Fork me on GitHub

source: git/modules/VertexFinder.cc@ f688c89

ImprovedOutputFile Timing dual_readout llp
Last change on this file since f688c89 was 2118a6a, checked in by Michele Selvaggi <michele.selvaggi@…>, 9 years ago

fixed missing vertexing variables

  • Property mode set to 100644
File size: 15.2 KB
Line 
1/** \class VertexFinder
2 *
3 * Cluster vertices from tracks
4 *
5 * \authors A. Hart, M. Selvaggi
6 *
7 */
8
9
10#include "modules/VertexFinder.h"
11#include "classes/DelphesClasses.h"
12#include "classes/DelphesFactory.h"
13#include "classes/DelphesFormula.h"
14#include "classes/DelphesPileUpReader.h"
15
16#include "ExRootAnalysis/ExRootResult.h"
17#include "ExRootAnalysis/ExRootFilter.h"
18#include "ExRootAnalysis/ExRootClassifier.h"
19
20#include "TMath.h"
21#include "TString.h"
22#include "TFormula.h"
23#include "TRandom3.h"
24#include "TObjArray.h"
25#include "TDatabasePDG.h"
26#include "TLorentzVector.h"
27#include "TMatrixT.h"
28#include "TVector3.h"
29
30static const Double_t mm = 1.;
31static const Double_t m = 1000.*mm;
32static const Double_t ns = 1.;
33static const Double_t s = 1.e+9 *ns;
34static const Double_t c_light = 2.99792458e+8 * m/s;
35
36//------------------------------------------------------------------------------
37
38VertexFinder::VertexFinder() :
39 fSigma(0), fMinPT(0), fMaxEta(0), fSeedMinPT(0), fMinNDF(0), fGrowSeeds(0)
40{
41}
42
43//------------------------------------------------------------------------------
44
45VertexFinder::~VertexFinder()
46{
47}
48
49//------------------------------------------------------------------------------
50
51void VertexFinder::Init()
52{
53 fSigma = GetDouble("Sigma", 3.0);
54 fMinPT = GetDouble("MinPT", 0.1);
55 fMaxEta = GetDouble("MaxEta", 10.0);
56 fSeedMinPT = GetDouble("SeedMinPT", 5.0);
57 fMinNDF = GetInt("MinNDF", 4);
58 fGrowSeeds = GetInt("GrowSeeds", 1);
59
60 fInputArray = ImportArray(GetString("InputArray", "TrackSmearing/tracks"));
61 fItInputArray = fInputArray->MakeIterator();
62
63 fOutputArray = ExportArray(GetString("OutputArray", "tracks"));
64 fVertexOutputArray = ExportArray(GetString("VertexOutputArray", "vertices"));
65}
66
67//------------------------------------------------------------------------------
68
69void VertexFinder::Finish()
70{
71 if(fItInputArray) delete fItInputArray;
72}
73
74//------------------------------------------------------------------------------
75//
76Bool_t VertexFinder::secondAscending (pair<UInt_t, Double_t> pair0, pair<UInt_t, Double_t> pair1)
77{
78 return (pair0.second < pair1.second);
79}
80
81Bool_t VertexFinder::secondDescending (pair<UInt_t, Double_t> pair0, pair<UInt_t, Double_t> pair1)
82{
83 return (pair0.second > pair1.second);
84}
85
86void VertexFinder::Process()
87{
88 Candidate *candidate;
89
90////////////////////////////////////////////////////////////////////////////////
91// Clear the track and cluster maps before starting
92////////////////////////////////////////////////////////////////////////////////
93 trackIDToDouble.clear ();
94 trackIDToInt.clear ();
95 trackIDToBool.clear ();
96 clusterIDToDouble.clear ();
97 clusterIDToInt.clear ();
98 clusterIDToBool.clear ();
99 trackPT.clear ();
100 clusterSumPT2.clear ();
101////////////////////////////////////////////////////////////////////////////////
102
103////////////////////////////////////////////////////////////////////////////////
104// Create the initial cluster seeds
105////////////////////////////////////////////////////////////////////////////////
106 createSeeds ();
107////////////////////////////////////////////////////////////////////////////////
108
109////////////////////////////////////////////////////////////////////////////////
110// In order of descending seed pt, grow each cluster. If a cluster ends up with
111// fewer than MinNDF tracks, release the tracks for other clusters to claim.
