]>
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" | |
a5a3321d | 49 | #include "info/AliTRDpidInfo.h" |
1ee39b3a | 50 | |
51 | ClassImp(AliTRDpidRefMakerNN) | |
52 | ||
705f8b0a | 53 | //________________________________________________________________________ |
9ec5bce7 | 54 | AliTRDpidRefMakerNN::AliTRDpidRefMakerNN() |
705f8b0a | 55 | :AliTRDpidRefMaker() |
56 | ,fNet(NULL) | |
57 | ,fTrainMomBin(kAll) | |
58 | ,fEpochs(1000) | |
59 | ,fMinTrain(100) | |
60 | ,fDate(0) | |
61 | ,fDoTraining(0) | |
62 | ,fContinueTraining(0) | |
63 | ,fTrainPath(NULL) | |
64 | ,fScale(0) | |
65 | ,fLy(0) | |
66 | ,fNtrkl(0) | |
67 | ,fRef(NULL) | |
68 | { | |
69 | // | |
70 | // Default constructor | |
71 | // | |
d80a6a00 | 72 | SetNameTitle("PIDrefMakerNN", "PID(NN) Reference Maker"); |
73 | ||
74 | memset(fTrain, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); | |
75 | memset(fTest, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); | |
76 | memset(fTrainData, 0, AliTRDCalPID::kNMom*sizeof(TTree*)); | |
77 | ||
78 | SetAbundance(.67); | |
79 | SetScaledEdx(Float_t(AliTRDCalPIDNN::kMLPscale)); | |
80 | TDatime datime; | |
81 | fDate = datime.GetDate(); | |
705f8b0a | 82 | } |
83 | ||
84 | //________________________________________________________________________ | |
85 | AliTRDpidRefMakerNN::AliTRDpidRefMakerNN(const char *name) | |
86 | :AliTRDpidRefMaker(name, "PID(NN) Reference Maker") | |
87 | ,fNet(NULL) | |
88 | ,fTrainMomBin(kAll) | |
89 | ,fEpochs(1000) | |
90 | ,fMinTrain(100) | |
91 | ,fDate(0) | |
92 | ,fDoTraining(0) | |
93 | ,fContinueTraining(0) | |
94 | ,fTrainPath(NULL) | |
95 | ,fScale(0) | |
96 | ,fLy(0) | |
97 | ,fNtrkl(0) | |
98 | ,fRef(NULL) | |
1ee39b3a | 99 | { |
100 | // | |
101 | // Default constructor | |
102 | // | |
103 | ||
9ec5bce7 | 104 | memset(fTrain, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); |
105 | memset(fTest, 0, AliTRDCalPID::kNMom*sizeof(TEventList*)); | |
106 | memset(fTrainData, 0, AliTRDCalPID::kNMom*sizeof(TTree*)); | |
1ee39b3a | 107 | |
108 | SetAbundance(.67); | |
109 | SetScaledEdx(Float_t(AliTRDCalPIDNN::kMLPscale)); | |
110 | TDatime datime; | |
111 | fDate = datime.GetDate(); | |
1ee39b3a | 112 | } |
113 | ||
114 | ||
115 | //________________________________________________________________________ | |
116 | AliTRDpidRefMakerNN::~AliTRDpidRefMakerNN() | |
117 | { | |
118 | } | |
119 | ||
120 | ||
121 | //________________________________________________________________________ | |
705f8b0a | 122 | void AliTRDpidRefMakerNN::MakeTrainTestTrees() |
1ee39b3a | 123 | { |
9ec5bce7 | 124 | // Create output file and tree |
1ee39b3a | 125 | // Called once |
126 | ||
705f8b0a | 127 | fRef = new TFile("TRD.CalibPIDrefMakerNN.root", "RECREATE"); |
9ec5bce7 | 128 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ |
129 | fTrainData[ip] = new TTree(Form("fTrainData_%d", ip), Form("NN Reference Data for MomBin %d", ip)); | |
130 | fTrainData[ip] -> Branch("fdEdx", fdEdx, Form("fdEdx[%d]/F", AliTRDpidUtil::kNNslices)); | |
131 | fTrainData[ip] -> Branch("fPID", fPID, Form("fPID[%d]/F", AliPID::kSPECIES)); | |
