1 / 22

What ORCA developers need to know about Persistent Analysis Objects

What ORCA developers need to know about Persistent Analysis Objects. Norbert Neumeister RPROM meeting February 2, 2004. Outline. What are analysis objects Why persistent analysis objects COBRA interfaces How to make analysis objects persistent HOWTO for ORCA developers Example.

newman
Télécharger la présentation

What ORCA developers need to know about Persistent Analysis Objects

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. What ORCA developers need to know about Persistent Analysis Objects Norbert Neumeister RPROM meeting February 2, 2004

  2. Outline • What are analysis objects • Why persistent analysis objects • COBRA interfaces • How to make analysis objects persistent • HOWTO for ORCA developers • Example Norbert Neumeister

  3. Introduction • Analysis Objects (reconstructed objects): • clusters, RecHits, track segments, etc. • tracks, vertices, muons, electrons, jets, etc. • Persistent analysis objects • store the result of event reconstruction DST/ESD • need a first prototype for DC04 • Persistency • Use COBRA/POOL • Use RootTrees (rootcint) • User defined persistency (e.g. ntuple like RootTree) Norbert Neumeister

  4. RecObject Persistency Requirements: • Ability to make the data part of the object persistent • Ability to store relations between RecObjects • Ability to query for the stored objects, and trigger reconstruction if the stored objects do not correspond to the query Norbert Neumeister

  5. Analysis Objects • Reconstructed objects are hierarchical: • High-level objects are constructed from lower level objects • Examples: Jets are constructed from Tracks, Vertices are constructed from Tracks, etc. • At the lowest level all objects are constructed from raw data (Digis) • A single reconstruction algorithm (aka RecUnit) may use several instances of lower level algorithms simultaneously • Reconstructed tracks from various algorithms (Kalman Filter, DAF, GSF) may be combined in a high level object. Norbert Neumeister

  6. COBRA/CARF • Up-to-now the only CARF mechanism for accessing RecObjects via a RecCollection was by string • Limitations: • No way to specify algorithm parameters in the query, or to get several RecCollections that differ only by algorithm parameter values • No way to specify reconstruction algorithms for lower level objects, like tracks for vertices • No way to know what parameters were used when reading a RecCollection Norbert Neumeister

  7. Problems • Collection of RecObjects is identified by a string • RecUnits (reconstruction algorithms) are just identified by a string (e.g. GlobalMuonReconstructor) • Dependencies on other RecObjects are not taken care of • The hierarchy of RecUnits is not handled. • A RecObject provided by a RecUnit may depend on • other RecUnits (in general a tree of components) • their associated parameters • the versions of the algorithms used It is necessary to get the configuration used to create the objects together with the object collection! Therefore a collection of reconstructed objects is not just identified by a string! Norbert Neumeister

  8. Requirements (I) • A collection of reconstructed objects is defined by: • Type of reconstructed object • Name of algorithm • similar to the current string • identifies the algorithm builder • Version of algorithm • the developer must set the version of an algorithm • avoid new version when only documentation is changed • ParameterSet(parameters of all algorithms but not of all components) • ComponentSet(dependency on other RecUnits) • Calibration • Save full dependency and check state of underlying objects • i.e. collection of jets depends on tracks and CaloClusters, tracks depend on tracker digis, etc. • Guarantee same results for same configuration Norbert Neumeister

  9. Requirements (II) • The user can access a collection of reconstructed objects by specifying the RecObject type and a RecQuery (see later) • Provide only one documented user interface to retrieve reconstructed objects: RecCollection • No pointer to G3EventProxy should be necessary • default is current event • new RecCollection interface Norbert Neumeister

  10. Configuration • The full configuration is defined by: • Name • Version • ParameterSet • ComponentSet • CalibrationSet • RecConfig defines the full configuration of a RecUnit • Defines “equal” operator • The operator returns true only if all parts compare equal • Every reconstruction algorithm needs a default configuration! • RecQuery • defines a subset of the full configuration • can obtain the full configuration from a registered RecUnit Norbert Neumeister

  11. RecConfig Settings Order of initialization: 1) defaultConfig of the algorithm • That’s the only place where parameter names and types are defined 2) .orcarc: Every parameter is configurable with the following syntax • AlgorithmName:ParameterName = value • overwrites 1) 3) The current RecQuery overwrites 2) For components same sequence as for parameters in a recursive way: AlgorithmName:ComponentName:ParameterName = value Norbert Neumeister

  12. ParameterSet (I) • A class that contains the complete set of RecUnit parameters exposed as configurable by the author • If a RecUnit is implemented by several algorithmic classes, like a TrackReconstructor which uses SeedGenerator, TrajectoryBuilder, etc. then all parameters of all such classes are grouped in a single parameter set • The Parameters of lower level algorithms are not part of the parameter set of a high-level algorithm Norbert Neumeister