112////////////////////////////////////////////////////////////////////////////////
113 sort (clusterSumPT2.begin (), clusterSumPT2.end (), secondDescending);
114 for (vector<pair<UInt_t, Double_t> >::const_iterator cluster = clusterSumPT2.begin (); cluster != clusterSumPT2.end (); cluster++)
115 {
116 // Skip the cluster if it no longer has any tracks
117 if (!clusterIDToInt.at (cluster->first).at ("ndf"))
118 continue;
119
120 // Grow the cluster if GrowSeeds is true
121 if (fGrowSeeds)
122 growCluster (cluster->first);
123
124 // If the cluster still has fewer than MinNDF tracks, release the tracks;
125 // otherwise, mark the seed track as claimed
126
127 if ((Int_t) clusterIDToInt.at (cluster->first).at ("ndf") < fMinNDF)
128 {
129 for (map<UInt_t, map<string, Int_t> >::iterator track = trackIDToInt.begin (); track != trackIDToInt.end (); track++)
130 {
131 if (track->second.at ("clusterIndex") != (Int_t) cluster->first)
132 continue;
133 track->second["clusterIndex"] = -1;
134 trackIDToBool[track->first]["claimed"] = false;
135 }
136 }
137 else
138 trackIDToBool[clusterIDToInt.at (cluster->first).at ("seed")]["claimed"] = true;
139 }
140////////////////////////////////////////////////////////////////////////////////
141
142////////////////////////////////////////////////////////////////////////////////
143// Add tracks to the output array after updating their ClusterIndex.
144////////////////////////////////////////////////////////////////////////////////
145 fItInputArray->Reset ();
146 while((candidate = static_cast<Candidate*>(fItInputArray->Next())))
147 {
148 if (candidate->Momentum.Pt () < fMinPT || fabs (candidate->Momentum.Eta ()) > fMaxEta)
149 continue;
150 candidate->ClusterIndex = trackIDToInt.at (candidate->GetUniqueID ()).at ("clusterIndex");
151 fOutputArray->Add(candidate);
152 }
153////////////////////////////////////////////////////////////////////////////////
154
155////////////////////////////////////////////////////////////////////////////////
156// Add clusters with at least MinNDF tracks to the output array in order of
157// descending sum(pt**2).
158////////////////////////////////////////////////////////////////////////////////
159 clusterSumPT2.clear ();
160 for (map<UInt_t, map<string, Int_t> >::const_iterator cluster = clusterIDToInt.begin (); cluster != clusterIDToInt.end (); cluster++)
161 {
162
163 if (cluster->second.at ("ndf") < fMinNDF)
164 continue;
165 clusterSumPT2.push_back (make_pair (cluster->first, clusterIDToDouble.at (cluster->first).at ("sumPT2")));
166 }
167 sort (clusterSumPT2.begin (), clusterSumPT2.end (), secondDescending);
168
169 for (vector<pair<UInt_t, Double_t> >::const_iterator cluster = clusterSumPT2.begin (); cluster != clusterSumPT2.end (); cluster++)
170 {
171 DelphesFactory *factory = GetFactory();
172 candidate = factory->NewCandidate();
173
174 candidate->ClusterIndex = cluster->first;
175 candidate->ClusterNDF = clusterIDToInt.at (cluster->first).at ("ndf");
176 candidate->ClusterSigma = fSigma;
177 candidate->SumPT2 = cluster->second;
178 candidate->Position.SetXYZT(0.0, 0.0, clusterIDToDouble.at (cluster->first).at ("z"), 0.0);
179 candidate->PositionError.SetXYZT(0.0, 0.0, clusterIDToDouble.at (cluster->first).at ("ez"), 0.0);
180
181 fVertexOutputArray->Add(candidate);
182 }
183////////////////////////////////////////////////////////////////////////////////
184}
185
186void
187VertexFinder::createSeeds ()
188{
189 Candidate *candidate;
190 UInt_t clusterIndex = 0, maxSeeds = 0;
191
192////////////////////////////////////////////////////////////////////////////////
193// Loop over all tracks, initializing some variables.
194////////////////////////////////////////////////////////////////////////////////
195 fItInputArray->Reset();
196 while((candidate = static_cast<Candidate*>(fItInputArray->Next())))
197 {
198 if (candidate->Momentum.Pt () < fMinPT || fabs (candidate->Momentum.Eta ()) > fMaxEta)
199 continue;
200
201 trackIDToDouble[candidate->GetUniqueID ()]["pt"] = candidate->Momentum.Pt ();
202 trackIDToDouble[candidate->GetUniqueID ()]["ept"] = candidate->ErrorPT ? candidate->ErrorPT : 1.0e-15;;
203 trackIDToDouble[candidate->GetUniqueID ()]["eta"] = candidate->Momentum.Eta ();
204
205 trackIDToDouble[candidate->GetUniqueID ()]["z"] = candidate->DZ;
206 trackIDToDouble[candidate->GetUniqueID ()]["ez"] = candidate->ErrorDZ ? candidate->ErrorDZ : 1.0e-15;
207
208 trackIDToInt[candidate->GetUniqueID ()]["clusterIndex"] = -1;
209 trackIDToInt[candidate->GetUniqueID ()]["interactionIndex"] = candidate->IsPU;
210
211 trackIDToBool[candidate->GetUniqueID ()]["claimed"] = false;
212
213 trackPT.push_back (make_pair (candidate->GetUniqueID (), candidate->Momentum.Pt ()));
214 }
215////////////////////////////////////////////////////////////////////////////////
216
217////////////////////////////////////////////////////////////////////////////////
218// Sort tracks by pt and leave only the SeedMinPT highest pt ones in the
219// trackPT vector.
