lar_cluster3d::DBScanAlg Class Reference

DBScanAlg class definiton. More...

Inheritance diagram for lar_cluster3d::DBScanAlg:

Public Types
enum	TimeValues {   BUILDTHREEDHITS = 0, BUILDHITTOHITMAP = 1, RUNDBSCAN = 2, BUILDCLUSTERINFO = 3,   PATHFINDING = 4, NUMTIMEVALUES }
	enumerate the possible values for time checking if monitoring timing More...

Public Member Functions
	DBScanAlg (fhicl::ParameterSet const &pset)
	Constructor. More...
void	Cluster3DHits (reco::HitPairList &hitPairList, reco::ClusterParametersList &clusterParametersList) const override
	Given a set of recob hits, run DBscan to form 3D clusters. More...
void	Cluster3DHits (reco::HitPairListPtr &hitPairList, reco::ClusterParametersList &clusterParametersList) const override
	Given a set of recob hits, run DBscan to form 3D clusters. More...
float	getTimeToExecute (IClusterAlg::TimeValues index) const override
	If monitoring, recover the time to execute a particular function. More...

Private Member Functions
void	expandCluster (const kdTree::KdTreeNode &, kdTree::CandPairList &, reco::ClusterParameters &, size_t) const
	the main routine for DBScan More...

Private Attributes
bool	m_enableMonitoring
	Data members to follow. More...
size_t	m_minPairPts
std::vector< float >	m_timeVector
std::unique_ptr< lar_cluster3d::IClusterParametersBuilder >	m_clusterBuilder
	Common cluster builder tool. More...
kdTree	m_kdTree

Detailed Description

DBScanAlg class definiton.

Definition at line 35 of file DBScanAlg_tool.cc.

Member Enumeration Documentation

enum lar_cluster3d::IClusterAlg::TimeValues

inherited

enumerate the possible values for time checking if monitoring timing

Enumerator
BUILDTHREEDHITS
BUILDHITTOHITMAP
RUNDBSCAN
BUILDCLUSTERINFO
PATHFINDING
NUMTIMEVALUES

Definition at line 49 of file IClusterAlg.h.

                    {
      BUILDTHREEDHITS = 0,
      BUILDHITTOHITMAP = 1,
      RUNDBSCAN = 2,
      BUILDCLUSTERINFO = 3,
      PATHFINDING = 4,
      NUMTIMEVALUES
    };

Constructor & Destructor Documentation

lar_cluster3d::DBScanAlg::DBScanAlg ( fhicl::ParameterSet const &  pset )

explicit

Constructor.

Parameters

pset

Definition at line 91 of file DBScanAlg_tool.cc.

References Get, fhicl::ParameterSet::get(), m_clusterBuilder, m_enableMonitoring, m_kdTree, m_minPairPts, and fhicl::ParameterSet::put_or_replace().

  {
    m_enableMonitoring = pset.get<bool>("EnableMonitoring", true);
    m_minPairPts = pset.get<size_t>("MinPairPts", 2);

    m_clusterBuilder = art::make_tool<lar_cluster3d::IClusterParametersBuilder>(
      pset.get<fhicl::ParameterSet>("ClusterParamsBuilder"));

    // Recover the parameter set for the kdTree
    fhicl::ParameterSet kdTreeParams(pset.get<fhicl::ParameterSet>("kdTree"));

    // Now work out the maximum wire pitch
    auto const& wireReadoutGeom = art::ServiceHandle<geo::WireReadout>()->Get();

    // Returns the wire pitch per plane assuming they will be the same for all TPCs
    constexpr geo::TPCID tpcid{0, 0};
    std::vector<double> const wirePitchVec{
      wireReadoutGeom.Plane(geo::PlaneID{tpcid, 0}).WirePitch(),
      wireReadoutGeom.Plane(geo::PlaneID{tpcid, 1}).WirePitch(),
      wireReadoutGeom.Plane(geo::PlaneID{tpcid, 2}).WirePitch()};

    float maxBestDist = 1.99 * *std::max_element(wirePitchVec.begin(), wirePitchVec.end());

    kdTreeParams.put_or_replace<float>("RefLeafBestDist", maxBestDist);

    m_kdTree = kdTree(kdTreeParams);
  }

Member Function Documentation

void lar_cluster3d::DBScanAlg::Cluster3DHits ( reco::HitPairList &  hitPairList, reco::ClusterParametersList &  clusterParametersList  ) const

overridevirtual

Given a set of recob hits, run DBscan to form 3D clusters.

