The trainer module for the neural networks of the CDC trigger. More...

#include <CDCTriggerNeuroTrainerModule.h>

Inheritance diagram for CDCTriggerNeuroTrainerModule:

Collaboration diagram for CDCTriggerNeuroTrainerModule:

[legend]

Public Types
enum	EModulePropFlags { c_Input = 1 , c_Output = 2 , c_ParallelProcessingCertified = 4 , c_HistogramManager = 8 , c_InternalSerializer = 16 , c_TerminateInAllProcesses = 32 , c_DontCollectStatistics = 64 }
	Each module can be tagged with property flags, which indicate certain features of the module. More...

typedef ModuleCondition::EAfterConditionPath	EAfterConditionPath
	Forward the EAfterConditionPath definition from the ModuleCondition.

Public Member Functions
	CDCTriggerNeuroTrainerModule ()
	Constructor, for setting module description and parameters.

virtual	~CDCTriggerNeuroTrainerModule ()
	Destructor.

virtual void	initialize () override
	Initialize the module. More...

virtual void	event () override
	Called once for each event. More...

virtual void	terminate () override
	Do the training for all sectors.

void	updateRelevantID (unsigned isector)
	calculate and set the relevant id range for given sector based on hit counters of the track segments.

void	train (unsigned isector)
	Train a single MLP.

void	saveTraindata (const std::string &filename, const std::string &arrayname="trainSets")
	Save all training samples. More...

bool	loadTraindata (const std::string &filename, const std::string &arrayname="trainSets")
	Load saved training samples. More...

virtual std::vector< std::string >	getFileNames (bool outputFiles)
	Return a list of output filenames for this modules. More...

virtual void	beginRun ()
	Called when entering a new run. More...

virtual void	endRun ()
	This method is called if the current run ends. More...

const std::string &	getName () const
	Returns the name of the module. More...

const std::string &	getType () const
	Returns the type of the module (i.e. More...

const std::string &	getPackage () const
	Returns the package this module is in.

const std::string &	getDescription () const
	Returns the description of the module.

void	setName (const std::string &name)
	Set the name of the module. More...

void	setPropertyFlags (unsigned int propertyFlags)
	Sets the flags for the module properties. More...

LogConfig &	getLogConfig ()
	Returns the log system configuration.

void	setLogConfig (const LogConfig &logConfig)
	Set the log system configuration.

void	setLogLevel (int logLevel)
	Configure the log level.

void	setDebugLevel (int debugLevel)
	Configure the debug messaging level.

void	setAbortLevel (int abortLevel)
	Configure the abort log level.

void	setLogInfo (int logLevel, unsigned int logInfo)
	Configure the printed log information for the given level. More...

void	if_value (const std::string &expression, const std::shared_ptr< Path > &path, EAfterConditionPath afterConditionPath=EAfterConditionPath::c_End)
	Add a condition to the module. More...

void	if_false (const std::shared_ptr< Path > &path, EAfterConditionPath afterConditionPath=EAfterConditionPath::c_End)
	A simplified version to add a condition to the module. More...

void	if_true (const std::shared_ptr< Path > &path, EAfterConditionPath afterConditionPath=EAfterConditionPath::c_End)
	A simplified version to set the condition of the module. More...

bool	hasCondition () const
	Returns true if at least one condition was set for the module.

const ModuleCondition *	getCondition () const
	Return a pointer to the first condition (or nullptr, if none was set)

const std::vector< ModuleCondition > &	getAllConditions () const
	Return all set conditions for this module.

bool	evalCondition () const
	If at least one condition was set, it is evaluated and true returned if at least one condition returns true. More...

std::shared_ptr< Path >	getConditionPath () const
	Returns the path of the last true condition (if there is at least one, else reaturn a null pointer). More...