132 | fTrainData[ip] -> Branch("fLy", &fLy, "fLy/I"); | |
133 | fTrainData[ip] -> Branch("fNtrkl", &fNtrkl, "fNtrkl/I"); | |
134 | ||
135 | fTrain[ip] = new TEventList(Form("fTrainMom%d", ip), Form("Training list for momentum intervall %d", ip)); | |
136 | fTest[ip] = new TEventList(Form("fTestMom%d", ip), Form("Test list for momentum intervall %d", ip)); | |
137 | } | |
1ee39b3a | 138 | } |
139 | ||
140 | ||
141 | ||
142 | //________________________________________________________________________ | |
143 | Bool_t AliTRDpidRefMakerNN::PostProcess() | |
144 | { | |
145 | // Draw result to the screen | |
146 | // Called once at the end of the query | |
147 | ||
d80a6a00 | 148 | TFile *fCalib = TFile::Open(Form("AnalysisResults.root")); |
9ec5bce7 | 149 | if (!fCalib) { |
150 | AliError("Calibration file not available"); | |
151 | return kFALSE; | |
152 | } | |
d80a6a00 | 153 | TDirectoryFile *dCalib = (TDirectoryFile*)fCalib->Get("TRD.CalibPIDrefMaker"); |
154 | if (!dCalib) { | |
155 | AliError("Calibration directory not available"); | |
156 | return kFALSE; | |
157 | } | |
158 | fData = (TTree*)dCalib->Get("RefPID"); | |
9ec5bce7 | 159 | if (!fData) { |
160 | AliError("Calibration data not available"); | |
161 | return kFALSE; | |
162 | } | |
705f8b0a | 163 | TObjArray *o = NULL; |
d80a6a00 | 164 | if(!(o = (TObjArray*)dCalib->Get("MonitorNN"))) { |
9ec5bce7 | 165 | AliWarning("Missing monitoring container."); |
166 | return kFALSE; | |
167 | } | |
168 | fContainer = (TObjArray*)o->Clone("monitor"); | |
169 | ||
d80a6a00 | 170 | |
9ec5bce7 | 171 | |
172 | if (!fRef) { | |
705f8b0a | 173 | AliDebug(2, "Loading file TRD.CalibPIDrefMakerNN.root"); |
174 | LoadFile("TRD.CalibPIDrefMakerNN.root"); | |
9ec5bce7 | 175 | } |
176 | else AliDebug(2, "file available"); | |
177 | ||
178 | if (!fRef) { | |
d80a6a00 | 179 | MakeTrainingSample(); |
9ec5bce7 | 180 | } |
181 | else AliDebug(2, "file available"); | |
182 | ||
183 | ||
184 | // build the training and the test list for the neural networks | |
185 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
186 | MakeTrainingLists(ip); | |
187 | } | |
1ee39b3a | 188 | if(!fDoTraining) return kTRUE; |
189 | ||
9ec5bce7 | 190 | |
191 | ||
192 | // train the neural networks | |
1ee39b3a | 193 | gSystem->Exec(Form("mkdir ./Networks_%d/",fDate)); |
194 | AliDebug(2, Form("TrainMomBin [%d] [%d]", fTrainMomBin, kAll)); | |
195 | ||
196 | // train single network for a single momentum (recommended) | |
197 | if(!(fTrainMomBin == kAll)){ | |
9ec5bce7 | 198 | if(fTrain[fTrainMomBin] -> GetN() < fMinTrain){ |
199 | AliError("Warning in AliTRDpidRefMakerNN::PostProcess : Not enough events for training available! Please check Data sample!"); | |
1ee39b3a | 200 | return kFALSE; |
201 | } | |
202 | MakeRefs(fTrainMomBin); | |
1ee39b3a | 203 | MonitorTraining(fTrainMomBin); |
204 | } | |
205 | // train all momenta | |
206 | else{ | |
207 | for(Int_t iMomBin = 0; iMomBin < AliTRDCalPID::kNMom; iMomBin++){ | |
9ec5bce7 | 208 | if(fTrain[iMomBin] -> GetN() < fMinTrain){ |
209 | AliError(Form("Warning in AliTRDpidRefMakerNN::PostProcess : Not enough events for training available for momentum bin [%d]! Please check Data sample!", iMomBin)); | |