Parameters

hitPairList	The input list of 3D hits to run clustering on
clusterParametersList	A list of cluster objects (parameters from associated hits)

Driver for processing input 2D hits, transforming to 3D hits and building lists of associated 3D hits (candidate 3D clusters)

Implements lar_cluster3d::IClusterAlg.

Definition at line 119 of file DBScanAlg_tool.cc.

References reco::ClusterParameters::addHit3D(), lar_cluster3d::IClusterAlg::BUILDCLUSTERINFO, lar_cluster3d::IClusterAlg::BUILDHITTOHITMAP, lar_cluster3d::kdTree::BuildKdTree(), reco::ClusterHit3D::CLUSTERATTACHED, reco::ClusterHit3D::CLUSTERNOISE, reco::ClusterHit3D::CLUSTERVISITED, expandCluster(), lar_cluster3d::kdTree::FindNearestNeighbors(), lar_cluster3d::kdTree::getTimeToExecute(), m_clusterBuilder, m_enableMonitoring, m_kdTree, m_minPairPts, m_timeVector, lar_cluster3d::IClusterAlg::NUMTIMEVALUES, and lar_cluster3d::IClusterAlg::RUNDBSCAN.

  {
    cet::cpu_timer theClockDBScan;

    m_timeVector.resize(NUMTIMEVALUES, 0.);

    // DBScan is driven of its "epsilon neighborhood". Computing adjacency within DBScan can be time
    // consuming so the idea is the prebuild the adjaceny map and then run DBScan.
    // We'll employ a kdTree to implement this scheme
    kdTree::KdTreeNodeList kdTreeNodeContainer;
    kdTree::KdTreeNode topNode = m_kdTree.BuildKdTree(hitPairList, kdTreeNodeContainer);

    if (m_enableMonitoring) m_timeVector[BUILDHITTOHITMAP] = m_kdTree.getTimeToExecute();

    if (m_enableMonitoring) theClockDBScan.start();

    // Ok, here we go!
    // The idea is to loop through all of the input 3D hits and do the clustering
    for (const auto& hit : hitPairList) {
      // Check if the hit has already been visited
      if (hit.getStatusBits() & reco::ClusterHit3D::CLUSTERVISITED) continue;

      // Mark as visited
      hit.setStatusBit(reco::ClusterHit3D::CLUSTERVISITED);

      // Find the neighborhood for this hit
      kdTree::CandPairList candPairList;
      float bestDistance(std::numeric_limits<float>::max());

      m_kdTree.FindNearestNeighbors(&hit, topNode, candPairList, bestDistance);

      if (candPairList.size() < m_minPairPts) {
        hit.setStatusBit(reco::ClusterHit3D::CLUSTERNOISE);
      }
      else {
        // "Create" a new cluster and get a reference to it
        clusterParametersList.push_back(reco::ClusterParameters());

        reco::ClusterParameters& curCluster = clusterParametersList.back();

        hit.setStatusBit(reco::ClusterHit3D::CLUSTERATTACHED);
        curCluster.addHit3D(&hit);

        // expand the cluster
        expandCluster(topNode, candPairList, curCluster, m_minPairPts);
      }
    }

    if (m_enableMonitoring) {
      theClockDBScan.stop();

      m_timeVector[RUNDBSCAN] = theClockDBScan.accumulated_real_time();
    }

    // Initial clustering is done, now trim the list and get output parameters
    cet::cpu_timer theClockBuildClusters;

    // Start clocks if requested
    if (m_enableMonitoring) theClockBuildClusters.start();

    m_clusterBuilder->BuildClusterInfo(clusterParametersList);

    if (m_enableMonitoring) {
      theClockBuildClusters.stop();

      m_timeVector[BUILDCLUSTERINFO] = theClockBuildClusters.accumulated_real_time();
    }

    mf::LogDebug("Cluster3D") << ">>>>> DBScan done, found " << clusterParametersList.size()
                              << " clusters" << std::endl;
  }

void lar_cluster3d::DBScanAlg::Cluster3DHits ( reco::HitPairListPtr &  hitPairList, reco::ClusterParametersList &  clusterParametersList  ) const

overridevirtual

Given a set of recob hits, run DBscan to form 3D clusters.