Module::EAfterConditionPath	getAfterConditionPath () const
	What to do after the conditional path is finished. More...

std::vector< std::shared_ptr< Path > >	getAllConditionPaths () const
	Return all condition paths currently set (no matter if the condition is true or not).

bool	hasProperties (unsigned int propertyFlags) const
	Returns true if all specified property flags are available in this module. More...

bool	hasUnsetForcedParams () const
	Returns true and prints error message if the module has unset parameters which the user has to set in the steering file.

const ModuleParamList &	getParamList () const
	Return module param list.

template<typename T >
ModuleParam< T > &	getParam (const std::string &name) const
	Returns a reference to a parameter. More...

bool	hasReturnValue () const
	Return true if this module has a valid return value set.

int	getReturnValue () const
	Return the return value set by this module. More...

std::shared_ptr< PathElement >	clone () const override
	Create an independent copy of this module. More...

std::shared_ptr< boost::python::list >	getParamInfoListPython () const
	Returns a python list of all parameters. More...

Static Public Member Functions
static void	exposePythonAPI ()
	Exposes methods of the Module class to Python.

Protected Member Functions
virtual void	def_initialize ()
	Wrappers to make the methods without "def_" prefix callable from Python. More...

virtual void	def_beginRun ()
	Wrapper method for the virtual function beginRun() that has the implementation to be used in a call from Python.

virtual void	def_event ()
	Wrapper method for the virtual function event() that has the implementation to be used in a call from Python.

virtual void	def_endRun ()
	This method can receive that the current run ends as a call from the Python side. More...

virtual void	def_terminate ()
	Wrapper method for the virtual function terminate() that has the implementation to be used in a call from Python.

void	setDescription (const std::string &description)
	Sets the description of the module. More...

void	setType (const std::string &type)
	Set the module type. More...

template<typename T >
void	addParam (const std::string &name, T &paramVariable, const std::string &description, const T &defaultValue)
	Adds a new parameter to the module. More...

template<typename T >
void	addParam (const std::string &name, T &paramVariable, const std::string &description)
	Adds a new enforced parameter to the module. More...

void	setReturnValue (int value)
	Sets the return value for this module as integer. More...

void	setReturnValue (bool value)
	Sets the return value for this module as bool. More...

void	setParamList (const ModuleParamList &params)
	Replace existing parameter list.

Protected Attributes
std::string	m_hitCollectionName
	Name of the StoreArray containing the input track segment hits.

std::string	m_EventTimeName
	name of the event time StoreObjPtr

std::string	m_inputCollectionName
	Name of the StoreArray containing the input 2D tracks.

bool	m_trainOnRecoTracks
	Switch between MCParticles or RecoTracks as targets.

std::string	m_targetCollectionName
	Name of the MCParticles/RecoTracks collection used as target values.

std::string	m_filename
	Name of file where network weights etc. More...

std::string	m_trainFilename
	Name of file where training samples are stored.

std::string	m_logFilename
	Name of file where training log is stored.

std::string	m_arrayname
	Name of the TObjArray holding the networks.

std::string	m_trainArrayname
	Name of the TObjArray holding the training samples.

bool	m_saveDebug
	If true, save training curve and parameter distribution of training data.

bool	m_load
	Switch to load saved parameters from a previous run.

NeuroTrigger::Parameters	m_parameters
	Parameters for the NeuroTrigger.

bool	m_selectSectorByMC
	Switch for sector selection during training.

int	m_nTrainPrepare
	Number of samples to prepare input ranges.

std::vector< std::vector< float > >	m_IDranges
	Input ranges given over the module parameters.

double	m_relevantCut
	Cut on the hit counters to get relevant ID ranges.

bool	m_cutSum
	Switch to apply cut to single hit counter or to sum over counters.

double	m_nTrainMin
	Minimal number of training samples.

double	m_nTrainMax
	Maximal number of training samples.

bool	m_multiplyNTrain
	Switch to multiply number of samples with number of weights.

int	m_nValid
	Number of validation samples.

int	m_nTest
	Number of test samples.

bool	m_stopLoop
	Switch for eary stopping.

bool	m_rescaleTarget
	Switch to rescale out of range target values or ignore them.

double	m_wMax
	Limit for weights.

int	m_nThreads
	Number of threads for training.

int	m_checkInterval
	Training is stopped if validation error is higher than checkInterval epochs ago, i.e. More...

int	m_maxEpochs
	Maximal number of training epochs.

int	m_repeatTrain
	Number of training runs with different random start weights.

bool	m_neuroTrackInputMode
	Use neurotracks instead of 2dtracks as input for the neurotrigger.

