]>
Commit | Line | Data |
---|---|---|
1ee39b3a | 1 | /************************************************************************* |
9ec5bce7 | 2 | * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. * |
3 | * * | |
4 | * Author: The ALICE Off-line Project. * | |
5 | * Contributors are mentioned in the code where appropriate. * | |
6 | * * | |
7 | * Permission to use, copy, modify and distribute this software and its * | |
8 | * documentation strictly for non-commercialf purposes is hereby granted * | |
9 | * without fee, provided that the above copyright notice appears in all * | |
10 | * copies and that both the copyright notice and this permission notice * | |
11 | * appear in the supporting documentation. The authors make no claims * | |
12 | * about the suitability of this software for any purpose. It is * | |
13 | * provided "as is" without express or implied warranty. * | |
14 | **************************************************************************/ | |
1ee39b3a | 15 | |
16 | /* $Id: AliTRDpidRefMakerNN.cxx 27496 2008-07-22 08:35:45Z cblume $ */ | |
17 | ||
18 | //////////////////////////////////////////////////////////////////////////// | |
19 | // // | |
20 | // Builds the reference tree for the training of neural networks // | |
21 | // // | |
22 | //////////////////////////////////////////////////////////////////////////// | |
23 | ||
24 | #include "TSystem.h" | |
25 | #include "TDatime.h" | |
26 | #include "TPDGCode.h" | |
27 | #include "TH1F.h" | |
28 | #include "TH2F.h" | |
29 | #include "TFile.h" | |
30 | #include "TGraphErrors.h" | |
31 | #include "TTree.h" | |
32 | #include "TEventList.h" | |
33 | #include "TMultiLayerPerceptron.h" | |
34 | ||
35 | #include "AliPID.h" | |
36 | #include "AliESDtrack.h" | |
37 | #include "AliTrackReference.h" | |
38 | ||
39 | #include "AliTRDtrackV1.h" | |
40 | #include "AliTRDReconstructor.h" | |
41 | #include "AliTRDpidUtil.h" | |
42 | #include "AliTRDpidRefMakerNN.h" | |
43 | #include "AliTRDpidUtil.h" | |
44 | ||
da27d983 | 45 | #include "Cal/AliTRDCalPID.h" |
46 | #include "Cal/AliTRDCalPIDNN.h" | |
1ee39b3a | 47 | #include "info/AliTRDtrackInfo.h" |
48 | #include "info/AliTRDv0Info.h" | |
49 | ||
50 | ClassImp(AliTRDpidRefMakerNN) | |
51 | ||
705f8b0a | 52 | //________________________________________________________________________ |
9ec5bce7 | 53 | AliTRDpidRefMakerNN::AliTRDpidRefMakerNN() |
705f8b0a | 54 | :AliTRDpidRefMaker() |
55 | ,fNet(NULL) | |
56 | ,fTrainMomBin(kAll) | |
57 | ,fEpochs(1000) | |
58 | ,fMinTrain(100) | |
59 | ,fDate(0) | |
60 | ,fDoTraining(0) | |
61 | ,fContinueTraining(0) | |
62 | ,fTrainPath(NULL) | |
63 | ,fScale(0) | |
64 | ,fLy(0) | |
65 | ,fNtrkl(0) | |
66 | ,fRef(NULL) | |
67 | { | |
68 | // | |
69 | // Default constructor | |
70 | // | |
71 | SetNameTitle("refMakerNN", "PID(NN) Reference Maker"); | |
72 | } | |
73 | ||
74 | //________________________________________________________________________ | |
75 | AliTRDpidRefMakerNN::AliTRDpidRefMakerNN(const char *name) | |
76 | :AliTRDpidRefMaker(name, "PID(NN) Reference Maker") | |
77 | ,fNet(NULL) | |
78 | ,fTrainMomBin(kAll) | |
79 | ,fEpochs(1000) | |
80 | ,fMinTrain(100) | |
81 | ,fDate(0) | |
82 | ,fDoTraining(0) | |
