void StatsCalcFitnessNMSEOp::calculateStatsDeme(Stats &outStats, Deme &ioDeme, Context &ioContext) const { Beagle_StackTraceBeginM(); outStats.clear(); outStats.clearItems(); int outputs = mConfig->getOutputs().size(); outStats.addItem("processed", ioContext.getProcessedDeme()); outStats.addItem("total-processed", ioContext.getTotalProcessedDeme()); if(ioDeme.size() == 0) { outStats.setGenerationValues(Beagle::string("deme")+uint2str(ioContext.getDemeIndex()+1), ioContext.getGeneration(), 0, true); outStats.resize(3+outputs*2); outStats[0].mId = "NMSE"; outStats[0].mAvg = 0.0; outStats[0].mStd = 0.0; outStats[0].mMax = 0.0; outStats[0].mMin = 0.0; outStats[1].mId = "treedepth"; outStats[1].mAvg = 0.0; outStats[1].mStd = 0.0; outStats[1].mMax = 0.0; outStats[1].mMin = 0.0; outStats[2].mId = "treesize"; outStats[2].mAvg = 0.0; outStats[2].mStd = 0.0; outStats[2].mMax = 0.0; outStats[2].mMin = 0.0; for(int i=0; i<outputs; ++i) { int nr = 3 + i*2; outStats[nr].mId = "y" + int2str(i) + "-NMSE"; outStats[nr].mAvg = 0.0; outStats[nr].mStd = 0.0; outStats[nr].mMax = 0.0; outStats[nr].mMin = 0.0; ++nr; outStats[nr].mId = "y" + int2str(i) + "-MSE"; outStats[nr].mAvg = 0.0; outStats[nr].mStd = 0.0; outStats[nr].mMax = 0.0; outStats[nr].mMin = 0.0; } return; } const GP::Deme& gpDeme = castObjectT<const GP::Deme&>(ioDeme); const FitnessNMSE::Handle firstIndivFitness = castHandleT<FitnessNMSE>(ioDeme[0]->getFitness()); if(ioDeme.size() == 1) { outStats.setGenerationValues(Beagle::string("deme")+uint2str(ioContext.getDemeIndex()+1), ioContext.getGeneration(), 1, true); outStats.resize(3+outputs*2); outStats[0].mId = "NMSE"; outStats[0].mAvg = firstIndivFitness->getNMSE(); outStats[0].mStd = 0.0; outStats[0].mMax = firstIndivFitness->getNMSE(); outStats[0].mMin = firstIndivFitness->getNMSE(); outStats[1].mId = "treedepth"; outStats[1].mAvg = gpDeme[0]->getMaxTreeDepth(); outStats[1].mStd = 0.0; outStats[1].mMax = outStats[1].mAvg; outStats[1].mMin = outStats[1].mAvg; outStats[2].mId = "treesize"; outStats[2].mAvg = gpDeme[0]->getTotalNodes(); outStats[2].mStd = 0.0; outStats[2].mMax = outStats[2].mAvg; outStats[2].mMin = outStats[2].mAvg; for(int i=0; i<outputs; ++i) { int nr = 3 + i*2; double nmse = (*firstIndivFitness)[i].nmse; double mse = (*firstIndivFitness)[i].mse; outStats[nr].mId = "y" + int2str(i) + "-NMSE"; outStats[nr].mAvg = nmse; outStats[nr].mStd = 0.0; outStats[nr].mMax = nmse; outStats[nr].mMin = nmse; ++nr; outStats[nr].mId = "y" + int2str(i) + "-MSE"; outStats[nr].mAvg = mse; outStats[nr].mStd = 0.0; outStats[nr].mMax = mse; outStats[nr].mMin = mse; } return; } double sum = firstIndivFitness->getNMSE(); double pow2sum = pow2Of<double>(sum); double max = sum; double min = sum; std::vector<double> sumNMSE(outputs, 0); std::vector<double> pow2sumNMSE(outputs, 0); std::vector<double> maxNMSE(outputs, 0); std::vector<double> minNMSE(outputs, 0); std::vector<double> sumMSE(outputs, 0); std::vector<double> pow2sumMSE(outputs, 0); std::vector<double> maxMSE(outputs, 0); std::vector<double> minMSE(outputs, 