NeuroTrigger	m_NeuroTrigger
	Instance of the NeuroTrigger.

std::vector< CDCTriggerMLPData >	m_trainSets
	Sets of training data for all sectors.

StoreArray< CDCTriggerTrack >	m_tracks
	List of input tracks.

std::vector< TH1D * >	phiHistsMC
	phi of MCParticles

std::vector< TH1D * >	ptHistsMC
	pt of MCParticles

std::vector< TH1D * >	thetaHistsMC
	theta of MCParticles

std::vector< TH1D * >	zHistsMC
	z of MCParticles

std::vector< TH1D * >	phiHists2D
	phi of 2D tracks

std::vector< TH1D * >	ptHists2D
	pt of 2D tracks

Private Member Functions
std::list< ModulePtr >	getModules () const override
	no submodules, return empty list

std::string	getPathString () const override
	return the module name.

void	setParamPython (const std::string &name, const boost::python::object &pyObj)
	Implements a method for setting boost::python objects. More...

void	setParamPythonDict (const boost::python::dict &dictionary)
	Implements a method for reading the parameter values from a boost::python dictionary. More...

Private Attributes
std::string	m_name
	The name of the module, saved as a string (user-modifiable)

std::string	m_type
	The type of the module, saved as a string.

std::string	m_package
	Package this module is found in (may be empty).

std::string	m_description
	The description of the module.

unsigned int	m_propertyFlags
	The properties of the module as bitwise or (with \|) of EModulePropFlags.

LogConfig	m_logConfig
	The log system configuration of the module.

ModuleParamList	m_moduleParamList
	List storing and managing all parameter of the module.

bool	m_hasReturnValue
	True, if the return value is set.

int	m_returnValue
	The return value.

std::vector< ModuleCondition >	m_conditions
	Module condition, only non-null if set.

Detailed Description

The trainer module for the neural networks of the CDC trigger.

Prepare training data for several neural networks and train them using the Fast Artificial Neural Network library (FANN). For documentation of FANN see http://leenissen.dk/fann/wp/

Definition at line 31 of file CDCTriggerNeuroTrainerModule.h.

Member Enumeration Documentation

◆ EModulePropFlags

enum EModulePropFlags

inherited

Each module can be tagged with property flags, which indicate certain features of the module.

Enumerator
c_Input	This module is an input module (reads data).
c_Output	This module is an output module (writes data).
c_ParallelProcessingCertified	This module can be run in parallel processing mode safely (All I/O must be done through the data store, in particular, the module must not write any files.)
c_HistogramManager	This module is used to manage histograms accumulated by other modules.
c_InternalSerializer	This module is an internal serializer/deserializer for parallel processing.
c_TerminateInAllProcesses	When using parallel processing, call this module's terminate() function in all processes(). This will also ensure that there is exactly one process (single-core if no parallel modules found) or at least one input, one main and one output process.
c_DontCollectStatistics	No statistics is collected for this module.

Definition at line 77 of file Module.h.

Member Function Documentation

◆ beginRun()

virtual void beginRun ( void )

inlinevirtualinherited

Called when entering a new run.

Called at the beginning of each run, the method gives you the chance to change run dependent constants like alignment parameters, etc.

This method can be implemented by subclasses.

Definition at line 147 of file Module.h.

◆ clone()

std::shared_ptr< PathElement > clone ( ) const

overridevirtualinherited

Create an independent copy of this module.

Note that parameters are shared, so changing them on a cloned module will also affect the original module.

Implements PathElement.

Definition at line 179 of file Module.cc.

◆ def_endRun()

virtual void def_endRun ( )

inlineprotectedvirtualinherited

This method can receive that the current run ends as a call from the Python side.

For regular C++-Modules that forwards the call to the regular endRun() method.