83 | ,fContinueTraining(0) | |
84 | ,fTrainPath(NULL) | |
85 | ,fScale(0) | |
86 | ,fLy(0) | |
87 | ,fNtrkl(0) | |
88 | ,fRef(NULL) | |
1ee39b3a | 89 | { |
90 | // | |
91 | // Default constructor | |
92 | // | |
93 | ||
9ec5bce7 | 94 | memset(fTrain, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); |
95 | memset(fTest, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); | |
96 | memset(fTrainData, 0, AliTRDCalPID::kNMom*sizeof(TTree*)); | |
1ee39b3a | 97 | |
98 | SetAbundance(.67); | |
99 | SetScaledEdx(Float_t(AliTRDCalPIDNN::kMLPscale)); | |
100 | TDatime datime; | |
101 | fDate = datime.GetDate(); | |
1ee39b3a | 102 | } |
103 | ||
104 | ||
105 | //________________________________________________________________________ | |
106 | AliTRDpidRefMakerNN::~AliTRDpidRefMakerNN() | |
107 | { | |
108 | } | |
109 | ||
110 | ||
111 | //________________________________________________________________________ | |
705f8b0a | 112 | void AliTRDpidRefMakerNN::MakeTrainTestTrees() |
1ee39b3a | 113 | { |
9ec5bce7 | 114 | // Create output file and tree |
1ee39b3a | 115 | // Called once |
116 | ||
705f8b0a | 117 | fRef = new TFile("TRD.CalibPIDrefMakerNN.root", "RECREATE"); |
9ec5bce7 | 118 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ |
119 | fTrainData[ip] = new TTree(Form("fTrainData_%d", ip), Form("NN Reference Data for MomBin %d", ip)); | |
120 | fTrainData[ip] -> Branch("fdEdx", fdEdx, Form("fdEdx[%d]/F", AliTRDpidUtil::kNNslices)); | |
121 | fTrainData[ip] -> Branch("fPID", fPID, Form("fPID[%d]/F", AliPID::kSPECIES)); | |
122 | fTrainData[ip] -> Branch("fLy", &fLy, "fLy/I"); | |
123 | fTrainData[ip] -> Branch("fNtrkl", &fNtrkl, "fNtrkl/I"); | |
124 | ||
125 | fTrain[ip] = new TEventList(Form("fTrainMom%d", ip), Form("Training list for momentum intervall %d", ip)); | |
126 | fTest[ip] = new TEventList(Form("fTestMom%d", ip), Form("Test list for momentum intervall %d", ip)); | |
127 | } | |
1ee39b3a | 128 | } |
129 | ||
130 | ||
131 | ||
132 | //________________________________________________________________________ | |
133 | Bool_t AliTRDpidRefMakerNN::PostProcess() | |
134 | { | |
135 | // Draw result to the screen | |
136 | // Called once at the end of the query | |
137 | ||
9ec5bce7 | 138 | TFile *fCalib = TFile::Open(Form("TRD.CalibPIDrefMaker.root")); |
139 | if (!fCalib) { | |
140 | AliError("Calibration file not available"); | |
141 | return kFALSE; | |
142 | } | |
705f8b0a | 143 | fData = (TTree*)fCalib->Get("refMakerNN"); |
9ec5bce7 | 144 | if (!fData) { |
145 | AliError("Calibration data not available"); | |
146 | return kFALSE; | |
147 | } | |
705f8b0a | 148 | TObjArray *o = NULL; |
149 | if(!(o = (TObjArray*)fCalib->Get("MonitorNN"))) { | |
9ec5bce7 | 150 | AliWarning("Missing monitoring container."); |
151 | return kFALSE; | |
152 | } | |
153 | fContainer = (TObjArray*)o->Clone("monitor"); | |
154 | ||
155 | ||
156 | if (!fRef) { | |
705f8b0a | 157 | AliDebug(2, "Loading file TRD.CalibPIDrefMakerNN.root"); |
158 | LoadFile("TRD.CalibPIDrefMakerNN.root"); | |
9ec5bce7 | 159 | } |
160 | else AliDebug(2, "file available"); | |
161 | ||
162 | if (!fRef) { | |
705f8b0a | 163 | MakeTrainTestTrees(); |
9ec5bce7 | 164 | |
