Things you need to know
- psyche: Propagation of SYstematics and CHaracterization of Events
- highland: HIGH Level ANalysis Development
- psyche is a generalization of the BANFF interface. It has:
- Event selections and systematic propagation
- Is the core of highland2:
- Basic data classes, tools, selections and systematics
- highland2 provides in addition:
- Corrections
- Functionality to create micro-trees for analysis
- Drawing tools (also for systematics)
- Data reduction tools: Flat Tree creation
- Analyses do not inherit from each other anymore, but use each other. i.e numuCCMultiPiAnalysis uses numuCCAnalysis
- All Analyses inherit ultimately from AnalysisAlgorithm in highlandCore.
- All tracker analyses inherit from baseTrackerAnalysis class (which inherits from baseAnalysis)
- All P0D analyses inherit from baseP0DAnalysis class (which inherits from baseAnalysis)
- The concept of selection has been separated from the one of analysis. A highland analysis could use several selections. Thus a new class SelectionBase has been introduced. The selection has to be defined in that class and not in the analysis it self as it was done in highland. Have a look at numuCCAnalysis.cxx (in highland2/numuCCAnalysis/v0r0/src) and numuCCSelection.cxx (in psyche/psycheSelections/vXrY/src), and this distiction should be clear.
- All corrections are in highland2: the input tree for psyche contains all corrections
- Mature selections and systematics are in psyche. However new selections and systematics MUST be developed first in highland2:
- You can have your mySelection.cxx file in your highland2/myAnalysis directory and move it to psycheSelections once it is mature enough.
- highland2 analysis packages should mainly take care of filling micro-tree variables for the selection and systematics development.
- highland2 is much faster than highland (>100 times)
- use of arrays instead of vectors in psyche and highland
- use of vectors instead of maps
- Many things were unecessarily computed several times
- direct access to branch addreses when filling micro-tree variables by specifying the variable index and not the variable name
Bugzilla
We already have separated entries for psyche and highland2 in bugzilla
Structure
nd280AnalysisTools: highland2 needs this to understand oaAnalysis files and make the true-reco association.
psyche
- psycheCore
- psycheEventModel
- psycheUtils
- psycheND280Utils
- psycheIO
- psycheSelections
- psycheSystematics
- psycheSteering
highland2
nd280Psyche: this is the psyche master package used for releases and recursive check out nd280Highland2: this is the highland2 master package used for releases and recursive check out
Event selection
- Every event selection inherits from SelectionBase class in psycheCore and implements 4 mandatory methods:
- DefineSteps: The selection is a collection of steps (cuts and actions) that msut be defined
- MakeBox: The box (AnaBoxB) contains information that has to be passed from one step to another. This method creates the box of the appropriate type (AnaBoxB or a derived one)
- FillEventSummary: The EventSummary class contains relevant info for the fitters (BANFF, MaCh3), as LeptonCandidate properties, True Vertex properties or event sample in which the event has been classified
- GetSampleEnum: It should return the appropriate sample type (from nd280amples.hxx in psycheCore)
- Several selections that are related (share some steps) can be run in parallel using the branch functionaliy. This was already available in highland but it is much better now.
Creating a new analysis in highland2
- Create a cmt package (i.e. myAnalysis) inside highland2.
- go outside highland2 directory
- cmt create myAnalysis v0r0 highland2
- This will create a cmt package highland2/myAnalysis
- In the src directory create:
- myAnalysis.cxx/hxx
- mySelection.cxx/hxx
- Fill those files taking as examples numuCCAnalysis or numuCCMultiPiAnalysis, both in highland2, and numuCCSelection or numuCCMultiPiSelection, both under psyche/psycheSelections
- Once mySelection is mature enough it could be officialized (moved into psycheSelections)
Migrating an analysis from highland to highland2
Following the previous instructions migration from highland to highland2 can proceed. Some recommendations:
- Don't expect to have same number of events with highland and highland2 if your selection is not based on the numuCCMultiPi or antiNumuCC selections.