210 | continue; | |
1ee39b3a | 211 | } |
212 | MakeRefs(fTrainMomBin); | |
1ee39b3a | 213 | MonitorTraining(iMomBin); |
214 | } | |
215 | } | |
216 | ||
217 | return kTRUE; // testing protection | |
218 | } | |
219 | ||
220 | ||
d80a6a00 | 221 | //________________________________________________________________________ |
222 | Bool_t AliTRDpidRefMakerNN::MakeTrainingSample() | |
223 | { | |
224 | ||
225 | TFile *fCalib = TFile::Open(Form("AnalysisResults.root")); | |
226 | if (!fCalib) { | |
227 | AliError("Calibration file not available"); | |
228 | return kFALSE; | |
229 | } | |
230 | TDirectoryFile *dCalib = (TDirectoryFile*)fCalib->Get("TRD.CalibPIDrefMaker"); | |
231 | if (!dCalib) { | |
232 | AliError("Calibration directory not available"); | |
233 | return kFALSE; | |
234 | } | |
235 | fData = (TTree*)dCalib->Get("RefPID"); | |
236 | if (!fData) { | |
237 | AliError("Calibration data not available"); | |
238 | return kFALSE; | |
239 | } | |
240 | TObjArray *o = NULL; | |
241 | if(!(o = (TObjArray*)dCalib->Get("MonitorNN"))) { | |
242 | AliWarning("Missing monitoring container."); | |
243 | return kFALSE; | |
244 | } | |
245 | fContainer = (TObjArray*)o->Clone("monitor"); | |
246 | ||
247 | if (!fRef) { | |
248 | MakeTrainTestTrees(); | |
249 | ||
250 | // Convert the CaliPIDrefMaker tree to 11 (different momentum bin) trees for NN training | |
251 | ||
252 | LinkPIDdata(); | |
253 | for(Int_t ip=0; ip < AliTRDCalPID::kNMom; ip++){ | |
254 | for(Int_t is=0; is < AliPID::kSPECIES; is++) { | |
255 | memset(fPID, 0, AliPID::kSPECIES*sizeof(Float_t)); | |
256 | fPID[is] = 1; | |
257 | Int_t n(0); // index of data | |
258 | for(Int_t itrk = 0; itrk<fData->GetEntries() && n<kMaxStat; itrk++){ | |
259 | if(!(fData->GetEntry(itrk))) continue; | |
260 | if(fPIDdataArray->GetPID()!=is) continue; | |
261 | fNtrkl = fPIDdataArray->GetNtracklets(); | |
262 | for(Int_t ily=fPIDdataArray->GetNtracklets(); ily--;){ | |
263 | fLy = ily; | |
264 | if(fPIDdataArray->GetData(ily)->Momentum()!= ip) continue; | |
265 | memset(fdEdx, 0, AliTRDpidUtil::kNNslices*sizeof(Float_t)); | |
266 | for(Int_t islice=AliTRDCalPID::kNSlicesNN; islice--;){ | |
267 | fdEdx[islice]+=fPIDdataArray->GetData(ily)->fdEdx[islice]; | |
268 | fdEdx[islice]/=fScale; | |
269 | } | |
270 | fTrainData[ip] -> Fill(); | |
271 | n++; | |
272 | } | |
273 | } | |
274 | AliDebug(2, Form("%d %d %d", ip, is, n)); | |
275 | } | |
276 | } | |
277 | ||
278 | ||
279 | fRef -> cd(); | |
280 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
281 | fTrainData[ip] -> Write(); | |
282 | } | |
283 | return kTRUE; | |
284 | ||
285 | } | |
286 | else AliWarning("Training file available. No conversion done!"); | |
287 | return kFALSE; | |
288 | ||
289 | } | |
290 | ||
1ee39b3a | 291 | //________________________________________________________________________ |
9ec5bce7 | 292 | void AliTRDpidRefMakerNN::MakeTrainingLists(Int_t mombin) |
1ee39b3a | 293 | { |
294 | // | |
295 | // build the training lists for the neural networks | |
296 | // | |
297 | ||
9ec5bce7 | 298 | if (!fRef) { |
