Difference between revisions of "Analysis FAQ"

From GlueXWiki
Jump to: navigation, search
(Track / Detector Matching)
(What do I need to know about multithreading with JANA?)
Line 65: Line 65:
 
</syntaxhighlight>
 
</syntaxhighlight>
  
* Except within the processor init() method, any time a ROOT histogram/file/tree/directory is created/modified/filled, it should occur within a ROOT lock:
+
* Except within the processor init() method, any time a ROOT histogram/file/tree/directory is created/modified/filled, it should occur within a lock.
 
+
<syntaxhighlight>
+
japp->RootWriteLock(); //ACQUIRE ROOT LOCK!!
+
{
+
  dMyHist->Fill(locValue);
+
}
+
japp->RootUnLock(); //RELEASE ROOT LOCK!!
+
</syntaxhighlight>
+
  
 
== Neutrals ==
 
== Neutrals ==

Revision as of 07:49, 3 November 2017

JANA

How do I skip events?

  • To skip the first 'X' events, add the following to the command line when running hd_root:
-PEVENTS_TO_SKIP=X

How do I limit the number of events analyzed?

  • To analyze only 'X' events, add the following to the command line when running hd_root:
-PEVENTS_TO_KEEP=X

How do I find out which factories are being called, and how much time they take?

  • Run with the 'janadot' plugin. For example:
hd_root dana_rest.hddm -PPLUGINS=b1pi_hists,monitoring_hists,janadot
  • When that finishes, use the dot program to convert jana.dot to postscript, and then convert to pdf:
dot -Tps2 jana.dot -o jana.ps
ps2pdf jana.ps

What do I need to know about multithreading with JANA?

  • To run JANA with 'X' threads, add the following to the command line when running hd_root:
-PNTHREADS=X
  • Event Processors (They drive JANA, and are in your plugins):
    • All threads share/use the same instance of an event processor. Thus, you must be careful when setting/modifying processor member variables.
    • The init() method is guaranteed to be called by only one thread. Any code in here will be thread safe.
    • The brun() method will be called by each thread. Be VERY careful when modifying processor member variables here, as one thread may still be analyzing an event with the previous run number.
    • The evnt() method will be called by each thread. See the histogramming section about how to safely fill histograms.
  • Factories (They create most JANA objects):
    • Each thread will have it's own instance of each factory. Thus, factory member variables can be safely modified without locks. You still need to be careful about ROOT objects (below).
  • ROOT:
    • Only one instance of a given ROOT file, directory, histogram, or tree should be created.
      • In processors, they should be created in the init() method, so that only one thread will attempt to create them.
      • In factories and analysis actions, each thread should acquire a ROOT lock (see histogramming section), and first check to see if the histogram/etc. has already been created. If so, it should grab a pointer to the existing histogram/etc. instead of creating a duplicate. Example:
japp->RootWriteLock(); //ACQUIRE ROOT LOCK!!
{
   if(gDirectory->Get("MyHistName") != NULL) //already created by another thread
      dMyHist = static_cast<TH1D*>(gDirectory->Get("MyHistName"); //grab the pointer to the pre-existing object
   else
      dMyHist = new TH1D("MyHistName", "MyHistTitle", 100, 0.0, 1.0); //create the histogram
}
japp->RootUnLock(); //RELEASE ROOT LOCK!!
  • Except within the processor init() method, any time a ROOT histogram/file/tree/directory is created/modified/filled, it should occur within a lock.

Neutrals

Given a DNeutralParticleHypothesis or DNeutralShower object, how do I get the FCAL/BCAL shower objects?

  • Answer: By calling the inherited JObject::GetSingle() function:
//given a hypothesis, get the neutral shower:
const DNeutralShower* locNeutralShower = NULL;
locNeutralParticleHypothesis->GetSingle(locNeutralShower);
 
//given a neutral shower, get the FCAL shower:
const DFCALShower* locFCALShower = NULL;
if(locNeutralShower->dDetectorSystem == SYS_FCAL)
   locNeutralShower->GetSingle(locFCALShower);
 
//given a neutral shower, get the BCAL shower:
const DBCALShower* locBCALShower = NULL;
if(locNeutralShower->dDetectorSystem == SYS_BCAL)
   locNeutralShower->GetSingle(locBCALShower);

Histogramming

How do I manually create and fill a histogram in my plugin's processor in a way that is thread-safe?

//If not already present, include the appropriate ROOT header file in your processor header file:
#include "TH1D.h" //1 dimensional ("1") histogram containing doubles ("D")
 
//Define your histogram in your processor class definition:
TH1D* dMyHist;
 
//Create your histogram in your processor init() method:
   //Note that you do not need a ROOT lock here, since the init() method of event processors is only called by one of the threads. 
   //If you were doing this in a factory or an analysis action, you would need a ROOT lock here. 
dMyHist = new TH1D("HistName", "Hist Title;X-axis Title;Y-axis Title", 100, 0.0, 1.0); //100 bins from 0 to 1
 
//Fill the histogram in your processor evnt() method, but within a ROOT lock:
double locValue = 1.0;
japp->RootWriteLock(); //ACQUIRE ROOT LOCK!!
{
   dMyHist->Fill(locValue);
}
japp->RootUnLock(); //RELEASE ROOT LOCK!!
 