Parameters

hitPairList	The input list of 3D hits to run clustering on
clusterParametersList	A list of cluster objects (parameters from associated hits)

Driver for processing input 2D hits, transforming to 3D hits and building lists of associated 3D hits (candidate 3D clusters)

Implements lar_cluster3d::IClusterAlg.

Definition at line 196 of file DBScanAlg_tool.cc.

References reco::ClusterParameters::addHit3D(), lar_cluster3d::IClusterAlg::BUILDCLUSTERINFO, lar_cluster3d::IClusterAlg::BUILDHITTOHITMAP, lar_cluster3d::kdTree::BuildKdTree(), reco::ClusterHit3D::CLUSTERATTACHED, reco::ClusterHit3D::CLUSTERNOISE, reco::ClusterHit3D::CLUSTERVISITED, expandCluster(), lar_cluster3d::kdTree::FindNearestNeighbors(), lar_cluster3d::kdTree::getTimeToExecute(), m_clusterBuilder, m_enableMonitoring, m_kdTree, m_minPairPts, m_timeVector, lar_cluster3d::IClusterAlg::NUMTIMEVALUES, and lar_cluster3d::IClusterAlg::RUNDBSCAN.

  {
    cet::cpu_timer theClockDBScan;

    m_timeVector.resize(NUMTIMEVALUES, 0.);

    // DBScan is driven of its "epsilon neighborhood". Computing adjacency within DBScan can be time
    // consuming so the idea is the prebuild the adjaceny map and then run DBScan.
    // We'll employ a kdTree to implement this scheme
    kdTree::KdTreeNodeList kdTreeNodeContainer;
    kdTree::KdTreeNode topNode = m_kdTree.BuildKdTree(hitPairList, kdTreeNodeContainer);

    if (m_enableMonitoring) m_timeVector[BUILDHITTOHITMAP] = m_kdTree.getTimeToExecute();

    if (m_enableMonitoring) theClockDBScan.start();

    // Ok, here we go!
    // The idea is to loop through all of the input 3D hits and do the clustering
    for (const auto& hit : hitPairList) {
      // Check if the hit has already been visited
      if (hit->getStatusBits() & reco::ClusterHit3D::CLUSTERVISITED) continue;

      // Mark as visited
      hit->setStatusBit(reco::ClusterHit3D::CLUSTERVISITED);

      // Find the neighborhood for this hit
      kdTree::CandPairList candPairList;
      float bestDistance(std::numeric_limits<float>::max());

      m_kdTree.FindNearestNeighbors(hit, topNode, candPairList, bestDistance);

      if (candPairList.size() < m_minPairPts) {
        hit->setStatusBit(reco::ClusterHit3D::CLUSTERNOISE);
      }
      else {
        // "Create" a new cluster and get a reference to it
        clusterParametersList.push_back(reco::ClusterParameters());

        reco::ClusterParameters& curCluster = clusterParametersList.back();

        hit->setStatusBit(reco::ClusterHit3D::CLUSTERATTACHED);
        curCluster.addHit3D(hit);

        // expand the cluster
        expandCluster(topNode, candPairList, curCluster, m_minPairPts);
      }
    }

    if (m_enableMonitoring) {
      theClockDBScan.stop();

      m_timeVector[RUNDBSCAN] = theClockDBScan.accumulated_real_time();
    }

    // Initial clustering is done, now trim the list and get output parameters
    cet::cpu_timer theClockBuildClusters;

    // Start clocks if requested
    if (m_enableMonitoring) theClockBuildClusters.start();

    m_clusterBuilder->BuildClusterInfo(clusterParametersList);

    if (m_enableMonitoring) {
      theClockBuildClusters.stop();

      m_timeVector[BUILDCLUSTERINFO] = theClockBuildClusters.accumulated_real_time();
    }

    mf::LogDebug("Cluster3D") << ">>>>> DBScan done, found " << clusterParametersList.size()
                              << " clusters" << std::endl;
  }

void lar_cluster3d::DBScanAlg::expandCluster ( const kdTree::KdTreeNode &  topNode, kdTree::CandPairList &  candPairList, reco::ClusterParameters &  cluster, size_t  minPts  ) const

private

the main routine for DBScan

Definition at line 273 of file DBScanAlg_tool.cc.