299 | LoadFile(Form("TRD.Calib%s.root", GetName())); | |
1ee39b3a | 300 | } |
301 | ||
9ec5bce7 | 302 | if (!fRef) { |
303 | AliError("ERROR file for building training list not available"); | |
1ee39b3a | 304 | return; |
305 | } | |
306 | ||
9ec5bce7 | 307 | AliDebug(2, "\n Making training lists! \n"); |
1ee39b3a | 308 | |
9ec5bce7 | 309 | Int_t nPart[AliPID::kSPECIES]; |
310 | memset(nPart, 0, AliPID::kSPECIES*sizeof(Int_t)); | |
1ee39b3a | 311 | |
312 | // set needed branches | |
9ec5bce7 | 313 | fTrainData[mombin] -> SetBranchAddress("fdEdx", fdEdx); |
314 | fTrainData[mombin] -> SetBranchAddress("fPID", fPID); | |
315 | fTrainData[mombin] -> SetBranchAddress("fLy", &fLy); | |
316 | fTrainData[mombin] -> SetBranchAddress("fNtrkl", &fNtrkl); | |
317 | ||
1ee39b3a | 318 | // start first loop to check total number of each particle type |
9ec5bce7 | 319 | for(Int_t iEv=0; iEv < fTrainData[mombin] -> GetEntries(); iEv++){ |
320 | fTrainData[mombin] -> GetEntry(iEv); | |
1ee39b3a | 321 | |
322 | // use only events with goes through 6 layers TRD | |
9ec5bce7 | 323 | if(fNtrkl != AliTRDgeometry::kNlayer) continue; |
324 | ||
325 | if(fPID[AliPID::kElectron] == 1) | |
326 | nPart[AliPID::kElectron]++; | |
327 | else if(fPID[AliPID::kMuon] == 1) | |
328 | nPart[AliPID::kMuon]++; | |
329 | else if(fPID[AliPID::kPion] == 1) | |
330 | nPart[AliPID::kPion]++; | |
331 | else if(fPID[AliPID::kKaon] == 1) | |
332 | nPart[AliPID::kKaon]++; | |
333 | else if(fPID[AliPID::kProton] == 1) | |
334 | nPart[AliPID::kProton]++; | |
1ee39b3a | 335 | } |
336 | ||
337 | AliDebug(2, "Particle multiplicities:"); | |
9ec5bce7 | 338 | 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])); |
339 | ||
340 | ||
1ee39b3a | 341 | |
9ec5bce7 | 342 | // // implement counter of training and test sample size |
343 | Int_t iTrain = 0, iTest = 0; | |
1ee39b3a | 344 | |
345 | // set training sample size per momentum interval to 2/3 | |
346 | // of smallest particle counter and test sample to 1/3 | |
9ec5bce7 | 347 | iTrain = nPart[0]; |
348 | for(Int_t iPart = 1; iPart < AliPID::kSPECIES; iPart++){ | |
349 | // exclude muons and kaons if not availyable | |
350 | // this is neeeded since we do not have v0 candiates | |
351 | if((iPart == AliPID::kMuon || iPart == AliPID::kKaon) && (nPart[AliPID::kMuon] == 0 || nPart[AliPID::kKaon] == 0)) continue; | |
352 | if(iTrain > nPart[iPart]) | |
353 | iTrain = nPart[iPart]; | |
354 | } | |
355 | iTest = Int_t( iTrain * (1-fFreq)); | |
356 | iTrain = Int_t(iTrain * fFreq); | |
357 | AliDebug(2, Form("Momentum[%d] Train[%d] Test[%d]", mombin, iTrain, iTest)); | |
358 | ||
1ee39b3a | 359 | |
360 | ||
361 | // reset couters | |
9ec5bce7 | 362 | memset(nPart, 0, AliPID::kSPECIES*sizeof(Int_t)); |
1ee39b3a | 363 | |
364 | // start second loop to set the event lists | |
9ec5bce7 | 365 | for(Int_t iEv = 0; iEv < fTrainData[mombin] -> GetEntries(); iEv++){ |
366 | fTrainData[mombin] -> GetEntry(iEv); | |
367 | ||
368 | // set event list | |
369 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ | |
370 | if(nPart[is] < iTrain && fPID[is] == 1){ | |
371 | fTrain[mombin] -> Enter(iEv); | |