220////////////////////////////////////////////////////////////////////////////////
221 sort (trackPT.begin (), trackPT.end (), secondDescending);
222 for (vector<pair<UInt_t, Double_t> >::const_iterator track = trackPT.begin (); track != trackPT.end (); track++, maxSeeds++)
223 {
224 if (track->second < fSeedMinPT)
225 break;
226 }
227 // If there are no tracks with pt above MinSeedPT, create just one seed from
228 // the highest pt track.
229 if (!maxSeeds)
230 maxSeeds++;
231 if (trackPT.size () > maxSeeds)
232 {
233 trackPT.erase (trackPT.begin () + maxSeeds, trackPT.end ());
234 }
235////////////////////////////////////////////////////////////////////////////////
236
237////////////////////////////////////////////////////////////////////////////////
238// Create the seeds from the SeedMinPT highest pt tracks.
239////////////////////////////////////////////////////////////////////////////////
240 for (vector<pair<UInt_t, Double_t> >::const_iterator track = trackPT.begin (); track != trackPT.end (); track++, clusterIndex++)
241 {
242 addTrackToCluster (track->first, clusterIndex);
243 clusterSumPT2.push_back (make_pair (clusterIndex, track->second * track->second));
244 }
245////////////////////////////////////////////////////////////////////////////////
246}
247
248void
249VertexFinder::growCluster (const UInt_t clusterIndex)
250{
251 Bool_t done = false;
252 UInt_t nearestID;
253 Int_t oldClusterIndex;
254 Double_t nearestDistance;
255 vector<UInt_t> nearTracks;
256 nearTracks.clear ();
257
258////////////////////////////////////////////////////////////////////////////////
259// Grow the cluster until there are no more tracks within Sigma standard
260// deviations of the cluster.
261////////////////////////////////////////////////////////////////////////////////
262 while (!done)
263 {
264 done = true;
265 nearestID = 0;
266 nearestDistance = -1.0;
267
268 // These two loops are for finding the nearest track to the cluster. The
269 // first time, the ID of each track within 10*Sigma of the cluster is
270 // saved in the nearTracks vector; subsequently, to save time, only the
271 // tracks in this vector are checked.
272 if (!nearTracks.size ())
273 {
274
275 for (map<UInt_t, map<string, Double_t> >::const_iterator track = trackIDToDouble.begin (); track != trackIDToDouble.end (); track++)
276 {
277 if (trackIDToBool.at (track->first).at ("claimed") || trackIDToInt.at (track->first).at ("clusterIndex") == (Int_t) clusterIndex)
278 continue;
279
280 Double_t distance = fabs (clusterIDToDouble.at (clusterIndex).at ("z") - track->second.at ("z")) / hypot (clusterIDToDouble.at (clusterIndex).at ("ez"), track->second.at ("ez"));
281 if (nearestDistance < 0.0 || distance < nearestDistance)
282 {
283 nearestID = track->first;
284 nearestDistance = distance;
285 }
286 if (distance < 10.0 * fSigma)
287 nearTracks.push_back (track->first);
288 }
289 }
290
291 else
292 {
293 for (vector<UInt_t>::const_iterator track = nearTracks.begin (); track != nearTracks.end (); track++)
294 {
295 if (trackIDToBool.at (*track).at ("claimed") || trackIDToInt.at (*track).at ("clusterIndex") == (Int_t) clusterIndex)
296 continue;
297 Double_t distance = fabs (clusterIDToDouble.at (clusterIndex).at ("z") - trackIDToDouble.at (*track).at ("z")) / hypot (clusterIDToDouble.at (clusterIndex).at ("ez"), trackIDToDouble.at (*track).at ("ez"));
298 if (nearestDistance < 0.0 || distance < nearestDistance)
299 {
300 nearestID = *track;
301 nearestDistance = distance;
302 }
303 }
304 }
305
306 // If no tracks within Sigma of the cluster were found, stop growing.
