Class to keep all parameters of an expert MLP for the neuro trigger. More...

#include <CDCTriggerMLP.h>

Inheritance diagram for CDCTriggerMLP:

Public Member Functions
	CDCTriggerMLP ()
	default constructor.

	CDCTriggerMLP (std::vector< unsigned short > &nodes, unsigned short targets, std::vector< float > &outputscale, std::vector< float > &phirangeUse, std::vector< float > &invptrangeUse, std::vector< float > &thetarangeUse, std::vector< float > &phirangeTrain, std::vector< float > &invptrangeTrain, std::vector< float > &thetarangeTrain, unsigned short maxHits, unsigned long pattern, unsigned long patternMask, unsigned short tmax, const std::string &etoption)
	constructor to set all parameters (not weights and relevantID ranges).

	~CDCTriggerMLP ()
	destructor, empty because we don't allocate memory anywhere.

bool	isTrained () const
	check if weights are default values or set by some trainer

unsigned	nLayers () const
	get number of layers

unsigned	nNodesLayer (unsigned iLayer) const
	get number of nodes in a layer

unsigned	nWeights () const
	get number of weights from length of weights vector

unsigned	nWeightsCal () const
	calculate number of weights from number of nodes

std::vector< float >	getWeights () const
	get weights vector

void	setWeights (std::vector< float > xweights)
	set weights vector

unsigned short	getMaxHitsPerSL () const
	get maximum hit number for a single super layer

unsigned long	getSLpattern () const
	get super layer pattern

unsigned long	getSLpatternMask () const
	get bitmask for super layer pattern

unsigned long	getSLpatternUnmasked () const
	get raw super layer pattern

unsigned short	getTMax () const
	get maximal drift time

std::vector< float >	getIDRange (unsigned iSL) const
	get relevant ID range for given super layer

void	setRelID (std::vector< float > relid)
	set and get total relevant ID range

std::vector< float >	getRelID () const

std::string	get_et_option () const
	Returns way of obtaining the event time.

bool	inPhiRangeUse (float phi) const
	check whether given phi value is in sector

bool	inPtRangeUse (float pt) const
	check whether given pt value is in sector

bool	inInvptRangeUse (float invpt) const
	check whether given 1/pt value is in sector

bool	inThetaRangeUse (float theta) const
	check whether given theta value is in sector

bool	inPhiRangeTrain (float phi) const
	check whether given phi value is in training sector

bool	inPtRangeTrain (float pt) const
	check whether given pt value is in training sector

bool	inInvptRangeTrain (float invpt) const
	check whether given 1/pt value is in training sector

bool	inThetaRangeTrain (float theta) const
	check whether given theta value is in training sector

bool	isRelevant (float relId, unsigned iSL) const
	check whether given relative TS ID is in relevant range

float	scaleId (double relId, unsigned iSL) const
	scale relative TS ID from relevant range to approximately [-1, 1] (to facilitate the FPGA implementation, the scale factor is rounded to a power of 2)

std::vector< float >	scaleTarget (std::vector< float > target) const
	scale target value from outputScale to [-1, 1]

std::vector< float >	unscaleTarget (std::vector< float > target) const
	scale target value from [-1, 1] to outputScale

int	zIndex () const
	get target index for z (-1 if no output is trained for z)

int	thetaIndex () const
	get target index for theta (-1 if no output is trained for theta)

Private Member Functions
	ClassDef (CDCTriggerMLP, 11)
	Needed to make the ROOT object storable.