372 | nPart[is]++; | |
373 | } else if(nPart[is] < iTest+iTrain && fPID[is] == 1){ | |
374 | fTest[mombin] -> Enter(iEv); | |
375 | nPart[is]++; | |
1ee39b3a | 376 | } else continue; |
377 | } | |
378 | } | |
379 | ||
380 | AliDebug(2, "Particle multiplicities in both lists:"); | |
9ec5bce7 | 381 | 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])); |
382 | ||
383 | return; | |
1ee39b3a | 384 | } |
385 | ||
386 | ||
387 | //________________________________________________________________________ | |
388 | void AliTRDpidRefMakerNN::MakeRefs(Int_t mombin) | |
389 | { | |
390 | // | |
391 | // train the neural networks | |
392 | // | |
393 | ||
394 | ||
d80a6a00 | 395 | if (!fTrainData[mombin]) LoadFile(Form("TRD.CalibPIDrefMakerNN.root")); |
1ee39b3a | 396 | |
9ec5bce7 | 397 | if (!fTrainData[mombin]) { |
1ee39b3a | 398 | AliError("Tree for training list not available"); |
399 | return; | |
400 | } | |
401 | ||
402 | TDatime datime; | |
403 | fDate = datime.GetDate(); | |
404 | ||
405 | AliDebug(2, Form("Training momentum bin %d", mombin)); | |
406 | ||
407 | // set variable to monitor the training and to save the development of the networks | |
408 | Int_t nEpochs = fEpochs/kMoniTrain; | |
409 | AliDebug(2, Form("Training %d times %d epochs", kMoniTrain, nEpochs)); | |
410 | ||
411 | // make directories to save the networks | |
412 | gSystem->Exec(Form("rm -r ./Networks_%d/MomBin_%d",fDate, mombin)); | |
413 | gSystem->Exec(Form("mkdir ./Networks_%d/MomBin_%d",fDate, mombin)); | |
414 | ||
415 | // variable to check if network can load weights from previous training | |
9ec5bce7 | 416 | Bool_t bFirstLoop = kTRUE; |
1ee39b3a | 417 | |
418 | // train networks over several loops and save them after each loop | |
419 | for(Int_t iLoop = 0; iLoop < kMoniTrain; iLoop++){ | |
9ec5bce7 | 420 | fTrainData[mombin] -> SetEventList(fTrain[mombin]); |
421 | fTrainData[mombin] -> SetEventList(fTest[mombin]); | |
1ee39b3a | 422 | |
9ec5bce7 | 423 | AliDebug(2, Form("Momentum[%d] Trainingloop[%d]", mombin, iLoop)); |
1ee39b3a | 424 | |
9ec5bce7 | 425 | // check if network is already implemented |
426 | if(bFirstLoop == kTRUE){ | |
427 | 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]); | |
428 | fNet -> SetLearningMethod(TMultiLayerPerceptron::kStochastic); // set learning method | |
429 | fNet -> TMultiLayerPerceptron::SetEta(0.001); // set learning speed | |
430 | if(!fContinueTraining){ | |
431 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph"); | |
432 | else fNet -> Train(nEpochs,""); | |
1ee39b3a | 433 | } |
9ec5bce7 | 434 | else{ |
435 | fNet -> LoadWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fTrainPath, mombin, kMoniTrain - 1)); | |
436 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph+"); | |
437 | else fNet -> Train(nEpochs,"+"); | |
1ee39b3a | 438 | } |
9ec5bce7 | 439 | bFirstLoop = kFALSE; |
440 | } | |
441 | else{ | |
442 | if(AliLog::GetDebugLevel("","AliTRDpidRefMakerNN")>=2) fNet -> Train(nEpochs,"text update=10, graph+"); | |
443 | else fNet -> Train(nEpochs,"+"); | |
444 | } | |
1ee39b3a | 445 | |
9ec5bce7 | 446 | // save weights for monitoring of the training |