165 | // Convert the CaliPIDrefMaker tree to 11 (different momentum bin) trees for NN training | |
166 | ||
167 | LinkPIDdata(); | |
168 | for(Int_t ip=0; ip < AliTRDCalPID::kNMom; ip++){ | |
169 | for(Int_t is=0; is < AliPID::kSPECIES; is++) { | |
170 | memset(fPID, 0, AliPID::kSPECIES*sizeof(Float_t)); | |
171 | fPID[is] = 1; | |
172 | Int_t n(0); // index of data | |
173 | for(Int_t itrk = 0; itrk<fData->GetEntries() && n<kMaxStat; itrk++){ | |
174 | if(!(fData->GetEntry(itrk))) continue; | |
175 | if(fPIDbin!=is) continue; | |
176 | for(Int_t ily=fPIDdataArray->fNtracklets; ily--;){ | |
177 | fLy = ily; | |
178 | fNtrkl = fPIDdataArray->fNtracklets; | |
179 | if((fPIDdataArray->fData[ily].fPLbin & 0xf)!= ip) continue; | |
180 | memset(fdEdx, 0, AliTRDpidUtil::kNNslices*sizeof(Float_t)); | |
181 | for(Int_t islice=AliTRDCalPID::kNSlicesNN; islice--;){ | |
182 | fdEdx[islice]+=fPIDdataArray->fData[ily].fdEdx[islice]; | |
183 | fdEdx[islice]/=fScale; | |
184 | } | |
185 | fTrainData[ip] -> Fill(); | |
186 | n++; | |
187 | } | |
188 | } | |
189 | AliDebug(2, Form("%d %d %d", ip, is, n)); | |
190 | } | |
191 | } | |
192 | ||
193 | ||
194 | fRef -> cd(); | |
195 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
196 | fTrainData[ip] -> Write(); | |
197 | } | |
198 | ||
199 | } | |
200 | else AliDebug(2, "file available"); | |
201 | ||
202 | ||
203 | // build the training and the test list for the neural networks | |
204 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
205 | MakeTrainingLists(ip); | |
206 | } | |
1ee39b3a | 207 | if(!fDoTraining) return kTRUE; |
208 | ||
9ec5bce7 | 209 | |
210 | ||
211 | // train the neural networks | |
1ee39b3a | 212 | gSystem->Exec(Form("mkdir ./Networks_%d/",fDate)); |
213 | AliDebug(2, Form("TrainMomBin [%d] [%d]", fTrainMomBin, kAll)); | |
214 | ||
215 | // train single network for a single momentum (recommended) | |
216 | if(!(fTrainMomBin == kAll)){ | |
9ec5bce7 | 217 | if(fTrain[fTrainMomBin] -> GetN() < fMinTrain){ |
218 | AliError("Warning in AliTRDpidRefMakerNN::PostProcess : Not enough events for training available! Please check Data sample!"); | |
1ee39b3a | 219 | return kFALSE; |
220 | } | |
221 | MakeRefs(fTrainMomBin); | |
1ee39b3a | 222 | MonitorTraining(fTrainMomBin); |
223 | } | |
224 | // train all momenta | |
225 | else{ | |
226 | for(Int_t iMomBin = 0; iMomBin < AliTRDCalPID::kNMom; iMomBin++){ | |
9ec5bce7 | 227 | if(fTrain[iMomBin] -> GetN() < fMinTrain){ |
228 | AliError(Form("Warning in AliTRDpidRefMakerNN::PostProcess : Not enough events for training available for momentum bin [%d]! Please check Data sample!", iMomBin)); | |
229 | continue; | |
1ee39b3a | 230 | } |
231 | MakeRefs(fTrainMomBin); | |
1ee39b3a | 232 | MonitorTraining(iMomBin); |
233 | } | |
234 | } | |
235 | ||
236 | return kTRUE; // testing protection | |
237 | } | |
238 | ||
239 | ||
240 | //________________________________________________________________________ | |
9ec5bce7 | 241 | void AliTRDpidRefMakerNN::MakeTrainingLists(Int_t mombin) |
1ee39b3a | 242 | { |
243 | // | |
244 | // build the training lists for the neural networks | |
245 | // | |
246 | ||
9ec5bce7 | 247 | if (!fRef) { |
248 | LoadFile(Form("TRD.Calib%s.root", GetName())); | |
1ee39b3a | 249 | } |
250 | ||