- Only those selectons have been validated
- Don't worry about systematics yet. Just make sure that you can run your selection with highland2 and produce some reasonable plots with the DrawingTools
- Once your analysis/selection is implemented, compiled and run, compare the results with the old highland
- Final number of selected events
- Events after each cut, since we should try to see at which point in the selection the results differ
- Events for each category
- Submit a bug report (there are separate entries for psyche and highland2 in bugzilla) specifying the problem and any clues you may have
Flat Trees
FlatTree is now fully working and validated in highland2. It is highly recomended to use highland2 files since few issues has been solved.
In previous versions oo highland2 it was possible to run over old highland FlatTrees, although to do that you had to comment ou the line
#define USEHIGHLAND2
in FlatTreeConverter.hxx
This functionality is not available anymore. Everyone is supposed to use highland2 FlatTrees when running highland2 analyses
DrawingTools
They are similar to the ones in highland. Some interesting new methods you can use:
- DumpConfigurations()
- DumpConfiguration(name)
- DumpSystematics()
- DumpCorrections()
- DumpSofwareVersions()
highland --> highland2 quick table
- TVector3 and TLorentzVector are now arrays, therefore
- .X() —> [0]
- .Y() —> [1]
- .Z() —> [2]
- For arguments in functions:
- const TVector3& pos —> const Float_t* pos (same for TLorenzVector)
- –> a 4-vector can be passed without .Vect() (that is not defined anymore)
- Vars in the microtree are now indexed in a enum in yourAnalysis.hxx (enum enumStandardMicroTrees_yourAnalysis):
- AddVar("varname", "I"); —> AddVarI(output(), varname, "dox"); (the last argument might be empty but not removed: WithDoc by default)
- AddToyVar("varname", "I"); —> AddToyVarI(output(), varname, "dox");
- AddVectorVar("varname", "dox", "I", 3); —> AddVar3VI(output(), varname, "dox"); (AddVar4VI for 4-vectors)
- for float, just replace I with F or D or C (all definitions in highlandTools/vXrY/src/OutputManager.hxx)
- in DefineTruthTree() same method, no need to add "truth" as first argument
- Similarly for filling the vars:
- FillVar("varname", varvalue); —> output().FillVar(varname, varvalue);
- FillToyVar("varname", varvalue); —> output().FillToyVar(varname, varvalue);
- FillVectorVar("varname", varvalue); —> output().FillVectorVar(varname, varvalue, vectorsize); (note the vectorsize to be added!)
- ND::box(). —> box().
- ND::your_box(). —> _your_box-> (to be defined in InitializeAnalyis as a class to be defined in yourSelection.hxx)
- numuCCAnalyis:: —> _numuCCAnalysis-> (to be defined in yourSelection.hxx)
- std::string det = "FGD1" —> SubDetId::SubDetEnum = SubDetId::kFGD1 (etc for other subdetectors)
-For categories
- COUTFV —> CATOUTFV
- COOUTFV —> COLOUTFV
- CNOTRUTH —> CATNOTRUTH
- CONOTRUTH —> COLNOTRUTH
- COTHER —> CATOTHER
- COOTHER —> COLOTHER
Some AnaTrueVertex need to become AnaTrueVertexB (if pointing to psyche, "B" stand for "Base", i.e. it is in psyche) and afterwards it might need to be casted to AnaTrueVertex with: static_cast<AnaTrueVertex*>(trueTrack->TrueVertex)
std_actions:: —> anaUtils:: (common methods might be in psycheUtils or in baseAnalysis/CategoriesUtils or in highlandUtils (just created in the head), with the same namespace anaUtils, and you have to #include the needed hxx, i.e. AnalysisUtils.hxx or CategoriesUtils.hxx or ??? (see in highlandUtils)
- In yourAnalysis.hxx / .cxx:
- #include yourSelection.hxx instead of StandardsActionsYour.hxx, StandardCutsYour.hxx and AnalysisBoxYour.hxx: yourSelection has to contain all of them
- #include OutputManager.hxx
- void -> virtual void
- DefineSelection -> DefineSelections (plural)
- Separate DefineCorrections and DefineSystematics from DefineConfigurations (that will just contain single/all parameters)
- MultiAnalysisTools -> DataSample