Reimplemented in PyModule.

Definition at line 439 of file Module.h.

◆ def_initialize()

virtual void def_initialize ( )

inlineprotectedvirtualinherited

Wrappers to make the methods without "def_" prefix callable from Python.

Overridden in PyModule. Wrapper method for the virtual function initialize() that has the implementation to be used in a call from Python.

Reimplemented in PyModule.

Definition at line 420 of file Module.h.

◆ endRun()

virtual void endRun ( void )

inlinevirtualinherited

This method is called if the current run ends.

Use this method to store information, which should be aggregated over one run.

This method can be implemented by subclasses.

Definition at line 166 of file Module.h.

◆ evalCondition()

bool evalCondition ( ) const

inherited

If at least one condition was set, it is evaluated and true returned if at least one condition returns true.

If no condition or result value was defined, the method returns false. Otherwise, the condition is evaluated and true returned, if at least one condition returns true. To speed up the evaluation, the condition strings were already parsed in the method if_value().

Returns: True if at least one condition and return value exists and at least one condition expression was evaluated to true.

Definition at line 96 of file Module.cc.

◆ event()

void event ( void )

overridevirtual

Called once for each event.

Prepare input and target for each track and store it.

Reimplemented from Module.

Definition at line 304 of file CDCTriggerNeuroTrainerModule.cc.

 {
   for (int itrack = 0; itrack < m_tracks.getEntries(); ++itrack) {
     // get related MCParticle/RecoTrack for target
     // and retrieve track parameters
     float phi0Target = 0;
     float invptTarget = 0;
     float thetaTarget = 0;
     float zTarget = 0;
     if (m_trainOnRecoTracks) {
       RecoTrack* recoTrack =
         m_tracks[itrack]->getRelatedTo<RecoTrack>(m_targetCollectionName);
       if (!recoTrack) {
         B2DEBUG(150, "Skipping CDCTriggerTrack without relation to RecoTrack.");
         continue;
       }
       // a RecoTrack has multiple representations for different particle hypothesis
       // -> just take the first one that does not give errors.
       const vector<genfit::AbsTrackRep*>& reps = recoTrack->getRepresentations();
       bool foundValidRep = false;
       for (unsigned irep = 0; irep < reps.size() && !foundValidRep; ++irep) {
         if (!recoTrack->wasFitSuccessful(reps[irep]))
           continue;
         // get state (position, momentum etc.) from hit closest to IP and
         // extrapolate to z-axis (may throw an exception -> continue to next representation)
         try {
           genfit::MeasuredStateOnPlane state =
             recoTrack->getMeasuredStateOnPlaneClosestTo(TVector3(0, 0, 0), reps[irep]);
           reps[irep]->extrapolateToLine(state, TVector3(0, 0, -1000), TVector3(0, 0, 2000));
           // flip tracks if necessary, such that trigger tracks and reco tracks
           // point in the same direction
           if (state.getMom().Dot(m_tracks[itrack]->getDirection()) < 0) {
             state.setPosMom(state.getPos(), -state.getMom());
             state.setChargeSign(-state.getCharge());
           }
           // get track parameters
           phi0Target = state.getMom().Phi();
           invptTarget = state.getCharge() / state.getMom().Pt();
           thetaTarget = state.getMom().Theta();
           zTarget = state.getPos().Z();
         } catch (...) {
           continue;
         }
         // break loop
         foundValidRep = true;
       }
       if (!foundValidRep) {
         B2DEBUG(150, "No valid representation found for RecoTrack, skipping.");
         continue;
       }
     } else {
       MCParticle* mcTrack =
         m_tracks[itrack]->getRelatedTo<MCParticle>(m_targetCollectionName);
       if (!mcTrack) {
         B2DEBUG(150, "Skipping CDCTriggerTrack without relation to MCParticle.");
         continue;
       }
       phi0Target = mcTrack->getMomentum().Phi();
       invptTarget = mcTrack->getCharge() / mcTrack->getMomentum().Pt();
       thetaTarget = mcTrack->getMomentum().Theta();
       zTarget = mcTrack->getProductionVertex().Z();
     }
  