447 | fNet -> DumpWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); | |
1ee39b3a | 448 | } // end training loop |
449 | } | |
450 | ||
451 | ||
452 | ||
453 | //________________________________________________________________________ | |
454 | void AliTRDpidRefMakerNN::MonitorTraining(Int_t mombin) | |
455 | { | |
456 | // | |
457 | // train the neural networks | |
458 | // | |
459 | ||
d80a6a00 | 460 | if (!fTrainData[mombin]) LoadFile(Form("TRD.CalibPIDrefMakerNN.root")); |
9ec5bce7 | 461 | if (!fTrainData[mombin]) { |
1ee39b3a | 462 | AliError("Tree for training list not available"); |
463 | return; | |
464 | } | |
465 | ||
466 | // init networks and set event list | |
467 | for(Int_t iChamb = 0; iChamb < AliTRDgeometry::kNlayer; iChamb++){ | |
9ec5bce7 | 468 | 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]); |
469 | fTrainData[mombin] -> SetEventList(fTrain[mombin]); | |
470 | fTrainData[mombin] -> SetEventList(fTest[mombin]); | |
1ee39b3a | 471 | } |
472 | ||
473 | // implement variables for likelihoods | |
474 | Float_t like[AliPID::kSPECIES][AliTRDgeometry::kNlayer]; | |
475 | memset(like, 0, AliPID::kSPECIES*AliTRDgeometry::kNlayer*sizeof(Float_t)); | |
476 | Float_t likeAll[AliPID::kSPECIES], totProb; | |
477 | ||
478 | Double_t pionEffiTrain[kMoniTrain], pionEffiErrTrain[kMoniTrain]; | |
479 | Double_t pionEffiTest[kMoniTrain], pionEffiErrTest[kMoniTrain]; | |
480 | memset(pionEffiTrain, 0, kMoniTrain*sizeof(Double_t)); | |
481 | memset(pionEffiErrTrain, 0, kMoniTrain*sizeof(Double_t)); | |
482 | memset(pionEffiTest, 0, kMoniTrain*sizeof(Double_t)); | |
483 | memset(pionEffiErrTest, 0, kMoniTrain*sizeof(Double_t)); | |
484 | ||
485 | // init histos | |
486 | const Float_t epsilon = 1/(2*(AliTRDpidUtil::kBins-1)); // get nice histos with bin center at 0 and 1 | |
487 | TH1F *hElecs, *hPions; | |
488 | hElecs = new TH1F("hElecs","Likelihood for electrons", AliTRDpidUtil::kBins, 0.-epsilon, 1.+epsilon); | |
489 | hPions = new TH1F("hPions","Likelihood for pions", AliTRDpidUtil::kBins, 0.-epsilon, 1.+epsilon); | |
490 | ||
9ec5bce7 | 491 | TGraphErrors *gEffisTrain = new TGraphErrors(kMoniTrain); |
492 | gEffisTrain -> SetLineColor(4); | |
493 | gEffisTrain -> SetMarkerColor(4); | |
494 | gEffisTrain -> SetMarkerStyle(29); | |
495 | gEffisTrain -> SetMarkerSize(1); | |
496 | ||
497 | TGraphErrors *gEffisTest = new TGraphErrors(kMoniTrain); | |
498 | gEffisTest -> SetLineColor(2); | |
499 | gEffisTest -> SetMarkerColor(2); | |
500 | gEffisTest -> SetMarkerStyle(29); | |
501 | gEffisTest -> SetMarkerSize(1); | |
1ee39b3a | 502 | |
503 | AliTRDpidUtil *util = new AliTRDpidUtil(); | |
504 | ||
505 | // monitor training progress | |
506 | for(Int_t iLoop = 0; iLoop < kMoniTrain; iLoop++){ | |
507 | ||
508 | // load weights | |
9ec5bce7 | 509 | fNet -> LoadWeights(Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); |
510 | AliDebug(2, Form("./Networks_%d/MomBin_%d/Net_%d",fDate, mombin, iLoop)); | |
1ee39b3a | 511 | |
512 | // event loop training list | |
1ee39b3a | 513 | |
9ec5bce7 | 514 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ |
1ee39b3a | 515 | |