Private Attributes
std::vector< unsigned short >	nNodes
	Number of nodes in each layer, not including bias nodes.

std::vector< float >	weights
	Weights of the network.

bool	trained
	Indicator whether the weights are just default values or have been set by some trainer (set to true when setWeights() is first called).

unsigned short	targetVars
	output variables: 1: z, 2: theta, 3: (z, theta)

std::vector< float >	outputScale
	Output[i] of the MLP is scaled from [-1, 1] to [outputScale[2i], outputScale[2i+1]].

std::vector< float >	phiRangeUse
	Phi region in radian for which this expert is used.

std::vector< float >	invptRangeUse
	Charge / Pt region in 1/GeV for which this expert is used.

std::vector< float >	thetaRangeUse
	Theta region in radian for which this expert is trained.

std::vector< float >	phiRangeTrain
	Phi region in radian for which this expert is used.

std::vector< float >	invptRangeTrain
	Charge / Pt region in 1/GeV for which this expert is trained.

std::vector< float >	thetaRangeTrain
	Theta region in radian for which this expert is trained.

unsigned short	maxHitsPerSL
	Maximum number of inputs for a single super layer.

unsigned long	SLpattern
	Super layer pattern for which this expert is trained.

unsigned long	SLpatternMask
	Bitmask for comparing the super layer pattern.

unsigned short	tMax
	Maximal drift time (for scaling), hits with larger values are ignored.

std::vector< float >	relevantID
	Hits must be within ID region around 2D track to be used as input.

std::string	et_option
	Returns way of obtaining the event time.

Friends
class	CDCTriggerNeuroTrainerModule

class	NeuroTrigger

Detailed Description

Class to keep all parameters of an expert MLP for the neuro trigger.

Definition at line 20 of file CDCTriggerMLP.h.

Constructor & Destructor Documentation

◆ CDCTriggerMLP() [1/2]

CDCTriggerMLP ( )

default constructor.

Definition at line 13 of file CDCTriggerMLP.cc.

                            :
  nNodes{27, 27, 2}, trained(false), targetVars(3), outputScale{ -1., 1., -1., 1.},
  phiRangeUse{0., 2. * M_PI}, invptRangeUse{ -5., 5.}, thetaRangeUse{0., M_PI},
  phiRangeTrain{0., 2. * M_PI}, invptRangeTrain{ -5., 5.}, thetaRangeTrain{0., M_PI},
  maxHitsPerSL(1), SLpattern(0), SLpatternMask(0), tMax(256),
  relevantID{ -1., 1.,
              -10., 1.,
              -1., 1.,
              -1., 10.,
              -1., 1.,
              -10.5, 1.,
              -1., 1.,
              -1., 11.,
              -1., 1.},
    et_option("etf_or_fastestpriority")
{
  weights.assign(nWeightsCal(), 0.);
}

◆ CDCTriggerMLP() [2/2]

CDCTriggerMLP	(	std::vector< unsigned short > &	nodes,
		unsigned short	targets,
		std::vector< float > &	outputscale,
		std::vector< float > &	phirangeUse,
		std::vector< float > &	invptrangeUse,
		std::vector< float > &	thetarangeUse,
		std::vector< float > &	phirangeTrain,
		std::vector< float > &	invptrangeTrain,
		std::vector< float > &	thetarangeTrain,
		unsigned short	maxHits,
		unsigned long	pattern,
		unsigned long	patternMask,
		unsigned short	tmax,
		const std::string &	etoption
	)

constructor to set all parameters (not weights and relevantID ranges).

Definition at line 32 of file CDCTriggerMLP.cc.

                                                        :
  nNodes(nodes), trained(false), targetVars(targets), outputScale(outputscale),
  phiRangeUse(phirangeUse), invptRangeUse(invptrangeUse), thetaRangeUse(thetarangeUse),
  phiRangeTrain(phirangeTrain), invptRangeTrain(invptrangeTrain), thetaRangeTrain(thetarangeTrain),
  maxHitsPerSL(maxHits), SLpattern(pattern), SLpatternMask(patternMask),
  tMax(tmax),
  relevantID{ -1., 1.,
              -10., 1.,
              -1., 1.,
              -1., 10.,
              -1., 1.,
              -10.5, 1.,
              -1., 1.,
              -1., 11.,
              -1., 1.},
    et_option(etoption)
{
  weights.assign(nWeightsCal(), 0.);
}

◆ ~CDCTriggerMLP()

~CDCTriggerMLP ( )

inline

destructor, empty because we don't allocate memory anywhere.