     // update 2D track variables
     m_NeuroTrigger.updateTrack(*m_tracks[itrack]);
  
     // find all matching sectors
     float phi0 = m_tracks[itrack]->getPhi0();
     float invpt = m_tracks[itrack]->getKappa(1.5);
     float theta = atan2(1., m_tracks[itrack]->getCotTheta());
     if (m_selectSectorByMC) {
       phi0 = phi0Target;
       invpt = invptTarget;
       theta = thetaTarget;
     }
     vector<int> sectors = m_NeuroTrigger.selectMLPs(phi0, invpt, theta);
     if (sectors.size() == 0) continue;
     // get target values
     vector<float> targetRaw = {};
     if (m_parameters.targetZ)
       targetRaw.push_back(zTarget);
     if (m_parameters.targetTheta)
       targetRaw.push_back(thetaTarget);
     for (unsigned i = 0; i < sectors.size(); ++i) {
       int isector = sectors[i];
       vector<float> target = m_NeuroTrigger[isector].scaleTarget(targetRaw);
       // skip out of range targets or rescale them
       bool outOfRange = false;
       for (unsigned itarget = 0; itarget < target.size(); ++itarget) {
         if (fabs(target[itarget]) > 1.) {
           outOfRange = true;
           target[itarget] /= fabs(target[itarget]);
         }
       }
       if (!m_rescaleTarget && outOfRange) continue;
  
       if (m_nTrainPrepare > 0 &&
           m_trainSets[isector].getTrackCounter() < m_nTrainPrepare) {
         // get relative ids for all hits related to the MCParticle / RecoTrack
         // and count them to find relevant id range
         // using only related hits suppresses background EXCEPT for curling tracks
         if (m_trainOnRecoTracks) {
           RecoTrack* recoTrack =
             m_tracks[itrack]->getRelatedTo<RecoTrack>(m_targetCollectionName);
           for (const CDCTriggerSegmentHit& hit :
                recoTrack->getRelationsTo<CDCTriggerSegmentHit>(m_hitCollectionName)) {
             // get relative id
             double relId = m_NeuroTrigger.getRelId(hit);
             m_trainSets[isector].addHit(hit.getISuperLayer(), round(relId));
           }
         } else {
           MCParticle* mcTrack =
             m_tracks[itrack]->getRelatedTo<MCParticle>(m_targetCollectionName);
           for (const CDCTriggerSegmentHit& hit :
                mcTrack->getRelationsTo<CDCTriggerSegmentHit>(m_hitCollectionName)) {
             // get relative id
             double relId = m_NeuroTrigger.getRelId(hit);
             m_trainSets[isector].addHit(hit.getISuperLayer(), round(relId));
           }
         }
         m_trainSets[isector].countTrack();
         // if required hit number is reached, get relevant ids
         if (m_trainSets[isector].getTrackCounter() >= m_nTrainPrepare) {
           updateRelevantID(isector);
         }
       } else {
         // check whether we already have enough samples
         float nTrainMax = m_multiplyNTrain ? m_nTrainMax * m_NeuroTrigger[isector].nWeights() : m_nTrainMax;
         if (m_trainSets[isector].nSamples() > (nTrainMax + m_nValid + m_nTest)) {
           continue;
         }
         // read out or determine event time
         m_NeuroTrigger.getEventTime(isector, *m_tracks[itrack], m_parameters.et_option, m_neuroTrackInputMode);
         // check hit pattern
         unsigned long hitPattern = m_NeuroTrigger.getInputPattern(isector, *m_tracks[itrack], m_neuroTrackInputMode);
         unsigned long sectorPattern = m_NeuroTrigger[isector].getSLpattern();
         B2DEBUG(250, "hitPattern " << hitPattern << " sectorPattern " << sectorPattern);
         if (sectorPattern > 0 && (sectorPattern & hitPattern) != sectorPattern) {
           B2DEBUG(250, "hitPattern not matching " << (sectorPattern & hitPattern));
           continue;
         }
         // get training data
         vector<unsigned> hitIds;
         if (m_neuroTrackInputMode) {
           hitIds = m_NeuroTrigger.selectHitsHWSim(isector, *m_tracks[itrack]);
         } else {
           hitIds = m_NeuroTrigger.selectHits(isector, *m_tracks[itrack]);
         }
         m_trainSets[isector].addSample(m_NeuroTrigger.getInputVector(isector, hitIds), target);
         if (m_saveDebug) {
           phiHistsMC[isector]->Fill(phi0Target);
           ptHistsMC[isector]->Fill(invptTarget);
           thetaHistsMC[isector]->Fill(thetaTarget);
           zHistsMC[isector]->Fill(zTarget);
           phiHists2D[isector]->Fill(m_tracks[itrack]->getPhi0());
           ptHists2D[isector]->Fill(m_tracks[itrack]->getKappa(1.5));
         }
         if (m_trainSets[isector].nSamples() % 1000 == 0) {
           B2DEBUG(50, m_trainSets[isector].nSamples() << " samples collected for sector " << isector);
         }
       }
     }
   }
   // check number of samples for all sectors
   if (m_stopLoop) {
     bool stop = true;
     for (unsigned isector = 0; isector < m_trainSets.size(); ++isector) {
       float nTrainMax = m_multiplyNTrain ? m_nTrainMax * m_NeuroTrigger[isector].nWeights() : m_nTrainMax;
       if (m_trainSets[isector].nSamples() < (nTrainMax + m_nValid + m_nTest)) {
         stop = false;
         break;
       }
     }
     if (stop) {
       B2INFO("Training sample preparation for NeuroTrigger finished, stopping event loop.");
       StoreObjPtr<EventMetaData> eventMetaData;
       eventMetaData->setEndOfData();
     }
   }
 }

