Determinstic Annealing Filter (DAF) implementation. More...

#include <DAF.h>

Inheritance diagram for DAF:

Collaboration diagram for DAF:

[legend]

Public Member Functions
	DAF (std::tuple< double, double, int > annealingScheme, int minIter, int maxIter, int minIterForPval, bool useRefKalman=true, double deltaPval=1e-3, double deltaWeight=1e-3, double probCut=1e-3)
	Create DAF. More...

	DAF (bool useRefKalman=true, double deltaPval=1e-3, double deltaWeight=1e-3)
	Create DAF. More...

	DAF (AbsKalmanFitter *kalman, double deltaPval=1e-3, double deltaWeight=1e-3)
	Create DAF. More...

void	processTrackWithRep (Track tr, const AbsTrackRep rep, bool resortHits=false) override
	Process a track using the DAF.

void	setProbCut (const double prob_cut)
	Set the probability cut for the weight calculation for the hits. More...

void	addProbCut (const double prob_cut, const int measDim)
	Set the probability cut for the weight calculation for the hits for a specific measurement dimensionality.

const std::vector< double > &	getBetas () const

void	setAnnealingScheme (double bStart, double bFinal, unsigned int nSteps)
	Configure the annealing scheme. More...

void	setAnnealingScheme (double bStart, double bFinal, unsigned int nSteps, unsigned int minIter, unsigned int maxIter)
	Configure the annealing scheme with custom values of min and max iterations. More...

void	setMaxIterations (unsigned int n) override
	Set the maximum number of iterations.

void	setConvergenceDeltaWeight (double delta)
	If all weights change less than delta between two iterations, the fit is regarded as converged.

AbsKalmanFitter *	getKalman () const

virtual void	setMaxFailedHits (int val) override

virtual void	setDebugLvl (unsigned int lvl=1) override

void	getChiSquNdf (const Track tr, const AbsTrackRep rep, double &bChi2, double &fChi2, double &bNdf, double &fNdf) const

double	getChiSqu (const Track tr, const AbsTrackRep rep, int direction=-1) const

double	getNdf (const Track tr, const AbsTrackRep rep, int direction=-1) const

double	getRedChiSqu (const Track tr, const AbsTrackRep rep, int direction=-1) const

double	getPVal (const Track tr, const AbsTrackRep rep, int direction=-1) const

unsigned int	getMinIterations () const

unsigned int	getMaxIterations () const

double	getDeltaPval () const

double	getRelChi2Change () const

double	getBlowUpFactor () const

bool	getResetOffDiagonals () const

double	getBlowUpMaxVal () const

eMultipleMeasurementHandling	getMultipleMeasurementHandling () const

int	getMaxFailedHits () const

virtual void	setMinIterations (unsigned int n)
	Set the minimum number of iterations.

void	setDeltaPval (double deltaPval)
	Set Convergence criterion. More...

void	setRelChi2Change (double relChi2Change)
	@ brief Set Non-convergence criterion More...

void	setBlowUpFactor (double blowUpFactor)

void	setResetOffDiagonals (bool resetOffDiagonals)

void	setBlowUpMaxVal (double blowUpMaxVal)
	Limit the cov entries to this maximum value when blowing up the cov. Set to negative value to disable. Default is 1.E6. More...

void	setMultipleMeasurementHandling (eMultipleMeasurementHandling mmh)
	How should multiple measurements be handled?

bool	isTrackPrepared (const Track tr, const AbsTrackRep rep) const

bool	isTrackFitted (const Track tr, const AbsTrackRep rep) const

bool	canIgnoreWeights () const
	returns if the fitter can ignore the weights and handle the MeasurementOnPlanes as if they had weight 1.

void	processTrack (Track *, bool resortHits=false)
	Process all reps. More...

Protected Member Functions
const std::vector< MeasurementOnPlane * >	getMeasurements (const KalmanFitterInfo fi, const TrackPoint tp, int direction) const
	get the measurementsOnPlane taking the multipleMeasurementHandling_ into account

Protected Attributes
unsigned int	minIterations_
	Minimum number of iterations to attempt. Forward and backward are counted as one iteration.

unsigned int	maxIterations_
	Maximum number of iterations to attempt. Forward and backward are counted as one iteration.

double	deltaPval_
	Convergence criterion. More...

double	relChi2Change_
	@ brief Non-convergence criterion More...

double	blowUpFactor_
	Blow up the covariance of the forward (backward) fit by this factor before seeding the backward (forward) fit.

bool	resetOffDiagonals_
	Reset the off-diagonals to 0 when blowing up the cov.

double	blowUpMaxVal_
	Limit the cov entries to this maxuimum value when blowing up the cov.

eMultipleMeasurementHandling	multipleMeasurementHandling_
	How to handle if there are multiple MeasurementsOnPlane.

int	maxFailedHits_
	after how many failed hits (exception during construction of plane, extrapolation etc.) should the fit be cancelled. More...

unsigned int	debugLvl_

Private Member Functions
	DAF (const DAF &)

DAF &	operator= (genfit::DAF const &)

bool	calcWeights (Track trk, const AbsTrackRep rep, double beta)
	Calculate and set the weights for the next fitting pass. More...