9ec5bce7 | 251 | if (!fRef) { |
252 | AliError("ERROR file for building training list not available"); | |
1ee39b3a | 253 | return; |
254 | } | |
255 | ||
9ec5bce7 | 256 | AliDebug(2, "\n Making training lists! \n"); |
1ee39b3a | 257 | |
9ec5bce7 | 258 | Int_t nPart[AliPID::kSPECIES]; |
259 | memset(nPart, 0, AliPID::kSPECIES*sizeof(Int_t)); | |
1ee39b3a | 260 | |
261 | // set needed branches | |
9ec5bce7 | 262 | fTrainData[mombin] -> SetBranchAddress("fdEdx", fdEdx); |
263 | fTrainData[mombin] -> SetBranchAddress("fPID", fPID); | |
264 | fTrainData[mombin] -> SetBranchAddress("fLy", &fLy); | |
265 | fTrainData[mombin] -> SetBranchAddress("fNtrkl", &fNtrkl); | |
266 | ||
1ee39b3a | 267 | // start first loop to check total number of each particle type |
9ec5bce7 | 268 | for(Int_t iEv=0; iEv < fTrainData[mombin] -> GetEntries(); iEv++){ |
269 | fTrainData[mombin] -> GetEntry(iEv); | |
1ee39b3a | 270 | |
271 | // use only events with goes through 6 layers TRD | |
9ec5bce7 | 272 | if(fNtrkl != AliTRDgeometry::kNlayer) continue; |
273 | ||
274 | if(fPID[AliPID::kElectron] == 1) | |
275 | nPart[AliPID::kElectron]++; | |
276 | else if(fPID[AliPID::kMuon] == 1) | |
277 | nPart[AliPID::kMuon]++; | |
278 | else if(fPID[AliPID::kPion] == 1) | |
279 | nPart[AliPID::kPion]++; | |
280 | else if(fPID[AliPID::kKaon] == 1) | |
281 | nPart[AliPID::kKaon]++; | |
282 | else if(fPID[AliPID::kProton] == 1) | |
283 | nPart[AliPID::kProton]++; | |
1ee39b3a | 284 | } |
285 | ||
286 | AliDebug(2, "Particle multiplicities:"); | |
9ec5bce7 | 287 | AliDebug(2, Form("Momentum[%d] Elecs[%d] Muons[%d] Pions[%d] Kaons[%d] Protons[%d]", mombin, nPart[AliPID::kElectron], nPart[AliPID::kMuon], nPart[AliPID::kPion], nPart[AliPID::kKaon], nPart[AliPID::kProton])); |
288 | ||
289 | ||
1ee39b3a | 290 | |
9ec5bce7 | 291 | // // implement counter of training and test sample size |
292 | Int_t iTrain = 0, iTest = 0; | |
1ee39b3a | 293 | |
294 | // set training sample size per momentum interval to 2/3 | |
295 | // of smallest particle counter and test sample to 1/3 | |
9ec5bce7 | 296 | iTrain = nPart[0]; |
297 | for(Int_t iPart = 1; iPart < AliPID::kSPECIES; iPart++){ | |
298 | // exclude muons and kaons if not availyable | |
299 | // this is neeeded since we do not have v0 candiates | |
300 | if((iPart == AliPID::kMuon || iPart == AliPID::kKaon) && (nPart[AliPID::kMuon] == 0 || nPart[AliPID::kKaon] == 0)) continue; | |
301 | if(iTrain > nPart[iPart]) | |
302 | iTrain = nPart[iPart]; | |
303 | } | |
304 | iTest = Int_t( iTrain * (1-fFreq)); | |
305 | iTrain = Int_t(iTrain * fFreq); | |
306 | AliDebug(2, Form("Momentum[%d] Train[%d] Test[%d]", mombin, iTrain, iTest)); | |
307 | ||
1ee39b3a | 308 | |
309 | ||
310 | // reset couters | |
9ec5bce7 | 311 | memset(nPart, 0, AliPID::kSPECIES*sizeof(Int_t)); |
1ee39b3a | 312 | |
313 | // start second loop to set the event lists | |
9ec5bce7 | 314 | for(Int_t iEv = 0; iEv < fTrainData[mombin] -> GetEntries(); iEv++){ |
315 | fTrainData[mombin] -> GetEntry(iEv); | |
316 | ||
317 | // set event list | |
318 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ | |
319 | if(nPart[is] < iTrain && fPID[is] == 1){ | |
320 | fTrain[mombin] -> Enter(iEv); | |
321 | nPart[is]++; | |