◆ getAfterConditionPath()

Module::EAfterConditionPath getAfterConditionPath ( ) const

inherited

What to do after the conditional path is finished.

(defaults to c_End if no condition is set)

Definition at line 133 of file Module.cc.

◆ getConditionPath()

std::shared_ptr< Path > getConditionPath ( ) const

inherited

Returns the path of the last true condition (if there is at least one, else reaturn a null pointer).

Definition at line 113 of file Module.cc.

◆ getFileNames()

virtual std::vector<std::string> getFileNames ( bool outputFiles )

inlinevirtualinherited

Return a list of output filenames for this modules.

This will be called when basf2 is run with "--dry-run" if the module has set either the c_Input or c_Output properties.

If the parameter outputFiles is false (for modules with c_Input) the list of input filenames should be returned (if any). If outputFiles is true (for modules with c_Output) the list of output files should be returned (if any).

If a module has sat both properties this member is called twice, once for each property.

The module should return the actual list of requested input or produced output filenames (including handling of input/output overrides) so that the grid system can handle input/output files correctly.

This function should return the same value when called multiple times. This is especially important when taking the input/output overrides from Environment as they get consumed when obtained so the finalized list of output files should be stored for subsequent calls.

Reimplemented in RootOutputModule, and RootInputModule.

Definition at line 134 of file Module.h.

◆ getName()

const std::string& getName ( ) const

inlineinherited

Returns the name of the module.

This can be changed via e.g. set_name() in the steering file to give more useful names if there is more than one module of the same type.

For identifying the type of a module, using getType() (or type() in Python) is recommended.

Definition at line 187 of file Module.h.

◆ getParamInfoListPython()

std::shared_ptr< boost::python::list > getParamInfoListPython ( ) const

inherited

Returns a python list of all parameters.

Each item in the list consists of the name of the parameter, a string describing its type, a python list of all default values and the description of the parameter.

Returns: A python list containing the parameters of this parameter list.

Definition at line 279 of file Module.cc.

◆ getReturnValue()

int getReturnValue ( ) const

inlineinherited

Return the return value set by this module.

This value is only meaningful if hasReturnValue() is true

Definition at line 381 of file Module.h.