//The histogram is automatically saved to hd_root.root, or the file you specified with "-o" on the command line.

Analysis Action Cuts

How do I place a cut prior to kinematic fitting?

  • Answer: Add an analysis action to the DReaction that performs the cut. It should be added prior to any action that uses the kinematic fit results. If no action uses the fit results, then it can be added anywhere.
   //E.g.: Place cut on pi0 invariant mass (assuming a Pi0 decay step is defined in your DReaction): 
      //the false means "used measured data" as opposed to kinfit data
      //0.04 and 0.24 specify the minimum and maximum invariant mass
   locReaction->Add_AnalysisAction(new DCutAction_InvariantMass(locReaction, Pi0, false, 0.04, 0.24)); //does not require kinfit results
 
   //Histogram kinematic fit results
   locReaction->Add_AnalysisAction(new DHistogramAction_KinFitResults(locReaction, 0.05)); //requires kinfit results
 
   //Thus, the kinematic fit is performed between execution of these actions. If a combo fails a cut prior to the kinfit, then the kinfit is not performed for that combo.

How do I use/histogram particle combos that failed a cut?

  • To get the particle combos that failed a cut, in your plugin's event processor evnt() method:
//Get Analysis Results
vector<const DAnalysisResults*> locAnalysisResults;
locEventLoop->Get(locAnalysisResults);
 
//Get Failed Combos
map<const DParticleCombo*, size_t>& locFailedParticleComboMap; //size_t is the action index where the combo failed the cut
locAnalysisResults->Get_FailedParticleComboMap(locFailedParticleComboMap);
 
//Loop over the map
map<const DParticleCombo*, size_t>::iterator locIterator = locFailedParticleComboMap.begin();
for(; locIterator != locFailedParticleComboMap.end(); ++locIterator)
{
   const DParticleCombo* locParticleCombo = locIterator->first;
   if(locParticleCombo->Get_Reaction()->Get_ReactionName() != "MyReactionName")
      continue; //not from the reaction you were interested in
   size_t locFailedActionIndex = locIterator->second; //this is the action where it was cut
   //check locFailedActionIndex to select combos that failed the action you wanted, then histogram or whatever. 
}

How do I perform an opposite (anti-, not-) cut?

  • After creating an action, set the inherited DAnalysisAction::dPerformAntiCut member to true (default false):
DMyAction* locAction = new DMyAction(...);
locAction->dPerformAntiCut = true;
locReaction->Add_AnalysisAction(locAction);

Analysis Actions (General)

How do I execute an analysis action from within another, custom action?

  • To encompass a reaction-independent action, just add it to your action like you would normally:
    • However, be careful that the action has a unique name if it creates objects in the output ROOT file. If you are executing it more than once, use it as a pointer and call a different constructor in Initialize(), one where you can pass a unique name.
//In your custom action header file:
#include "ANALYSIS/DHistogramActions.h"
#include "ANALYSIS/DCutActions.h"
 
//In the custom action class definition:
DHistogramAction_DetectedParticleKinematics dMyAction; //or whichever action you wanted to add
 
//In the custom action Initialize() method:
dMyAction.Initialize(locEventLoop);
 
//In the custom action Perform_Action() method, execute the action:
dMyAction(locEventLoop);
  • To encompass a reaction-dependent action:
//In your custom action header file:
#include "ANALYSIS/DHistogramActions.h"
#include "ANALYSIS/DCutActions.h"
 
//In the custom action class definition:
DHistogramAction_InvariantMass* dMyAction; //or whichever action you wanted to add //Note it's a POINTER
 
//In the custom action Initialize() method:
dMyAction = new DHistogramAction_InvariantMass(Get_Reaction(), Pi0, false, 1000, 0.1, 0.18); //setup the action like you would normally (unique name rules still apply!)
dMyAction->Initialize(locEventLoop);
 
//In the custom action Perform_Action() method, execute the action:
bool locActionResult = (*dMyAction)(locEventLoop, locParticleCombo); //locActionResult is true/false if it passed/failed the action's cut (if any)
  • Note that, with a little modification to the above examples, you can execute actions directly in your plugin's event processor instead.

How do I branch an analysis?

  • You cannot directly branch an analysis, where actions are only applied to a subset of the combos. Instead, you need to either write a custom action that encompasses all of the desired branching, or create a separate DReaction and place different cuts on it.
  • To save memory, it is HIGHLY RECOMMENDED that you reuse the pointers to the exact same DReactionStep objects for both DReactions. This allows the particle combos to share DParticleComboStep objects between them.
  • Don't worry, kinematic fits won't be performed twice; if it detects that the result will be identical to a previously-fit combo (from another DReaction), it will skip the fit and just copy the results.