9ec5bce7 | 516 | if(!((is == AliPID::kElectron) || (is == AliPID::kPion))) continue; |
1ee39b3a | 517 | |
9ec5bce7 | 518 | Int_t iChamb = 0; |
519 | // reset particle probabilities | |
1ee39b3a | 520 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ |
9ec5bce7 | 521 | likeAll[iPart] = 1./AliPID::kSPECIES; |
1ee39b3a | 522 | } |
9ec5bce7 | 523 | totProb = 0.; |
1ee39b3a | 524 | |
9ec5bce7 | 525 | AliDebug(2, Form("%d",fTrain[mombin] -> GetN())); |
526 | for(Int_t iEvent = 0; iEvent < fTrain[mombin] -> GetN(); iEvent++ ){ | |
527 | fTrainData[mombin] -> GetEntry(fTrain[mombin] -> GetEntry(iEvent)); | |
528 | // use event only if it is electron or pion | |
529 | if(!(fPID[is] == 1.0)) continue; | |
530 | // get the probabilities for each particle type in each chamber | |
531 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
532 | like[iPart][iChamb] = fNet -> Result(fTrain[mombin] -> GetEntry(iEvent), iPart); | |
533 | likeAll[iPart] *= like[iPart][iChamb]; | |
534 | } | |
535 | //end chamber loop | |
536 | iChamb++; | |
537 | ||
538 | // get total probability and normalize it | |
539 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
540 | totProb += likeAll[iPart]; | |
541 | } | |
542 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
543 | likeAll[iPart] /= totProb; | |
544 | } | |
545 | ||
546 | if(iChamb == 5){ | |
547 | // fill likelihood distributions | |
548 | if(fPID[AliPID::kElectron] == 1) | |
549 | hElecs -> Fill(likeAll[AliPID::kElectron]); | |
550 | if(fPID[AliPID::kPion] == 1) | |
551 | hPions -> Fill(likeAll[AliPID::kElectron]); | |
552 | iChamb = 0; | |
553 | // reset particle probabilities | |
554 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
555 | likeAll[iPart] = 1./AliPID::kSPECIES; | |
556 | } | |
557 | totProb = 0.; | |
558 | } | |
559 | } // end event loop | |
560 | } // end species loop | |
561 | ||
1ee39b3a | 562 | |
563 | // calculate the pion efficiency and fill the graph | |
564 | util -> CalculatePionEffi(hElecs, hPions); | |
565 | pionEffiTrain[iLoop] = util -> GetPionEfficiency(); | |
566 | pionEffiErrTrain[iLoop] = util -> GetError(); | |
567 | ||
568 | gEffisTrain -> SetPoint(iLoop, iLoop+1, pionEffiTrain[iLoop]); | |
569 | gEffisTrain -> SetPointError(iLoop, 0, pionEffiErrTrain[iLoop]); | |
570 | hElecs -> Reset(); | |
571 | hPions -> Reset(); | |
572 | AliDebug(2, Form("TrainingLoop[%d] PionEfficiency[%f +/- %f]", iLoop, pionEffiTrain[iLoop], pionEffiErrTrain[iLoop])); | |
573 | // end training loop | |
574 | ||
575 | ||
576 | ||
9ec5bce7 | 577 | // monitor validation progress |
578 | for(Int_t is = 0; is < AliPID::kSPECIES; is++){ | |
1ee39b3a | 579 | |
9ec5bce7 | 580 | if(!((is == AliPID::kElectron) || (is == AliPID::kPion))) continue; |
581 | ||
582 | Int_t iChamb = 0; | |
1ee39b3a | 583 | // reset particle probabilities |
584 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
9ec5bce7 | 585 | likeAll[iPart] = 1./AliPID::kSPECIES; |
1ee39b3a | 586 | } |
587 | totProb = 0.; | |
588 | ||
9ec5bce7 | 589 | for(Int_t iEvent = 0; iEvent < fTest[mombin] -> GetN(); iEvent++ ){ |
590 | fTrainData[mombin] -> GetEntry(fTest[mombin] -> GetEntry(iEvent)); | |