◆ getType()

const std::string & getType ( ) const

inherited

Returns the type of the module (i.e.

class name minus 'Module')

Definition at line 41 of file Module.cc.

◆ hasProperties()

bool hasProperties ( unsigned int propertyFlags ) const

inherited

Returns true if all specified property flags are available in this module.

Parameters

propertyFlags Ored EModulePropFlags which should be compared with the module flags.

Definition at line 160 of file Module.cc.

◆ if_false()

void if_false	(	const std::shared_ptr< Path > &	path,
		EAfterConditionPath	afterConditionPath = `EAfterConditionPath::c_End`
	)

inherited

A simplified version to add a condition to the module.

Please note that successive calls of this function will add more than one condition to the module. If more than one condition results in true, only the last of them will be used.

Please be careful: Avoid creating cyclic paths, e.g. by linking a condition to a path which is processed before the path where this module is located in.

It is equivalent to the if_value() method, using the expression "<1". This method is meant to be used together with the setReturnValue(bool value) method.

Parameters

path	Shared pointer to the Path which will be executed if the return value is false.
afterConditionPath	What to do after executing 'path'.

Definition at line 85 of file Module.cc.

◆ if_true()

void if_true	(	const std::shared_ptr< Path > &	path,
		EAfterConditionPath	afterConditionPath = `EAfterConditionPath::c_End`
	)

inherited

A simplified version to set the condition of the module.

Please note that successive calls of this function will add more than one condition to the module. If more than one condition results in true, only the last of them will be used.

Please be careful: Avoid creating cyclic paths, e.g. by linking a condition to a path which is processed before the path where this module is located in.

It is equivalent to the if_value() method, using the expression ">=1". This method is meant to be used together with the setReturnValue(bool value) method.

Parameters

path	Shared pointer to the Path which will be executed if the return value is true.
afterConditionPath	What to do after executing 'path'.

Definition at line 90 of file Module.cc.

◆ if_value()

void if_value	(	const std::string &	expression,
		const std::shared_ptr< Path > &	path,
		EAfterConditionPath	afterConditionPath = `EAfterConditionPath::c_End`
	)

inherited

Add a condition to the module.

Please note that successive calls of this function will add more than one condition to the module. If more than one condition results in true, only the last of them will be used.

See https://confluence.desy.de/display/BI/Software+ModCondTut or ModuleCondition for a description of the syntax.

Please be careful: Avoid creating cyclic paths, e.g. by linking a condition to a path which is processed before the path where this module is located in.

Parameters

expression	The expression of the condition.
path	Shared pointer to the Path which will be executed if the condition is evaluated to true.
afterConditionPath	What to do after executing 'path'.

Definition at line 79 of file Module.cc.

◆ initialize()

void initialize ( void )

overridevirtual

Initialize the module.

Initialize the networks and register datastore objects.

Reimplemented from Module.

Definition at line 216 of file CDCTriggerNeuroTrainerModule.cc.

◆ loadTraindata()

bool loadTraindata	(	const std::string &	filename,
		const std::string &	arrayname = `"trainSets"`
	)

Load saved training samples.

Parameters

filename	name of the TFile to read from
arrayname	name of the TObjArray holding the training samples in the file

Returns: true if the training sets were loaded correctly

Definition at line 804 of file CDCTriggerNeuroTrainerModule.cc.

◆ saveTraindata()

void saveTraindata	(	const std::string &	filename,
		const std::string &	arrayname = `"trainSets"`
	)

Save all training samples.

Parameters

filename	name of the TFile to write to
arrayname	name of the TObjArray holding the training samples in the file

Definition at line 767 of file CDCTriggerNeuroTrainerModule.cc.

◆ setDescription()

void setDescription ( const std::string & description )

protectedinherited

Sets the description of the module.

Parameters

description A description of the module.

Definition at line 214 of file Module.cc.

◆ setLogInfo()

void setLogInfo	(	int	logLevel,
		unsigned int	logInfo
	)

inherited

Configure the printed log information for the given level.

Parameters

logLevel	The log level (one of LogConfig::ELogLevel)
logInfo	What kind of info should be printed? ORed combination of LogConfig::ELogInfo flags.