307 done = nearestDistance > fSigma || nearestDistance < 0.0;
308 if (done)
309 {
310 continue;
311 }
312
313 // Add the nearest track within Sigma to the cluster. If it already
314 // belonged to another cluster, remove it from that cluster first.
315 if (nearestDistance < fSigma)
316 {
317 oldClusterIndex = trackIDToInt.at (nearestID).at ("clusterIndex");
318 if (oldClusterIndex >= 0)
319 removeTrackFromCluster (nearestID, oldClusterIndex);
320
321 trackIDToBool[nearestID]["claimed"] = true;
322 addTrackToCluster (nearestID, clusterIndex);
323 }
324 }
325////////////////////////////////////////////////////////////////////////////////
326}
327
328Double_t
329VertexFinder::weight (const UInt_t trackID)
330{
331 return ((trackIDToDouble.at (trackID).at ("pt") / (trackIDToDouble.at (trackID).at ("ept") * trackIDToDouble.at (trackID).at ("ez"))) * (trackIDToDouble.at (trackID).at ("pt") / (trackIDToDouble.at (trackID).at ("ept") * trackIDToDouble.at (trackID).at ("ez"))));
332}
333
334void
335VertexFinder::removeTrackFromCluster (const UInt_t trackID, const UInt_t clusterID)
336{
337 Double_t wz = weight (trackID);
338
339 trackIDToInt[trackID]["clusterIndex"] = -1;
340 clusterIDToInt[clusterID]["ndf"]--;
341
342 clusterIDToDouble[clusterID]["sumZ"] -= wz * trackIDToDouble.at (trackID).at ("z");
343 clusterIDToDouble[clusterID]["errorSumZ"] -= wz * trackIDToDouble.at (trackID).at ("ez") * trackIDToDouble.at (trackID).at ("ez");
344 clusterIDToDouble[clusterID]["sumOfWeightsZ"] -= wz;
345 clusterIDToDouble[clusterID]["z"] = clusterIDToDouble.at (clusterID).at ("sumZ") / clusterIDToDouble.at (clusterID).at ("sumOfWeightsZ");
346 clusterIDToDouble[clusterID]["ez"] = sqrt ((1.0 / clusterIDToInt.at (clusterID).at ("ndf")) * (clusterIDToDouble.at (clusterID).at ("errorSumZ") / clusterIDToDouble.at (clusterID).at ("sumOfWeightsZ")));
347 clusterIDToDouble[clusterID]["sumPT2"] -= trackIDToDouble.at (trackID).at ("pt") * trackIDToDouble.at (trackID).at ("pt");
348}
349
350void
351VertexFinder::addTrackToCluster (const UInt_t trackID, const UInt_t clusterID)
352{
353 Double_t wz = weight (trackID);
354
355 if (!clusterIDToInt.count (clusterID))
356 {
357 clusterIDToInt[clusterID]["ndf"] = 0;
358 clusterIDToInt[clusterID]["seed"] = trackID;
359 clusterIDToDouble[clusterID]["sumZ"] = 0.0;
360 clusterIDToDouble[clusterID]["errorSumZ"] = 0.0;
361 clusterIDToDouble[clusterID]["sumOfWeightsZ"] = 0.0;
362 clusterIDToDouble[clusterID]["sumPT2"] = 0.0;
363 }
364
365 trackIDToInt[trackID]["clusterIndex"] = clusterID;
366 clusterIDToInt[clusterID]["ndf"]++;
367
368 clusterIDToDouble[clusterID]["sumZ"] += wz * trackIDToDouble.at (trackID).at ("z");
369 clusterIDToDouble[clusterID]["errorSumZ"] += wz * trackIDToDouble.at (trackID).at ("ez") * trackIDToDouble.at (trackID).at ("ez");
370 clusterIDToDouble[clusterID]["sumOfWeightsZ"] += wz;
371 clusterIDToDouble[clusterID]["z"] = clusterIDToDouble.at (clusterID).at ("sumZ") / clusterIDToDouble.at (clusterID).at ("sumOfWeightsZ");
372 clusterIDToDouble[clusterID]["ez"] = sqrt ((1.0 / clusterIDToInt.at (clusterID).at ("ndf")) * (clusterIDToDouble.at (clusterID).at ("errorSumZ") / clusterIDToDouble.at (clusterID).at ("sumOfWeightsZ")));
373 clusterIDToDouble[clusterID]["sumPT2"] += trackIDToDouble.at (trackID).at ("pt") * trackIDToDouble.at (trackID).at ("pt");
374}
375
376//------------------------------------------------------------------------------
Note: See TracBrowser for help on using the repository browser.