Definition at line 47 of file CDCTriggerMLP.h.

47{ }

Member Function Documentation

◆ get_et_option()

std::string get_et_option ( ) const

inline

Returns way of obtaining the event time.

Definition at line 82 of file CDCTriggerMLP.h.

82{ return et_option; }

◆ getIDRange()

std::vector< float > getIDRange ( unsigned iSL ) const

inline

get relevant ID range for given super layer

Definition at line 74 of file CDCTriggerMLP.h.

    {
      return {relevantID[2 * iSL], relevantID[2 * iSL + 1]};
    }

◆ getMaxHitsPerSL()

unsigned short getMaxHitsPerSL ( ) const

inline

get maximum hit number for a single super layer

Definition at line 64 of file CDCTriggerMLP.h.

64{ return maxHitsPerSL; }

◆ getRelID()

std::vector< float > getRelID ( ) const

inline

Definition at line 80 of file CDCTriggerMLP.h.

80{return relevantID;}

◆ getSLpattern()

unsigned long getSLpattern ( ) const

inline

get super layer pattern

Definition at line 66 of file CDCTriggerMLP.h.

66{ return SLpattern & SLpatternMask; }

◆ getSLpatternMask()

unsigned long getSLpatternMask ( ) const

inline

get bitmask for super layer pattern

Definition at line 68 of file CDCTriggerMLP.h.

68{ return SLpatternMask; }

◆ getSLpatternUnmasked()

unsigned long getSLpatternUnmasked ( ) const

inline

get raw super layer pattern

Definition at line 70 of file CDCTriggerMLP.h.

70{ return SLpattern; }

◆ getTMax()

unsigned short getTMax ( ) const

inline

get maximal drift time

Definition at line 72 of file CDCTriggerMLP.h.

72{ return tMax; }

◆ getWeights()

std::vector< float > getWeights ( ) const

inline

get weights vector

Definition at line 60 of file CDCTriggerMLP.h.

60{ return weights; }

◆ inInvptRangeTrain()

bool inInvptRangeTrain ( float invpt ) const

check whether given 1/pt value is in training sector

Definition at line 119 of file CDCTriggerMLP.cc.

{
  return (invptRangeTrain[0] <= invpt && invpt <= invptRangeTrain[1]);
}

◆ inInvptRangeUse()

bool inInvptRangeUse ( float invpt ) const

check whether given 1/pt value is in sector

Definition at line 93 of file CDCTriggerMLP.cc.

{
  return (invptRangeUse[0] <= invpt && invpt <= invptRangeUse[1]);
}

◆ inPhiRangeTrain()

bool inPhiRangeTrain ( float phi ) const

check whether given phi value is in training sector

Definition at line 105 of file CDCTriggerMLP.cc.

{
  return ((phiRangeTrain[0] <= (phi - 2. * M_PI) && (phi - 2. * M_PI) <= phiRangeTrain[1]) ||
          (phiRangeTrain[0] <= phi && phi <= phiRangeTrain[1]) ||
          (phiRangeTrain[0] <= (phi + 2. * M_PI) && (phi + 2. * M_PI) <= phiRangeTrain[1]));
}

◆ inPhiRangeUse()

bool inPhiRangeUse ( float phi ) const

check whether given phi value is in sector

Definition at line 79 of file CDCTriggerMLP.cc.

{
  return ((phiRangeUse[0] <= (phi - 2. * M_PI) && (phi - 2. * M_PI) <= phiRangeUse[1]) ||
          (phiRangeUse[0] <= phi && phi <= phiRangeUse[1]) ||
          (phiRangeUse[0] <= (phi + 2. * M_PI) && (phi + 2. * M_PI) <= phiRangeUse[1]));
}

◆ inPtRangeTrain()

bool inPtRangeTrain ( float pt ) const

check whether given pt value is in training sector

Definition at line 113 of file CDCTriggerMLP.cc.