322 | } else if(nPart[is] < iTest+iTrain && fPID[is] == 1){ | |
323 | fTest[mombin] -> Enter(iEv); | |
324 | nPart[is]++; | |
1ee39b3a | 325 | } else continue; |
326 | } | |
327 | } | |
328 | ||
329 | AliDebug(2, "Particle multiplicities in both lists:"); | |
9ec5bce7 | 330 | AliDebug(2, Form("Momentum[%d] Elecs[%d] Muons[%d] Pions[%d] Kaons[%d] Protons[%d]", mombin, nPart[AliPID::kElectron], nPart[AliPID::kMuon], nPart[AliPID::kPion], nPart[AliPID::kKaon], nPart[AliPID::kProton])); |
331 | ||
332 | return; | |
1ee39b3a | 333 | } |
334 | ||
335 | ||
336 | //________________________________________________________________________ | |
337 | void AliTRDpidRefMakerNN::MakeRefs(Int_t mombin) | |
338 | { | |
339 | // | |
340 | // train the neural networks | |
341 | // | |
342 | ||
343 | ||
9ec5bce7 | 344 | if (!fTrainData[mombin]) LoadFile(Form("TRD.Calib%s.root", GetName())); |
1ee39b3a | 345 | |
9ec5bce7 | 346 | if (!fTrainData[mombin]) { |
1ee39b3a | 347 | AliError("Tree for training list not available"); |
348 | return; | |
349 | } | |
350 | ||
351 | TDatime datime; | |
352 | fDate = datime.GetDate(); | |
353 | ||
354 | AliDebug(2, Form("Training momentum bin %d", mombin)); | |
355 | ||
356 | // set variable to monitor the training and to save the development of the networks | |
357 | Int_t nEpochs = fEpochs/kMoniTrain; | |
358 | AliDebug(2, Form("Training %d times %d epochs", kMoniTrain, nEpochs)); | |
359 | ||
360 | // make directories to save the networks | |
361 | gSystem->Exec(Form("rm -r ./Networks_%d/MomBin_%d",fDate, mombin)); | |
362 | gSystem->Exec(Form("mkdir ./Networks_%d/MomBin_%d",fDate, mombin)); | |
363 | ||
364 | // variable to check if network can load weights from previous training | |
9ec5bce7 | 365 | Bool_t bFirstLoop = kTRUE; |
1ee39b3a | 366 | |
367 | // train networks over several loops and save them after each loop | |
368 | for(Int_t iLoop = 0; iLoop < kMoniTrain; iLoop++){ | |
9ec5bce7 | 369 | fTrainData[mombin] -> SetEventList(fTrain[mombin]); |
370 | fTrainData[mombin] -> SetEventList(fTest[mombin]); | |
1ee39b3a | 371 | |
9ec5bce7 | 372 | AliDebug(2, Form("Momentum[%d] Trainingloop[%d]", mombin, iLoop)); |
1ee39b3a | 373 | |
9ec5bce7 | 374 | // check if network is already implemented |
375 | if(bFirstLoop == kTRUE){ | |
376 | fNet = new TMultiLayerPerceptron("fdEdx[0],fdEdx[1],fdEdx[2],fdEdx[3],fdEdx[4],fdEdx[5],fdEdx[6],fdEdx[7]:15:7:fPID[0],fPID[1],fPID[2],fPID[3],fPID[4]!",fTrainData[mombin],fTrain[mombin],fTest[mombin]); | |
377 | fNet -> SetLearningMethod(TMultiLayerPerceptron::kStochastic); // set learning method | |
378 | fNet -> TMultiLayerPerceptron::SetEta(0.001); // set learning speed | |
379 | if(!fContinueTraining){ | |
380 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph"); | |
381 | else fNet -> Train(nEpochs,""); | |
1ee39b3a | 382 | } |
9ec5bce7 | 383 | else{ |
384 | fNet -> LoadWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fTrainPath, mombin, kMoniTrain - 1)); | |
385 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph+"); | |
386 | else fNet -> Train(nEpochs,"+"); | |
1ee39b3a | 387 | } |
9ec5bce7 | 388 | bFirstLoop = kFALSE; |
389 | } | |
390 | else{ | |
391 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph+"); | |