  13. ParameterSet (II) • Parameters are organized as pair of string and value (name, value) • The name is in the scope of the ParameterSet and can be reused in other ParameterSets • The only way to instantiate a non-empty ParameterSet is by using a ParameterSetBuilder (prevent from changing an existing ParameterSet) • The following types are supported: int, bool, double, string • For double the equal operator allows tolerances • If nothing set explicitly all parameters are default values • Usage: ParameterSetBuilder builder; // add integer builder.addParameter(“numberOfTracks”,10); // add double with tolerance builder addParameter(“deltaRCut”,0.6,0.002) ParameterSet ps = builder.result(); int nTracks = ps.value<int>(“numberOfTracks”); Norbert Neumeister

  14. ComponetSet A B C D E F G • If a RecUnit uses the result of other RecUnits, like a Vertex algorithm uses reconstructed tracks, it should fully specify the configuration of the lower level RecUnits via RecConfigs. • A ComponentSet is a set of pairs (name, RecQuery) • The name is in the scope of the concrete component • The ComponentSet is a set of all RecQuery objects • use a recursive query! RecUnits: A, B, C, D, E, F, G B, C, D, E, F, G are components of A In addition RecUnit A can use a list of algorithms (TrajectoryBuilder, TrajectoryCleaner, TrajectorySmoother). Norbert Neumeister

  15. RecCollection Retrieving (persistent) objects: MyRecObj inherits from RecObject RecCollection<MyRecObj> myobjects(RecQuery("Name_of_algorithm“)); cout << myobjects.query() << endl; RecCollection<MyRecObj>::const_iterator it; Recommendation: Use RecCollection instead of AutoRecCollection or RecItr in ORCA. Norbert Neumeister

  16. Usage RecQuery q(“JetFinder”); q.setParameter("ConeSize", 0.77); RecQuery myTF(“TrackFinder”); q.setComponent("TrackReconstructor",myTF); ... RecCollection<Jet> myjets(q); cout << myjets.size() << endl; cout << myjets.query() << endl; Norbert Neumeister

  17. Recipe for Developers (I) • All objects you want to make persistent must inherit from the class RecObj • class Track : public RecObj • class Jet: public RecObj • For references to other reconstructed objects one must use TRecRef or TRecVec instead of pointers • E.g.: The class Jet owns a TRecRef<Track> for the leading track and a TRecVec<Track> for the jet constituents • The reconstruction algorithm must inherit from the class RecAlgorithm templated with the object type it has to reconstruct • class JetAlgo : public RecAlgorithm<Jet> Norbert Neumeister

  18. Recipe for Developers (II) • For all algorithm classes on has to implement the methods: • static RecConfig defaultConfig(); • virtual Name name() const { return defaultConfig().name(); } • virtual Version version() const { return defaultConfig().version(); } • virtual void reconstruct() • In the static method defaultConfig() the developer must fully specify the configuration of the algorithm. The configuration consists of: • name (string) • version (string) • list of parameters (pair of string and type) • list of components (pair of string and RecQuery) • … see example Norbert Neumeister

  19. Recipe for Developers (III) • A builder class must be provided using the class RecBuilder: • It will create a RecUnitFactory with same name as the RecUnit • It will establish the connection between transient and persistent objects • The simplest way is to do it like this: static PKBuilder< RecBuilder<JetAlgo> > jetbuilder("JetBuilder"); RecBuilder<JetAlgo> is equivalent to RecBuilder<JetAlgo,Jet> RecBuilder<JetAlgo, PJet> in case one needs a persistent jet class • In the src directory one has to provide two additional files • classes_def.xml, • classes.h Norbert Neumeister

  20. RecBuilder • The RecBuilder registers itself • One can only query for an RecCollection corresponding to a registered RecBuilder • The RecAlgorithm is instantiated at the time of the query • That’s where the names of parameters and components are checked. An exception is thrown if they don’t match. Norbert Neumeister

  21. Example • Trivial example which shows how to make reconstructed objects persistent in the CARF framework. • The example is based on a simple jet object which is built from track objects. • A jet contains a container of the tracks (constituents) and a reference to the most energetic track. • I addition to the object classes (Track, Jet) there are two fake track reconstruction classes TrackAlgo1 and TrackAlgo2 inheriting from a common base class TrackAlgo and a jet reconstruction algorithm (JetAlgo) • Have a look: /afs/cern.ch/cms/Releases/ORCA/prerelease/ORCA_7_7_0_pre1/src/CARF/PersistencyExample Norbert Neumeister

  22. Summary • Proposal for a new RecObject interface • Status • Prototype has been implemented (PRS + CCS) • Define a clean interface: • Review now • Use it for DC04: get experience in producing DSTs • Should be used by all high-level subsystems • For all analysis objects: Tracks, Muons, Electrons, Jets, etc. Norbert Neumeister

More Related