{
  return (invptRangeTrain[0] <= 1. / pt && 1. / pt <= invptRangeTrain[1]);
}

◆ inPtRangeUse()

bool inPtRangeUse ( float pt ) const

check whether given pt value is in sector

Definition at line 87 of file CDCTriggerMLP.cc.

{
  return (invptRangeUse[0] <= 1. / pt && 1. / pt <= invptRangeUse[1]);
}

◆ inThetaRangeTrain()

bool inThetaRangeTrain ( float theta ) const

check whether given theta value is in training sector

Definition at line 125 of file CDCTriggerMLP.cc.

{
  return (thetaRangeTrain[0] <= theta && theta <= thetaRangeTrain[1]);
}

◆ inThetaRangeUse()

bool inThetaRangeUse ( float theta ) const

check whether given theta value is in sector

Definition at line 99 of file CDCTriggerMLP.cc.

{
  return (thetaRangeUse[0] <= theta && theta <= thetaRangeUse[1]);
}

◆ isRelevant()

bool isRelevant	(	float	relId,
		unsigned	iSL
	)		const

check whether given relative TS ID is in relevant range

Definition at line 131 of file CDCTriggerMLP.cc.

{
  return (relevantID[2 * iSL] <= relId && relId <= relevantID[2 * iSL + 1]);
}

◆ isTrained()

bool isTrained ( ) const

inline

check if weights are default values or set by some trainer

Definition at line 50 of file CDCTriggerMLP.h.

50{ return trained; }

◆ nLayers()

unsigned nLayers ( ) const

inline

get number of layers

Definition at line 52 of file CDCTriggerMLP.h.

52{ return nNodes.size(); }

◆ nNodesLayer()

unsigned nNodesLayer ( unsigned iLayer ) const

inline

get number of nodes in a layer

Definition at line 54 of file CDCTriggerMLP.h.

54{ return nNodes[iLayer]; }

◆ nWeights()

unsigned nWeights ( ) const

inline

get number of weights from length of weights vector

Definition at line 56 of file CDCTriggerMLP.h.

56{ return weights.size(); }

◆ nWeightsCal()

unsigned nWeightsCal ( ) const

calculate number of weights from number of nodes

Definition at line 66 of file CDCTriggerMLP.cc.

{
  unsigned nWeights = 0;
  if (nLayers() > 1) {
    nWeights = (nNodes[0] + 1) * nNodes[1];
    for (unsigned il = 1; il < nLayers() - 1; ++il) {
      nWeights += (nNodes[il] + 1) * nNodes[il + 1];
    }
  }
  return nWeights;
}

◆ scaleId()

float scaleId	(	double	relId,
		unsigned	iSL
	)		const

scale relative TS ID from relevant range to approximately [-1, 1] (to facilitate the FPGA implementation, the scale factor is rounded to a power of 2)

Definition at line 137 of file CDCTriggerMLP.cc.

{
  float scale = 2. / (relevantID[2 * iSL + 1] - relevantID[2 * iSL]);
  // round down to nearest power of 2
  scale = pow(2, floor(log2(scale)));
  float offset = (relevantID[2 * iSL] + relevantID[2 * iSL + 1]) / 2.;
  return scale * (relId - offset);
}

◆ scaleTarget()

std::vector< float > scaleTarget ( std::vector< float > target ) const

scale target value from outputScale to [-1, 1]

Definition at line 147 of file CDCTriggerMLP.cc.