392 | else fNet -> Train(nEpochs,"+"); | |
393 | } | |
1ee39b3a | 394 | |
9ec5bce7 | 395 | // save weights for monitoring of the training |
396 | fNet -> DumpWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); | |
1ee39b3a | 397 | } // end training loop |
398 | } | |
399 | ||
400 | ||
401 | ||
402 | //________________________________________________________________________ | |
403 | void AliTRDpidRefMakerNN::MonitorTraining(Int_t mombin) | |
404 | { | |
405 | // | |
406 | // train the neural networks | |
407 | // | |
408 | ||
9ec5bce7 | 409 | if (!fTrainData[mombin]) LoadFile(Form("TRD.Calib%s.root", GetName())); |
410 | if (!fTrainData[mombin]) { | |
1ee39b3a | 411 | AliError("Tree for training list not available"); |
412 | return; | |
413 | } | |
414 | ||
415 | // init networks and set event list | |
416 | for(Int_t iChamb = 0; iChamb < AliTRDgeometry::kNlayer; iChamb++){ | |
9ec5bce7 | 417 | fNet = new TMultiLayerPerceptron("fdEdx[0],fdEdx[1],fdEdx[2],fdEdx[3],fdEdx[4],fdEdx[5],fdEdx[6],fdEdx[7]:15:7:fPID[0],fPID[1],fPID[2],fPID[3],fPID[4]!",fTrainData[mombin],fTrain[mombin],fTest[mombin]); |
418 | fTrainData[mombin] -> SetEventList(fTrain[mombin]); | |
419 | fTrainData[mombin] -> SetEventList(fTest[mombin]); | |
1ee39b3a | 420 | } |
421 | ||
422 | // implement variables for likelihoods | |
423 | Float_t like[AliPID::kSPECIES][AliTRDgeometry::kNlayer]; | |
424 | memset(like, 0, AliPID::kSPECIES*AliTRDgeometry::kNlayer*sizeof(Float_t)); | |
425 | Float_t likeAll[AliPID::kSPECIES], totProb; | |
426 | ||
427 | Double_t pionEffiTrain[kMoniTrain], pionEffiErrTrain[kMoniTrain]; | |
428 | Double_t pionEffiTest[kMoniTrain], pionEffiErrTest[kMoniTrain]; | |
429 | memset(pionEffiTrain, 0, kMoniTrain*sizeof(Double_t)); | |
430 | memset(pionEffiErrTrain, 0, kMoniTrain*sizeof(Double_t)); | |
431 | memset(pionEffiTest, 0, kMoniTrain*sizeof(Double_t)); | |
432 | memset(pionEffiErrTest, 0, kMoniTrain*sizeof(Double_t)); | |
433 | ||
434 | // init histos | |
435 | const Float_t epsilon = 1/(2*(AliTRDpidUtil::kBins-1)); // get nice histos with bin center at 0 and 1 | |
436 | TH1F *hElecs, *hPions; | |
437 | hElecs = new TH1F("hElecs","Likelihood for electrons", AliTRDpidUtil::kBins, 0.-epsilon, 1.+epsilon); | |
438 | hPions = new TH1F("hPions","Likelihood for pions", AliTRDpidUtil::kBins, 0.-epsilon, 1.+epsilon); | |
439 | ||
9ec5bce7 | 440 | TGraphErrors *gEffisTrain = new TGraphErrors(kMoniTrain); |
441 | gEffisTrain -> SetLineColor(4); | |
442 | gEffisTrain -> SetMarkerColor(4); | |
443 | gEffisTrain -> SetMarkerStyle(29); | |
444 | gEffisTrain -> SetMarkerSize(1); | |
445 | ||
446 | TGraphErrors *gEffisTest = new TGraphErrors(kMoniTrain); | |
447 | gEffisTest -> SetLineColor(2); | |
448 | gEffisTest -> SetMarkerColor(2); | |
449 | gEffisTest -> SetMarkerStyle(29); | |
450 | gEffisTest -> SetMarkerSize(1); | |
1ee39b3a | 451 | |
452 | AliTRDpidUtil *util = new AliTRDpidUtil(); | |
453 | ||
454 | // monitor training progress | |
455 | for(Int_t iLoop = 0; iLoop < kMoniTrain; iLoop++){ | |
456 | ||
457 | // load weights | |
9ec5bce7 | 458 | fNet -> LoadWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); |
459 | AliDebug(2, Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); | |