References reco::ClusterParameters::addHit3D(), reco::ClusterHit3D::CLUSTERATTACHED, reco::ClusterHit3D::CLUSTERVISITED, DEFINE_ART_CLASS_TOOL, lar_cluster3d::kdTree::FindNearestNeighbors(), reco::ClusterHit3D::getStatusBits(), m_kdTree, and reco::ClusterHit3D::setStatusBit().

Referenced by Cluster3DHits(), and getTimeToExecute().

  {
    // This is the main inside loop for the DBScan based clustering algorithm

    // Loop over added hits until list has been exhausted
    while (!candPairList.empty()) {
      // Dereference the point so we can see in the debugger...
      const reco::ClusterHit3D* neighborHit = candPairList.front().second;

      // Process if we've not been here before
      if (!(neighborHit->getStatusBits() & reco::ClusterHit3D::CLUSTERVISITED)) {
        // set as visited
        neighborHit->setStatusBit(reco::ClusterHit3D::CLUSTERVISITED);

        // get the neighborhood around this point
        kdTree::CandPairList neighborCandPairList;
        float bestDistance(std::numeric_limits<float>::max());

        m_kdTree.FindNearestNeighbors(neighborHit, topNode, neighborCandPairList, bestDistance);

        // If the epsilon neighborhood of this point is large enough then add its points to our list
        if (neighborCandPairList.size() >= minPts) {
          std::copy(neighborCandPairList.begin(),
                    neighborCandPairList.end(),
                    std::back_inserter(candPairList));
        }
      }

      // If the point is not yet in a cluster then we now add
      if (!(neighborHit->getStatusBits() & reco::ClusterHit3D::CLUSTERATTACHED)) {
        neighborHit->setStatusBit(reco::ClusterHit3D::CLUSTERATTACHED);
        cluster.addHit3D(neighborHit);
      }

      candPairList.pop_front();
    }
  }

float lar_cluster3d::DBScanAlg::getTimeToExecute ( IClusterAlg::TimeValues  index ) const

inlineoverridevirtual

If monitoring, recover the time to execute a particular function.

Implements lar_cluster3d::IClusterAlg.

Definition at line 65 of file DBScanAlg_tool.cc.

References expandCluster(), and m_timeVector.

    {
      return m_timeVector[index];
    }

Member Data Documentation

std::unique_ptr<lar_cluster3d::IClusterParametersBuilder> lar_cluster3d::DBScanAlg::m_clusterBuilder

private

Common cluster builder tool.

Definition at line 87 of file DBScanAlg_tool.cc.

Referenced by Cluster3DHits(), and DBScanAlg().

bool lar_cluster3d::DBScanAlg::m_enableMonitoring

private

Data members to follow.

Definition at line 82 of file DBScanAlg_tool.cc.

Referenced by Cluster3DHits(), and DBScanAlg().

kdTree lar_cluster3d::DBScanAlg::m_kdTree

private

Definition at line 88 of file DBScanAlg_tool.cc.

Referenced by Cluster3DHits(), DBScanAlg(), and expandCluster().

size_t lar_cluster3d::DBScanAlg::m_minPairPts

private

Definition at line 83 of file DBScanAlg_tool.cc.

Referenced by Cluster3DHits(), and DBScanAlg().

std::vector<float> lar_cluster3d::DBScanAlg::m_timeVector

mutableprivate

Definition at line 84 of file DBScanAlg_tool.cc.

Referenced by Cluster3DHits(), and getTimeToExecute().

---

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

• larreco/v10_01_07/source/larreco/RecoAlg/Cluster3DAlgs/DBScanAlg_tool.cc