Private Attributes
int	minIterForPval_

double	deltaWeight_

std::vector< double >	betas_

double	chi2Cuts_ [7]

std::unique_ptr< AbsKalmanFitter >	kalman_

Detailed Description

Determinstic Annealing Filter (DAF) implementation.

Author: Christian Höppner (Technische Universität München, original author); Karl Bicker (Technische Universität München)

The DAF is an iterative Kalman filter with annealing. It is capable of fitting tracks which are contaminated with noise hits. The algorithm is taken from the references R. Fruehwirth & A. Strandlie, Computer Physics Communications 120 (1999) 197-214 and CERN thesis: Dissertation by Matthias Winkler.

The weights which were assigned to the hits by the DAF are accessible in the MeasurementOnPlane objects in the KalmanFitterInfo objects.

Definition at line 49 of file DAF.h.

Constructor & Destructor Documentation

◆ DAF() [1/3]

DAF	(	std::tuple< double, double, int >	annealingScheme,
		int	minIter,
		int	maxIter,
		int	minIterForPval,
		bool	useRefKalman = `true`,
		double	deltaPval = `1e-3`,
		double	deltaWeight = `1e-3`,
		double	probCut = `1e-3`
	)

Create DAF.

Per default, use KalmanFitterRefTrack as fitter, this constructor should be used for additional configuration of the DAF, allows to provide custom parameters

Parameters

annealingScheme	Start and Final temperatures, and number of steps for the annealing scheme
minIter	Minimum number of iterations for the annealing scheme
maxIter	Maximum number of iterations for the annealing scheme
minIterForPval	Minimum number of iterations before checking the convergence by pvalue
useRefKalman	If false, use KalmanFitter as fitter.
deltaPval	Threshold value for pvalue convergence criterion
deltaWeight	Threshold value for weight convergence criterion
probCut	Probability cut for weight calculation

Definition at line 43 of file DAF.cc.

     : AbsKalmanFitter(10, deltaPval), deltaWeight_(deltaWeight), minIterForPval_(minIterForPval)
   {
     if (useRefKalman) {
       kalman_.reset(new KalmanFitterRefTrack());
       static_cast<KalmanFitterRefTrack*>(kalman_.get())->setRefitAll();
     } else
       kalman_.reset(new KalmanFitter());
  
     kalman_->setMultipleMeasurementHandling(weightedAverage);
     kalman_->setMaxIterations(1);
  
     setAnnealingScheme(std::get<0>(annealingScheme),
                        std::get<1>(annealingScheme),
                        std::get<2>(annealingScheme),
                        minIter, maxIter); // also sets maxIterations_
     setProbCut(probCut);
   }

◆ DAF() [2/3]

DAF	(	bool	useRefKalman = `true`,
		double	deltaPval = `1e-3`,
		double	deltaWeight = `1e-3`
	)

Create DAF.

Per default, use KalmanFitterRefTrack as fitter.

Parameters

useRefKalman If false, use KalmanFitter as fitter.

Definition at line 63 of file DAF.cc.

◆ DAF() [3/3]

DAF	(	AbsKalmanFitter *	kalman,
		double	deltaPval = `1e-3`,
		double	deltaWeight = `1e-3`
	)

Create DAF.

Use the provided AbsKalmanFitter as fitter.

Definition at line 79 of file DAF.cc.

Member Function Documentation

◆ calcWeights()

bool calcWeights	(	Track *	trk,
		const AbsTrackRep *	rep,
		double	beta
	)

private

Calculate and set the weights for the next fitting pass.

Return if convergence is met. The convergence criterium is the largest absolute change of all weights.

Definition at line 262 of file DAF.cc.

◆ processTrack()

void processTrack	(	Track *	tr,
		bool	resortHits = `false`
	)

inherited

Process all reps.

Start with the cardinalRep and resort the hits if necessary (and supported by the fitter)

Definition at line 25 of file AbsFitter.cc.