591 | // use event only if it is electron or pion | |
592 | if(!(fPID[is] == 1.0)) continue; | |
593 | ||
594 | // get the probabilities for each particle type in each chamber | |
595 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
596 | like[iPart][iChamb] = fNet -> Result(fTest[mombin] -> GetEntry(iEvent), iPart); | |
597 | likeAll[iPart] *= like[iPart][iChamb]; | |
598 | } | |
599 | iChamb++; | |
600 | ||
601 | // get total probability and normalize it | |
602 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
603 | totProb += likeAll[iPart]; | |
604 | } | |
605 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
606 | likeAll[iPart] /= totProb; | |
607 | } | |
608 | ||
609 | if(iChamb == 5){ | |
610 | // fill likelihood distributions | |
611 | if(fPID[AliPID::kElectron] == 1) | |
612 | hElecs -> Fill(likeAll[AliPID::kElectron]); | |
613 | if(fPID[AliPID::kPion] == 1) | |
614 | hPions -> Fill(likeAll[AliPID::kElectron]); | |
615 | iChamb = 0; | |
616 | // reset particle probabilities | |
617 | for(Int_t iPart = 0; iPart < AliPID::kSPECIES; iPart++){ | |
618 | likeAll[iPart] = 1./AliPID::kSPECIES; | |
619 | } | |
620 | totProb = 0.; | |
621 | } | |
622 | } // end event loop | |
623 | } // end species loop | |
624 | ||
1ee39b3a | 625 | // calculate the pion efficiency and fill the graph |
626 | util -> CalculatePionEffi(hElecs, hPions); | |
627 | pionEffiTest[iLoop] = util -> GetPionEfficiency(); | |
628 | pionEffiErrTest[iLoop] = util -> GetError(); | |
629 | ||
630 | gEffisTest -> SetPoint(iLoop, iLoop+1, pionEffiTest[iLoop]); | |
631 | gEffisTest -> SetPointError(iLoop, 0, pionEffiErrTest[iLoop]); | |
632 | hElecs -> Reset(); | |
633 | hPions -> Reset(); | |
634 | AliDebug(2, Form("TestLoop[%d] PionEfficiency[%f +/- %f] \n", iLoop, pionEffiTest[iLoop], pionEffiErrTest[iLoop])); | |
635 | ||
9ec5bce7 | 636 | } // end validation loop |
1ee39b3a | 637 | |
638 | util -> Delete(); | |
639 | ||
640 | gEffisTest -> Draw("PAL"); | |
641 | gEffisTrain -> Draw("PL"); | |
642 | ||
643 | } | |
644 | ||
645 | ||
646 | //________________________________________________________________________ | |
9ec5bce7 | 647 | Bool_t AliTRDpidRefMakerNN::LoadFile(const Char_t *InFileNN) |
1ee39b3a | 648 | { |
649 | // | |
650 | // Loads the files and sets the event list | |
9ec5bce7 | 651 | // for neural network training. |
1ee39b3a | 652 | // Useable for training outside of the makeResults.C macro |
653 | // | |
654 | ||
9ec5bce7 | 655 | fRef = TFile::Open(Form("%s", InFileNN)); |
656 | if(!fRef) return 0; | |
657 | for(Int_t ip = 0; ip < AliTRDCalPID::kNMom; ip++){ | |
658 | fTrainData[ip] = (TTree*)fRef -> Get(Form("fTrainData_%d", ip)); | |
659 | } | |
1ee39b3a | 660 | |
661 | for(Int_t iMom = 0; iMom < AliTRDCalPID::kNMom; iMom++){ | |
9ec5bce7 | 662 | fTrain[iMom] = new TEventList(Form("fTrainMom%d", iMom), Form("Training list for momentum intervall %d", iMom)); |
663 | fTest[iMom] = new TEventList(Form("fTestMom%d", iMom), Form("Test list for momentum intervall %d", iMom)); | |
1ee39b3a | 664 | } |
9ec5bce7 | 665 | return 1; |
1ee39b3a | 666 | } |
667 | ||
668 |