---

TrainProcessor.cc

Go to the documentation of this file.

#include <limits>
#include <string>

#include <TH1.h>

#include "PhysicsTools/MVAComputer/interface/AtomicId.h"

#include "PhysicsTools/MVATrainer/interface/MVATrainer.h"
#include "PhysicsTools/MVATrainer/interface/TrainProcessor.h"

namespace PhysicsTools {

TrainProcessor::TrainProcessor(const char *name,
                               const AtomicId *id,
                               MVATrainer *trainer) :
        Source(*id), name(name), trainer(trainer), monitoring(0), monModule(0)
{
}

TrainProcessor::~TrainProcessor()
{
}

void TrainProcessor::doTrainBegin()
{
        bool booked = false;
        unsigned int nBins = 50;

        if (!monitoring) {
                const char *source = getName();
                if (source) {
                        monitoring = trainer->bookMonitor(name + "_" + source);
                        monModule = trainer->bookMonitor(std::string("input_") +
                                                         source);
                } else {
                        monModule = trainer->bookMonitor("output");
                        nBins = 400;
                }

                booked = monModule != 0;
        }

        if (booked) {
                std::vector<SourceVariable*> inputs = getInputs().get();
                for(std::vector<SourceVariable*>::const_iterator iter =
                        inputs.begin(); iter != inputs.end(); ++iter) {

                        SourceVariable *var = *iter;
                        std::string name =
                                (const char*)var->getSource()->getName()
                                + std::string("_")
                                + (const char*)var->getName();

                        SigBkg pair;
                        pair.entries[0] = pair.entries[1] = 0;
                        pair.histo[0] = monModule->book<TH1F>(name + "_bkg",
                                (name + "_bkg").c_str(),
                                (name + " background").c_str(), nBins, 0, 0);
                        pair.histo[1] = monModule->book<TH1F>(name + "_sig",
                                (name + "_sig").c_str(),
                                (name + " signal").c_str(), nBins, 0, 0);
                        pair.underflow[0] = pair.underflow[1] = 0.0;
                        pair.overflow[0] = pair.overflow[1] = 0.0;

                        pair.sameBinning = true;        // use as default
                        if (monitoring) {
                                pair.min = -std::numeric_limits<double>::infinity();
                                pair.max = +std::numeric_limits<double>::infinity();
                        } else {
                                pair.min = -99999.0;
                                pair.max = +99999.0;
                        }

                        monHistos.push_back(pair);
                }
        }

        trainBegin();
}

void TrainProcessor::doTrainData(const std::vector<double> *values,
                                 bool target, double weight,
                                 bool train, bool test)
{
        if (monModule && test) {
                for(std::vector<SigBkg>::iterator iter = monHistos.begin();
                    iter != monHistos.end(); ++iter) {
                        const std::vector<double> &vals =
                                        values[iter - monHistos.begin()];
                        for(std::vector<double>::const_iterator value =
                                vals.begin(); value != vals.end(); ++value) {

                                iter->entries[target]++;

                                if (*value <= iter->min) {
                                        iter->underflow[target] += weight;
                                        continue;
                                } else if (*value >= iter->max) {
                                        iter->overflow[target] += weight;
                                        continue;
                                }

                                iter->histo[target]->Fill(*value, weight);

                                if (iter->sameBinning)
                                        iter->histo[!target]->Fill(*value, 0);
                        }
                }
        }

        if (train)
                trainData(values, target, weight);
        if (test)
                testData(values, target, weight, train);
}

void TrainProcessor::doTrainEnd()
{
        trainEnd();

        if (monModule) {
                for(std::vector<SigBkg>::const_iterator iter =
                        monHistos.begin(); iter != monHistos.end(); ++iter) {

                        for(unsigned int i = 0; i < 2; i++) {
                                Int_t oBin = iter->histo[i]->GetNbinsX() + 1;
                                iter->histo[i]->SetBinContent(0,
                                        iter->histo[i]->GetBinContent(0) +
                                        iter->underflow[i]);
                                iter->histo[i]->SetBinContent(oBin,
                                        iter->histo[i]->GetBinContent(oBin) +
                                        iter->overflow[i]);
                                iter->histo[i]->SetEntries(iter->entries[i]);
                        }
                }

                monModule = 0;
        }
}

} // namespace PhysicsTools

---