Definition at line 73 of file Module.cc.

◆ setName()

void setName ( const std::string & name )

inlineinherited

Set the name of the module.

Note: The module name is set when using the REG_MODULE macro, but the module can be renamed before calling process() using the set_name() function in your steering file.

Parameters

name	The name of the module

Definition at line 214 of file Module.h.

◆ setParamPython()

void setParamPython	(	const std::string &	name,
		const boost::python::object &	pyObj
	)

privateinherited

Implements a method for setting boost::python objects.

The method supports the following types: list, dict, int, double, string, bool The conversion of the python object to the C++ type and the final storage of the parameter value is done in the ModuleParam class.

Parameters

name	The unique name of the parameter.
pyObj	The object which should be converted and stored as the parameter value.

Definition at line 234 of file Module.cc.

◆ setParamPythonDict()

void setParamPythonDict ( const boost::python::dict & dictionary )

privateinherited

Implements a method for reading the parameter values from a boost::python dictionary.

The key of the dictionary has to be the name of the parameter and the value has to be of one of the supported parameter types.

Parameters

dictionary The python dictionary from which the parameter values are read.

Definition at line 249 of file Module.cc.

◆ setPropertyFlags()

void setPropertyFlags ( unsigned int propertyFlags )

inherited

Sets the flags for the module properties.

Parameters

propertyFlags bitwise OR of EModulePropFlags

Definition at line 208 of file Module.cc.

◆ setReturnValue() [1/2]

void setReturnValue ( bool value )

protectedinherited

Sets the return value for this module as bool.

The bool value is saved as an integer with the convention 1 meaning true and 0 meaning false. The value can be used in the steering file to divide the analysis chain into several paths.

Parameters

value The value of the return value.

Definition at line 227 of file Module.cc.

◆ setReturnValue() [2/2]

void setReturnValue ( int value )

protectedinherited

Sets the return value for this module as integer.

The value can be used in the steering file to divide the analysis chain into several paths.

Parameters

value The value of the return value.

Definition at line 220 of file Module.cc.

◆ setType()

void setType ( const std::string & type )

protectedinherited

Set the module type.

Only for use by internal modules (which don't use the normal REG_MODULE mechanism).

Definition at line 48 of file Module.cc.

Member Data Documentation

◆ m_checkInterval

int m_checkInterval

protected

Training is stopped if validation error is higher than checkInterval epochs ago, i.e.

either the validation error is increasing or the gain is less than the fluctuations.

Definition at line 131 of file CDCTriggerNeuroTrainerModule.h.

◆ m_filename

std::string m_filename

protected

Name of file where network weights etc.

are stored after training.

Definition at line 85 of file CDCTriggerNeuroTrainerModule.h.

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

trg/cdc/modules/neurotrigger/include/CDCTriggerNeuroTrainerModule.h
trg/cdc/modules/neurotrigger/src/CDCTriggerNeuroTrainerModule.cc

Public Types

Public Member Functions

Static Public Member Functions

Protected Member Functions

Protected Attributes

Private Member Functions

Private Attributes

Detailed Description

Member Enumeration Documentation

◆ EModulePropFlags

Member Function Documentation

◆ beginRun()

◆ clone()

◆ def_endRun()

◆ def_initialize()

◆ endRun()

◆ evalCondition()

◆ event()

◆ getAfterConditionPath()

◆ getConditionPath()

◆ getFileNames()

◆ getName()

◆ getParamInfoListPython()

◆ getReturnValue()

◆ getType()

◆ hasProperties()

◆ if_false()

◆ if_true()

◆ if_value()

◆ initialize()

◆ loadTraindata()

◆ saveTraindata()

◆ setDescription()

◆ setLogInfo()

◆ setName()

◆ setParamPython()

◆ setParamPythonDict()

◆ setPropertyFlags()

◆ setReturnValue() [1/2]

◆ setReturnValue() [2/2]

◆ setType()

Member Data Documentation

◆ m_checkInterval

◆ m_filename