TrainingData Class Reference

Public Member Functions
	__init__ (self, typing.Sequence[config.Particle] particles, config.FeiConfiguration config, typing.Mapping[int, typing.Mapping[str, float]] mc_counts)
pybasf2.Path	reconstruct (self)

Public Attributes
	particles = particles
	list of config.Particle objects
	config = config
	config.FeiConfiguration object
	mc_counts = mc_counts
	containing number of MC Particles

Detailed Description

Steers the creation of the training data.
The training data is used to train a multivariate classifier for each channel.
The training of the FEI at its core is just generating this training data for each channel.
After we created the training data for a stage, we have to train the classifiers (see Teacher class further down).

Definition at line 206 of file core.py.

Constructor & Destructor Documentation

__init__()

__init__	(		self,
		typing.Sequence[config.Particle]	particles,
		config.FeiConfiguration	config,
		typing.Mapping[int, typing.Mapping[str, float]]	mc_counts )

Create a new TrainingData object
@param particles list of config.Particle objects
@param config config.FeiConfiguration object
@param mc_counts containing number of MC Particles

Definition at line 214 of file core.py.

                 mc_counts: typing.Mapping[int, typing.Mapping[str, float]]):
        """
        Create a new TrainingData object
        @param particles list of config.Particle objects
        @param config config.FeiConfiguration object
        @param mc_counts containing number of MC Particles
        """
        
        self.particles = particles
        
        self.config = config
        
        self.mc_counts = mc_counts
 

Member Function Documentation

reconstruct()

pybasf2.Path reconstruct

(

self

)

Returns pybasf2.Path which creates the training data for the given particles

Definition at line 229 of file core.py.

    def reconstruct(self) -> pybasf2.Path:
        """
        Returns pybasf2.Path which creates the training data for the given particles
        """
        import ROOT  # noqa
        path = basf2.create_path()
 
        for particle in self.particles:
            pdgcode = abs(pdg.from_name(particle.name))
            nSignal = self.mc_counts[pdgcode]['sum']
            print(f"FEI-core: TrainingData: nSignal for {particle.name}: {nSignal}")
 
            # For D-Mesons we usually have a efficiency of 10^-3 including branching fraction
            if pdgcode > 400:
                nSignal /= 1000
            # For B-Mesons we usually have a efficiency of 10^-4 including branching fraction
            if pdgcode > 500:
                nSignal /= 10000
 
            for channel in particle.channels:
                weightfile = f'{channel.label}.xml'
                if basf2_mva.available(weightfile):
                    B2INFO(f"FEI-core: Skipping preparing Training Data for {weightfile}, already available")
                    continue
                filename = 'training_input.root'
 
                # nBackground = nEvents * nBestCandidates
                nBackground = self.mc_counts[0]['sum'] * channel.preCutConfig.bestCandidateCut
                inverseSamplingRates = {}
                # For some very pure channels (Jpsi), this sampling can be too aggressive and training fails.
                # It can therefore be disabled in the preCutConfig.
                if nBackground > Teacher.MaximumNumberOfMVASamples and not channel.preCutConfig.noBackgroundSampling:
                    inverseSamplingRates[0] = max(
                        1, int((int(nBackground / Teacher.MaximumNumberOfMVASamples) + 1) * channel.preCutConfig.bkgSamplingFactor))
                elif channel.preCutConfig.bkgSamplingFactor > 1:
                    inverseSamplingRates[0] = int(channel.preCutConfig.bkgSamplingFactor)
 
                if nSignal > Teacher.MaximumNumberOfMVASamples and not channel.preCutConfig.noSignalSampling:
                    inverseSamplingRates[1] = int(nSignal / Teacher.MaximumNumberOfMVASamples) + 1
 
                spectators = [channel.mvaConfig.target] + list(channel.mvaConfig.spectators.keys())
                if channel.mvaConfig.sPlotVariable is not None:
                    spectators.append(channel.mvaConfig.sPlotVariable)
 
                if self.config.monitor:
                    hist_variables = ['mcErrors', 'mcParticleStatus'] + channel.mvaConfig.variables + spectators
                    hist_variables_2d = [(x, channel.mvaConfig.target)
                                         for x in channel.mvaConfig.variables + spectators if x is not channel.mvaConfig.target]
                    hist_filename = os.path.join(self.config.monitoring_path, 'Monitor_TrainingData.root')
                    ma.variablesToHistogram(channel.name, variables=config.variables2binnings(hist_variables),
                                            variables_2d=config.variables2binnings_2d(hist_variables_2d),
                                            filename=hist_filename,
                                            ignoreCommandLineOverride=True,
                                            directory=config.removeJPsiSlash(f'{channel.label}'), path=path)
 
                teacher = basf2.register_module('VariablesToNtuple')
                teacher.set_name(f'VariablesToNtuple_{channel.name}')
                teacher.param('fileName', filename)
                teacher.param('treeName', ROOT.Belle2.MakeROOTCompatible.makeROOTCompatible(f'{channel.label} variables'))
                teacher.param('variables', channel.mvaConfig.variables + spectators)
                teacher.param('particleList', channel.name)
                teacher.param('sampling', (channel.mvaConfig.target, inverseSamplingRates))
                teacher.param('ignoreCommandLineOverride', True)
                path.add_module(teacher)
        return path
 
 

Member Data Documentation

config

config = config

config.FeiConfiguration object

Definition at line 225 of file core.py.

mc_counts

mc_counts = mc_counts

containing number of MC Particles

Definition at line 227 of file core.py.

particles

particles = particles

list of config.Particle objects

Definition at line 223 of file core.py.

---

The documentation for this class was generated from the following file:

• analysis/scripts/fei/core.py