1ee39b3a | 460 | |
461 | // event loop training list | |
1ee39b3a | 462 | |
9ec5bce7 | 463 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ |
1ee39b3a | 464 | |
9ec5bce7 | 465 | if(!((is == AliPID::kElectron) || (is == AliPID::kPion))) continue; |
1ee39b3a | 466 | |
9ec5bce7 | 467 | Int_t iChamb = 0; |
468 | // reset particle probabilities | |
1ee39b3a | 469 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ |
9ec5bce7 | 470 | likeAll[iPart] = 1./AliPID::kSPECIES; |
1ee39b3a | 471 | } |
9ec5bce7 | 472 | totProb = 0.; |
1ee39b3a | 473 | |
9ec5bce7 | 474 | AliDebug(2, Form("%d",fTrain[mombin] -> GetN())); |
475 | for(Int_t iEvent = 0; iEvent < fTrain[mombin] -> GetN(); iEvent++ ){ | |
476 | fTrainData[mombin] -> GetEntry(fTrain[mombin] -> GetEntry(iEvent)); | |
477 | // use event only if it is electron or pion | |
478 | if(!(fPID[is] == 1.0)) continue; | |
479 | // get the probabilities for each particle type in each chamber | |
480 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
481 | like[iPart][iChamb] = fNet -> Result(fTrain[mombin] -> GetEntry(iEvent), iPart); | |
482 | likeAll[iPart] *= like[iPart][iChamb]; | |
483 | } | |
484 | //end chamber loop | |
485 | iChamb++; | |
486 | ||
487 | // get total probability and normalize it | |
488 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
489 | totProb += likeAll[iPart]; | |
490 | } | |
491 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
492 | likeAll[iPart] /= totProb; | |
493 | } | |
494 | ||
495 | if(iChamb == 5){ | |
496 | // fill likelihood distributions | |
497 | if(fPID[AliPID::kElectron] == 1) | |
498 | hElecs -> Fill(likeAll[AliPID::kElectron]); | |
499 | if(fPID[AliPID::kPion] == 1) | |
500 | hPions -> Fill(likeAll[AliPID::kElectron]); | |
501 | iChamb = 0; | |
502 | // reset particle probabilities | |
503 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
504 | likeAll[iPart] = 1./AliPID::kSPECIES; | |
505 | } | |
506 | totProb = 0.; | |
507 | } | |
508 | } // end event loop | |
509 | } // end species loop | |
510 | ||
1ee39b3a | 511 | |
512 | // calculate the pion efficiency and fill the graph | |
513 | util -> CalculatePionEffi(hElecs, hPions); | |
514 | pionEffiTrain[iLoop] = util -> GetPionEfficiency(); | |
515 | pionEffiErrTrain[iLoop] = util -> GetError(); | |
516 | ||
517 | gEffisTrain -> SetPoint(iLoop, iLoop+1, pionEffiTrain[iLoop]); | |
518 | gEffisTrain -> SetPointError(iLoop, 0, pionEffiErrTrain[iLoop]); | |
519 | hElecs -> Reset(); | |
520 | hPions -> Reset(); | |
521 | AliDebug(2, Form("TrainingLoop[%d] PionEfficiency[%f +/- %f]", iLoop, pionEffiTrain[iLoop], pionEffiErrTrain[iLoop])); | |
522 | // end training loop | |
523 | ||
524 | ||
525 | ||
9ec5bce7 | 526 | // monitor validation progress |
527 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ | |
1ee39b3a | 528 | |
9ec5bce7 | 529 | if(!((is == AliPID::kElectron) || (is == AliPID::kPion))) continue; |
530 | ||
531 | Int_t iChamb = 0; | |
1ee39b3a | 532 | // reset particle probabilities |
533 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
9ec5bce7 | 534 | likeAll[iPart] = 1./AliPID::kSPECIES; |
1ee39b3a | 535 | } |
536 | totProb = 0.; | |
537 | ||