0); for(int i=0; i<outputs; ++i) { double nmse = (*firstIndivFitness)[i].nmse; double mse = (*firstIndivFitness)[i].mse; sumNMSE[i] = nmse; pow2sumNMSE[i] = pow2Of<double>(nmse); maxNMSE[i] = nmse; minNMSE[i] = nmse; sumMSE[i] = mse; pow2sumMSE[i] = pow2Of<double>(mse); maxMSE[i] = mse; minMSE[i] = mse; } unsigned int lMaxDepth = gpDeme[0]->getMaxTreeDepth(); unsigned int lMinDepth = lMaxDepth; double lSumDepth = (double)lMaxDepth; double lPow2SumDepth = pow2Of<double>(lSumDepth); unsigned int lMaxSize = gpDeme[0]->getTotalNodes(); unsigned int lMinSize = lMaxSize; double lSumSize = (double)lMaxSize; double lPow2SumSize = pow2Of<double>(lSumSize); unsigned indivs = 0; for(unsigned int i=1; i<ioDeme.size(); i++) { const FitnessNMSE::Handle indivFitness = castHandleT<FitnessNMSE>(ioDeme[i]->getFitness()); double tNmse = indivFitness->getNMSE(); //cout << endl << i << "\t" << tNmse << "\t" << indivFitness->size(); if(tNmse != tNmse || tNmse == numeric_limits<double>::infinity()) { continue; } ++indivs; sum += tNmse; pow2sum += pow2Of<double>(tNmse); max = maxOf<double>(max, tNmse); min = minOf<double>(min, tNmse); for(int j=0; j<outputs; ++j) { double nmse = (*indivFitness)[j].nmse; double mse = (*indivFitness)[j].mse; sumNMSE[j] += nmse; pow2sumNMSE[j] += pow2Of<double>(nmse); maxNMSE[j] = maxOf<double>(maxNMSE[j], nmse); minNMSE[j] = minOf<double>(minNMSE[j], nmse); sumMSE[j] += mse; pow2sumMSE[j] += pow2Of<double>(mse); maxMSE[j] = maxOf<double>(maxMSE[j], mse); minMSE[j] = minOf<double>(minMSE[j], mse); } unsigned int lTmpDepth = gpDeme[i]->getMaxTreeDepth(); lSumDepth += (double)lTmpDepth; lPow2SumDepth += pow2Of<double>((double)lTmpDepth); lMaxDepth = maxOf(lMaxDepth, lTmpDepth); lMinDepth = minOf(lMinDepth, lTmpDepth); unsigned int lTmpSize = gpDeme[i]->getTotalNodes(); lSumSize += (double)lTmpSize; lPow2SumSize += pow2Of<double>((double)lTmpSize); lMaxSize = maxOf(lMaxSize, lTmpSize); lMinSize = minOf(lMinSize, lTmpSize); } float lDepthStdError = (float)(lPow2SumDepth - (pow2Of<double>(lSumDepth) / indivs)) / (indivs - 1); lDepthStdError = sqrt(lDepthStdError); float lSizeStdError = (float)(lPow2SumSize - (pow2Of<double>(lSumSize) / indivs)) / (indivs - 1); lSizeStdError = sqrt(lSizeStdError); outStats.setGenerationValues(Beagle::string("deme")+uint2str(ioContext.getDemeIndex()+1), ioContext.getGeneration(), ioDeme.size(), true); outStats.resize(3+outputs*2); outStats[0].mId = "NMSE"; outStats[0].mAvg = sum / indivs; outStats[0].mStd = sqrt((pow2sum - (pow2Of<double>(sum)/indivs)) / (indivs-1)); outStats[0].mMax = max; outStats[0].mMin = min; outStats[1].mId = "treedepth"; outStats[1].mAvg = (float)(lSumDepth / indivs); outStats[1].mStd = lDepthStdError; outStats[1].mMax = (float)lMaxDepth; outStats[1].mMin = (float)lMinDepth; outStats[2].mId = "treesize"; outStats[2].mAvg = (float)(lSumSize / indivs); outStats[2].mStd = lSizeStdError; outStats[2].mMax = (float)lMaxSize; outStats[2].mMin = (float)lMinSize; for(int j=0; j<outputs; ++j) { int nr = 3 + j*2; outStats[nr].mId = "y" + int2str(j) + "-NMSE"; outStats[nr].mAvg = sumNMSE[j] / indivs; outStats[nr].mStd = sqrt((pow2sumNMSE[j] - (pow2Of<double>(sumNMSE[j])/indivs)) / (indivs-1)); outStats[nr].mMax = maxNMSE[j]; outStats[nr].mMin = minNMSE[j]; ++nr; outStats[nr].mId = "y" + int2str(j) + "-MSE"; outStats[nr].mAvg = sumMSE[j] / indivs; outStats[nr].mStd = sqrt((pow2sumMSE[j] - (pow2Of<double>(sumMSE[j])/indivs)) / (indivs-1)); outStats[nr].mMax = maxMSE[j]; outStats[nr].mMin = minMSE[j]; } Beagle_StackTraceEndM("void SinsGP::StatsCalcFitnessNMSEOp::calculateStatsDeme(Beagle::Stats& outStats, Beagle::Deme& ioDeme, Beagle::Context& ioContext) const"); }
/*! * \brief Calculate statistics of a given vivarium. * \param outStats Computed stats of the deme. * \param ioVivarium Vivarium to evaluate the statistics. * \param ioContext Context of the evolution. */ void StatsCalculateOp::calculateStatsVivarium(Stats& outStats, Vivarium& ioVivarium, Context& ioContext) const { Beagle_StackTraceBeginM(); outStats.clear(); outStats.clearItems(); unsigned int lPopSize = 0; for(unsigned int i=0; i<ioVivarium.size(); ++i) { Beagle::Stats::Handle lDemeStats = ioVivarium[i]->getStats(); Beagle_NonNullPointerAssertM(lDemeStats); lPopSize += lDemeStats->getPopSize(); } outStats.setGenerationValues("vivarium", ioContext.getGeneration(), lPopSize, true); outStats.addItem("processed", ioContext.getProcessedVivarium()); outStats.addItem("total-processed", ioContext.getTotalProcessedVivarium()); if(ioVivarium.size() == 0) return; outStats.resize(ioVivarium[0]->getStats()->size()); for(unsigned int i=0; i<outStats.size(); ++i) { double lSumXi = 0.0; double lSumXiPow2 = 0.0; double lMax = (*ioVivarium[0]->getStats())[i].mMax; double lMin = (*ioVivarium[0]->getStats())[i].mMin; for(unsigned int j=0; j<ioVivarium.size(); ++j) { Beagle_AssertM(outStats.size() == ioVivarium[j]->getStats()->size()); const Measure& lMeasure = (*ioVivarium[j]->getStats())[i]; unsigned int lDemeSize = ioVivarium[j]->getStats()->getPopSize(); if(lDemeSize != 0) { double lSumFit = lMeasure.mAvg * lDemeSize; double lTmpVar1 = pow2Of<double>(lSumFit) / lDemeSize; double lTmpVar2 = pow2Of<double>(lMeasure.mStd) * (lDemeSize-1); lSumXi += lSumFit; lSumXiPow2 += lTmpVar1 + lTmpVar2; lMax = maxOf<double>(lMax, lMeasure.mMax); lMin = minOf<double>(lMin, lMeasure.mMin); } } outStats[i].mID = (*ioVivarium[0]->getStats())[i].mID; if(lPopSize == 0) { outStats[i].mAvg = 0.0; outStats[i].mStd = 0.0; outStats[i].mMax = 0.0; outStats[i].mMin = 0.0; } else if(lPopSize == 1) { outStats[i].mAvg = lSumXi; outStats[i].mStd = 0.0; outStats[i].mMax = lMax; outStats[i].mMin = lMin; } else { double lAverage = lSumXi / lPopSize; double lVariance = lSumXiPow2 - (pow2Of<double>(lAverage*((double)lPopSize)) / ((double)lPopSize)); double lStdError = sqrt(lVariance / (lPopSize - 1)); outStats[i].mAvg = lAverage; outStats[i].mStd = lStdError; outStats[i].mMax = lMax; outStats[i].mMin = lMin; } } Beagle_StackTraceEndM("void StatsCalculateOp::calculateStatsVivarium(Stats& outStats, Vivarium& ioVivarium, Context& ioContext) const"); }