{
  std::vector<float> scaled;
  scaled.assign(target.size(), 0.);
  for (unsigned i = 0; i < target.size(); ++i) {
    scaled[i] = 2. * (target[i] - outputScale[2 * i]) / (outputScale[2 * i + 1] - outputScale[2 * i]) - 1.;
  }
  return scaled;
}

◆ setRelID()

void setRelID ( std::vector< float > relid )

inline

set and get total relevant ID range

Definition at line 79 of file CDCTriggerMLP.h.

79{relevantID = relid;}

◆ setWeights()

void setWeights ( std::vector< float > xweights )

inline

set weights vector

Definition at line 62 of file CDCTriggerMLP.h.

62{weights = xweights; }

◆ thetaIndex()

int thetaIndex ( ) const

get target index for theta (-1 if no output is trained for theta)

Definition at line 175 of file CDCTriggerMLP.cc.

{
  return (targetVars & 2) ? (targetVars & 1) : -1;
}

◆ unscaleTarget()

std::vector< float > unscaleTarget ( std::vector< float > target ) const

scale target value from [-1, 1] to outputScale

Definition at line 158 of file CDCTriggerMLP.cc.

{
  std::vector<float> unscaled;
  unscaled.assign(target.size(), 0.);
  for (unsigned i = 0; i < target.size(); ++i) {
    unscaled[i] = (target[i] + 1.) * (outputScale[2 * i + 1] - outputScale[2 * i]) / 2. + outputScale[2 * i];
  }
  return unscaled;
}

◆ zIndex()

int zIndex ( ) const

get target index for z (-1 if no output is trained for z)

Definition at line 169 of file CDCTriggerMLP.cc.

{
  return (targetVars & 1) ? 0 : -1;
}

Friends And Related Function Documentation

◆ CDCTriggerNeuroTrainerModule

friend class CDCTriggerNeuroTrainerModule

friend

Definition at line 23 of file CDCTriggerMLP.h.

◆ NeuroTrigger

friend class NeuroTrigger

friend

Definition at line 24 of file CDCTriggerMLP.h.

Member Data Documentation

◆ et_option

std::string et_option

private

Returns way of obtaining the event time.

The different options are: "etf_only" : only ETF info is used, otherwise an error is thrown. "fastestpriority" : event time is estimated by fastest priority time in selected track segments. if something fails, it is set to 0. "zero" : the event time is set to 0. "etf_or_fastestpriority" : the event time is obtained by the ETF, if not possible, the flag "fastestppriority" is used. "etf_or_zero" : the event time is obtained by the ETF, if

Definition at line 182 of file CDCTriggerMLP.h.

◆ invptRangeTrain

std::vector<float> invptRangeTrain

private

Charge / Pt region in 1/GeV for which this expert is trained.

Taking 1 / Pt instead of Pt means that straight tracks are at 0, i.e. there is a smooth transition from positive to negative charge.

Definition at line 146 of file CDCTriggerMLP.h.

◆ invptRangeUse

std::vector<float> invptRangeUse

private

Charge / Pt region in 1/GeV for which this expert is used.

Taking 1 / Pt instead of Pt means that straight tracks are at 0, i.e. there is a smooth transition from positive to negative charge.

Definition at line 137 of file CDCTriggerMLP.h.

◆ maxHitsPerSL

unsigned short maxHitsPerSL

private

Maximum number of inputs for a single super layer.

Definition at line 151 of file CDCTriggerMLP.h.

◆ nNodes

std::vector<unsigned short> nNodes

private

Number of nodes in each layer, not including bias nodes.

Definition at line 118 of file CDCTriggerMLP.h.

◆ outputScale

std::vector<float> outputScale

private

Output[i] of the MLP is scaled from [-1, 1] to [outputScale[2i], outputScale[2i+1]].

Definition at line 129 of file CDCTriggerMLP.h.

◆ phiRangeTrain

std::vector<float> phiRangeTrain

private

Phi region in radian for which this expert is used.

Valid ranges go from -2pi to 2pi.

Definition at line 142 of file CDCTriggerMLP.h.