◆ setAnnealingScheme() [1/2]

void setAnnealingScheme	(	double	bStart,
		double	bFinal,
		unsigned int	nSteps
	)

Configure the annealing scheme.

Set a start and end temperature and the number of steps. A logarithmic sequence of temperatures will be calculated. Also sets minIterations_, maxIterations_ and #minIterForPval as a function of the number of steps

Definition at line 216 of file DAF.cc.

◆ setAnnealingScheme() [2/2]

void setAnnealingScheme	(	double	bStart,
		double	bFinal,
		unsigned int	nSteps,
		unsigned int	minIter,
		unsigned int	maxIter
	)

Configure the annealing scheme with custom values of min and max iterations.

Set a start and end temperature and the number of steps. A logarithmic sequence of temperatures will be calculated. Also sets minIterations_ and maxIterations_ to the values provided instead of values depending on nSteps

Definition at line 239 of file DAF.cc.

◆ setBlowUpMaxVal()

void setBlowUpMaxVal ( double blowUpMaxVal )

inlineinherited

Limit the cov entries to this maximum value when blowing up the cov. Set to negative value to disable. Default is 1.E6.

This is especially useful for fits where the measurements do not constrain one direction, e.g. parallel wire measurements. The fit will not be constrained along the wire direction. This also means that the covariance along the wire direction will not get smaller during the fit. However, it will be blown up before the next iteration, leading to an exponential growth of the covariance element(s) corresponding to the wire direction. This can then lead to numerical problems. To prevent this, the maximum value of the covariance elements after blowing up can be limited.

Definition at line 119 of file AbsKalmanFitter.h.

◆ setDeltaPval()

void setDeltaPval ( double deltaPval )

inlineinherited

Set Convergence criterion.

if track total P-value changes less than this between consecutive iterations, consider the track converged. chi² from the backwards fit is used.

Definition at line 95 of file AbsKalmanFitter.h.

◆ setProbCut()

void setProbCut ( const double prob_cut )

Set the probability cut for the weight calculation for the hits.

By default the cut values for measurements of dimensionality from 1 to 5 are calculated. If you what to have cut values for an arbitrary measurement dimensionality use addProbCut(double prob_cut, int maxDim);

Definition at line 194 of file DAF.cc.

◆ setRelChi2Change()

void setRelChi2Change ( double relChi2Change )

inlineinherited

@ brief Set Non-convergence criterion

if the relative chi^2 between two iterations is larger than relChi2Change_, the fit is NOT converged and further iterations will be done, even if the deltaPval_ convergence criterium is met. This is especially useful for fits which have a p-value of almost 0 (possibly due to bad start values), where the p-value from one iteration to the next might not change much. However, a significant change in chi^2 tells us, that the fit might not yet be converged.

Definition at line 106 of file AbsKalmanFitter.h.

Member Data Documentation

◆ deltaPval_

double deltaPval_

protectedinherited

Convergence criterion.

if track total P-value changes less than this between consecutive iterations, consider the track converged. chi² from the backwards fit is used.

Definition at line 148 of file AbsKalmanFitter.h.

◆ maxFailedHits_

int maxFailedHits_

protectedinherited

after how many failed hits (exception during construction of plane, extrapolation etc.) should the fit be cancelled.

-1 means don't cancel

Definition at line 172 of file AbsKalmanFitter.h.

◆ relChi2Change_

double relChi2Change_

protectedinherited

@ brief Non-convergence criterion

if the relative chi^2 between two iterations is larger than relChi2Change_, the fit is NOT converged and further iterations will be done, even if the deltaPval_ convergence criterium is met. This is especially useful for fits which have a p-value of almost 0 (possibly due to bad start values), where the p-value from one iteration to the next might not change much. However, a significant change in chi^2 tells us, that the fit might not yet be converged.

Definition at line 158 of file AbsKalmanFitter.h.

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

genfit2/code2/fitters/include/DAF.h
genfit2/code2/fitters/src/DAF.cc

Public Member Functions

Protected Member Functions

Protected Attributes

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ DAF() [1/3]

◆ DAF() [2/3]

◆ DAF() [3/3]

Member Function Documentation

◆ calcWeights()

◆ processTrack()

◆ setAnnealingScheme() [1/2]

◆ setAnnealingScheme() [2/2]

◆ setBlowUpMaxVal()

◆ setDeltaPval()

◆ setProbCut()

◆ setRelChi2Change()

Member Data Documentation

◆ deltaPval_

◆ maxFailedHits_

◆ relChi2Change_