9ec5bce7 | 538 | for(Int_t iEvent = 0; iEvent < fTest[mombin] -> GetN(); iEvent++ ){ |
539 | fTrainData[mombin] -> GetEntry(fTest[mombin] -> GetEntry(iEvent)); | |
540 | // use event only if it is electron or pion | |
541 | if(!(fPID[is] == 1.0)) continue; | |
542 | ||
543 | // get the probabilities for each particle type in each chamber | |
544 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
545 | like[iPart][iChamb] = fNet -> Result(fTest[mombin] -> GetEntry(iEvent), iPart); | |
546 | likeAll[iPart] *= like[iPart][iChamb]; | |
547 | } | |
548 | iChamb++; | |
549 | ||
550 | // get total probability and normalize it | |
551 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
552 | totProb += likeAll[iPart]; | |
553 | } | |
554 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
555 | likeAll[iPart] /= totProb; | |
556 | } | |
557 | ||
558 | if(iChamb == 5){ | |
559 | // fill likelihood distributions | |
560 | if(fPID[AliPID::kElectron] == 1) | |
561 | hElecs -> Fill(likeAll[AliPID::kElectron]); | |
562 | if(fPID[AliPID::kPion] == 1) | |
563 | hPions -> Fill(likeAll[AliPID::kElectron]); | |
564 | iChamb = 0; | |
565 | // reset particle probabilities | |
566 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
567 | likeAll[iPart] = 1./AliPID::kSPECIES; | |
568 | } | |
569 | totProb = 0.; | |
570 | } | |
571 | } // end event loop | |
572 | } // end species loop | |
573 | ||
1ee39b3a | 574 | // calculate the pion efficiency and fill the graph |
575 | util -> CalculatePionEffi(hElecs, hPions); | |
576 | pionEffiTest[iLoop] = util -> GetPionEfficiency(); | |
577 | pionEffiErrTest[iLoop] = util -> GetError(); | |
578 | ||
579 | gEffisTest -> SetPoint(iLoop, iLoop+1, pionEffiTest[iLoop]); | |
580 | gEffisTest -> SetPointError(iLoop, 0, pionEffiErrTest[iLoop]); | |
581 | hElecs -> Reset(); | |
582 | hPions -> Reset(); | |
583 | AliDebug(2, Form("TestLoop[%d] PionEfficiency[%f +/- %f] \n", iLoop, pionEffiTest[iLoop], pionEffiErrTest[iLoop])); | |
584 | ||
9ec5bce7 | 585 | } // end validation loop |
1ee39b3a | 586 | |
587 | util -> Delete(); | |
588 | ||
589 | gEffisTest -> Draw("PAL"); | |
590 | gEffisTrain -> Draw("PL"); | |
591 | ||
592 | } | |
593 | ||
594 | ||
595 | //________________________________________________________________________ | |
9ec5bce7 | 596 | Bool_t AliTRDpidRefMakerNN::LoadFile(const Char_t *InFileNN) |
1ee39b3a | 597 | { |
598 | // | |
599 | // Loads the files and sets the event list | |
9ec5bce7 | 600 | // for neural network training. |
1ee39b3a | 601 | // Useable for training outside of the makeResults.C macro |
602 | // | |
603 | ||
9ec5bce7 | 604 | fRef = TFile::Open(Form("%s", InFileNN)); |
605 | if(!fRef) return 0; | |
606 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
607 | fTrainData[ip] = (TTree*)fRef -> Get(Form("fTrainData_%d", ip)); | |
608 | } | |
1ee39b3a | 609 | |
610 | for(Int_t iMom = 0; iMom < AliTRDCalPID::kNMom; iMom++){ | |
9ec5bce7 | 611 | fTrain[iMom] = new TEventList(Form("fTrainMom%d", iMom), Form("Training list for momentum intervall %d", iMom)); |
612 | fTest[iMom] = new TEventList(Form("fTestMom%d", iMom), Form("Test list for momentum intervall %d", iMom)); | |
1ee39b3a | 613 | } |
9ec5bce7 | 614 | return 1; |
1ee39b3a | 615 | } |
616 | ||
617 |