0.2+t->sp.cs/t->sp.damps); */
fprintf(stderr, "NoEffectAxis: standard deviation 0.1*%7.2e in principal axis %d without effect\n", fac/0.1, iAchse);
sw.init=true; //set the init flag, so all the CMA-ES variables are reset
myLearn(sw); //learn again using the default values as base
break;
}
}
//TolXUp
double stopTolUpXFactor=1e3;
double initialStds=0.3;
int i;
for(i=0; i<decisionNumber; ++i) {
if (sw.sigma * sqrt(sw.C[i][i]) > stopTolUpXFactor * initialStds)
break;
}
if (i < decisionNumber) {
fprintf(stderr, "TolUpX: standard deviation increased by more than %7.2e, larger initial standard deviation recommended \n", stopTolUpXFactor);
sw.init=true; //set the init flag, so all the CMA-ES variables are reset
myLearn(sw); //learn again using the default values as base
}
//ConditionCov
double dMaxSignifKond=1e13;
if (rgdouMax(sw.D, decisionNumber) >= rgdouMin(sw.D, decisionNumber) * dMaxSignifKond) {
fprintf(stderr, "ConditionNumber: maximal condition number %7.2e reached. maxEW=%7.2e,minEW=%7.2e\n", dMaxSignifKond, rgdouMax(sw.D, decisionNumber), rgdouMin(sw.D, decisionNumber));
sw.init=true; //set the init flag, so all the CMA-ES variables are reset
myLearn(sw); //learn again using the default values as base
}
//************************************* RESAMPLE SOLUTIONS FROM THE GOOD FRONTS ******************************//
//re-learn and re-sample when errors in the covariance matrix are detected in the sampling phase
bool success=mySample(sw);
while(!success){
myLearn(sw);
success=mySample(sw);
fprintf(stderr,"Resample\n");
}
// Neighbor orderedSwarms[swarmNumber];
// for(int i=0;i<swarmNumber;i++){
// orderedSwarms[i].index=i;
// orderedSwarms[i].distance=swarms[i].modelQuality*-1;//inverted model quality because we will use the preexisting sort, that sorts based on the smallest value (distance)
// }
// std::sort(orderedSwarms, orderedSwarms+swarmNumber, neighborsComparator);
//
// // for(int i=0;i<swarmNumber;i++){
// // printf("swarm: %d has quality %f sz: %d\n", orderedSwarms[i].index, orderedSwarms[i].distance*-1, repGlobal->getActualSize());
// // }
// // printf("\n\n\n");
//
// bool good=false;
// for(int i=0;i<100;i++){
// if(sw.neighborhood[0].index == orderedSwarms[i].index){
// good=true;
// break;
// }
// }
//
// if(good){
// //re-learn and re-sample when errors in the covariance matrix are detected in the sampling phase
// bool success=mySampleReplacement(sw, 10);
// while(!success){
// myLearn(sw);
// success=mySampleReplacement(sw, 10);
// fprintf(stderr,"Resample\n");
// }
// }else{
// for(int i=0;i<sw.getSize();i++)
// sw.particles[i].solution.evalSolution=false;
// // //re-learn and re-sample when errors in the covariance matrix are detected in the sampling phase
// // bool success=mySample(sw);
// // while(!success){
// // myLearn(sw);
// // success=mySample(sw);
// // fprintf(stderr,"Resample\n");
// // }
// }
//************************************* END OF RESAMPLING SOLUTIONS FROM THE GOOD FRONTS ******************************//
if(logCMAES){
printf("\n\npop after \n\n");
for(int i=0; i<sw.getSize();i++){
//printVector(evo.rgrgx[i],N+2);
printVector(sw.particles[i].solution.decisionVector,decisionNumber);
printf("\n");
}
}
}