◆ phiRangeUse

std::vector<float> phiRangeUse

private

Phi region in radian for which this expert is used.

Valid ranges go from -2pi to 2pi.

Definition at line 133 of file CDCTriggerMLP.h.

◆ relevantID

std::vector<float> relevantID

private

Hits must be within ID region around 2D track to be used as input.

Default for axial layers is +- 1 wire, default for stereo layers is region spanned by stereos +- 1 wire.

Definition at line 166 of file CDCTriggerMLP.h.

◆ SLpattern

unsigned long SLpattern

private

Super layer pattern for which this expert is trained.

Binary pattern of 9 * maxHitsPerSL bits (on/off for each hit). 0 means no restriction rather than no inputs.

Definition at line 155 of file CDCTriggerMLP.h.

◆ SLpatternMask

unsigned long SLpatternMask

private

Bitmask for comparing the super layer pattern.

A track matches a sector, if SLpattern & SLpatternMask == hitPattern(track) & SLpatternMask.

Definition at line 159 of file CDCTriggerMLP.h.

◆ targetVars

unsigned short targetVars

private

output variables: 1: z, 2: theta, 3: (z, theta)

Definition at line 126 of file CDCTriggerMLP.h.

◆ thetaRangeTrain

std::vector<float> thetaRangeTrain

private

Theta region in radian for which this expert is trained.

Definition at line 148 of file CDCTriggerMLP.h.

◆ thetaRangeUse

std::vector<float> thetaRangeUse

private

Theta region in radian for which this expert is trained.

Definition at line 139 of file CDCTriggerMLP.h.

◆ tMax

unsigned short tMax

private

Maximal drift time (for scaling), hits with larger values are ignored.

Definition at line 162 of file CDCTriggerMLP.h.

◆ trained

bool trained

private

Indicator whether the weights are just default values or have been set by some trainer (set to true when setWeights() is first called).

Definition at line 123 of file CDCTriggerMLP.h.

◆ weights

std::vector<float> weights

private

Weights of the network.

Definition at line 120 of file CDCTriggerMLP.h.

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

trg/cdc/dataobjects/include/CDCTriggerMLP.h
trg/cdc/dataobjects/src/CDCTriggerMLP.cc

Public Member Functions

Private Member Functions

Private Attributes

Friends

Detailed Description

Constructor & Destructor Documentation

◆ CDCTriggerMLP() [1/2]

◆ CDCTriggerMLP() [2/2]

◆ ~CDCTriggerMLP()

Member Function Documentation

◆ get_et_option()

◆ getIDRange()

◆ getMaxHitsPerSL()

◆ getRelID()

◆ getSLpattern()

◆ getSLpatternMask()

◆ getSLpatternUnmasked()

◆ getTMax()

◆ getWeights()

◆ inInvptRangeTrain()

◆ inInvptRangeUse()

◆ inPhiRangeTrain()

◆ inPhiRangeUse()

◆ inPtRangeTrain()

◆ inPtRangeUse()

◆ inThetaRangeTrain()

◆ inThetaRangeUse()

◆ isRelevant()

◆ isTrained()

◆ nLayers()

◆ nNodesLayer()

◆ nWeights()

◆ nWeightsCal()

◆ scaleId()

◆ scaleTarget()

◆ setRelID()

◆ setWeights()

◆ thetaIndex()

◆ unscaleTarget()

◆ zIndex()

Friends And Related Function Documentation

◆ CDCTriggerNeuroTrainerModule

◆ NeuroTrigger

Member Data Documentation

◆ et_option

◆ invptRangeTrain

◆ invptRangeUse

◆ maxHitsPerSL

◆ nNodes

◆ outputScale

◆ phiRangeTrain

◆ phiRangeUse

◆ relevantID

◆ SLpattern

◆ SLpatternMask

◆ targetVars

◆ thetaRangeTrain

◆ thetaRangeUse

◆ tMax

◆ trained

◆ weights