Logo ROOT   6.16/01
Reference Guide
MethodPDEFoam.cxx
Go to the documentation of this file.
1// @(#)root/tmva $Id$
2// Author: Tancredi Carli, Dominik Dannheim, Alexander Voigt
3
4/**********************************************************************************
5 * Project: TMVA - a Root-integrated toolkit for multivariate Data analysis *
6 * Package: TMVA *
7 * Class : MethodPDEFoam *
8 * Web : http://tmva.sourceforge.net *
9 * *
10 * Description: *
11 * Implementation (see header for description) *
12 * *
13 * Authors (alphabetical): *
14 * Tancredi Carli - CERN, Switzerland *
15 * Dominik Dannheim - CERN, Switzerland *
16 * Alexander Voigt - TU Dresden, Germany *
17 * Peter Speckmayer - CERN, Switzerland *
18 * *
19 * Original author of the TFoam implementation: *
20 * S. Jadach - Institute of Nuclear Physics, Cracow, Poland *
21 * *
22 * Copyright (c) 2008, 2010: *
23 * CERN, Switzerland *
24 * MPI-K Heidelberg, Germany *
25 * *
26 * Redistribution and use in source and binary forms, with or without *
27 * modification, are permitted according to the terms listed in LICENSE *
28 * (http://tmva.sourceforge.net/LICENSE) *
29 **********************************************************************************/
30
31/*! \class TMVA::MethodPDEFoam
32\ingroup TMVA
33
34The PDEFoam method is an extension of the PDERS method, which
35divides the multi-dimensional phase space in a finite number of
36hyper-rectangles (cells) of constant event density. This "foam" of
37cells is filled with averaged probability-density information
38sampled from a training event sample.
39
40For a given number of cells, the binning algorithm adjusts the size
41and position of the cells inside the multidimensional phase space
42based on a binary-split algorithm, minimizing the variance of the
43event density in the cell.
44The binned event density information of the final foam is stored in
45binary trees, allowing for a fast and memory-efficient
46classification of events.
47
48The implementation of PDEFoam is based on the Monte-Carlo
49integration package TFoam included in the analysis package ROOT.
50*/
51
52#include "TMVA/MethodPDEFoam.h"
53
55#include "TMVA/Config.h"
56#include "TMVA/Configurable.h"
57#include "TMVA/CrossEntropy.h"
58#include "TMVA/DataSet.h"
59#include "TMVA/DataSetInfo.h"
60#include "TMVA/Event.h"
61#include "TMVA/GiniIndex.h"
63#include "TMVA/IMethod.h"
65#include "TMVA/MethodBase.h"
66#include "TMVA/MsgLogger.h"
67#include "TMVA/Ranking.h"
68#include "TMVA/SdivSqrtSplusB.h"
69#include "TMVA/SeparationBase.h"
70#include "TMVA/Tools.h"
71#include "TMVA/Types.h"
72#include "TMVA/VariableInfo.h"
73
74#include "TMath.h"
75#include "TH1F.h"
76#include "TFile.h"
77
78REGISTER_METHOD(PDEFoam)
79
81
82////////////////////////////////////////////////////////////////////////////////
83/// init PDEFoam objects
84
86 const TString& methodTitle,
87 DataSetInfo& dsi,
88 const TString& theOption ) :
89 MethodBase( jobName, Types::kPDEFoam, methodTitle, dsi, theOption)
90 , fSigBgSeparated(kFALSE)
91 , fFrac(0.001)
92 , fDiscrErrCut(-1.0)
93 , fVolFrac(1.0/15.0)
94 , fnCells(999)
95 , fnActiveCells(500)
96 , fnSampl(2000)
97 , fnBin(5)
98 , fEvPerBin(10000)
99 , fCompress(kTRUE)
100 , fMultiTargetRegression(kFALSE)
101 , fNmin(100)
102 , fCutNmin(kTRUE)
103 , fMaxDepth(0)
104 , fKernelStr("None")
105 , fKernel(kNone)
106 , fKernelEstimator(NULL)
107 , fTargetSelectionStr("Mean")
108 , fTargetSelection(kMean)
109 , fFillFoamWithOrigWeights(kFALSE)
110 , fUseYesNoCell(kFALSE)
111 , fDTLogic("None")
112 , fDTSeparation(kFoam)
113 , fPeekMax(kTRUE)
114 , fXmin()
115 , fXmax()
116 , fFoam()
117{
118}
119
120////////////////////////////////////////////////////////////////////////////////
121/// constructor from weight file
122
124 const TString& theWeightFile) :
125 MethodBase( Types::kPDEFoam, dsi, theWeightFile)
126 , fSigBgSeparated(kFALSE)
127 , fFrac(0.001)
128 , fDiscrErrCut(-1.0)
129 , fVolFrac(1.0/15.0)
130 , fnCells(999)
131 , fnActiveCells(500)
132 , fnSampl(2000)
133 , fnBin(5)
134 , fEvPerBin(10000)
135 , fCompress(kTRUE)
136 , fMultiTargetRegression(kFALSE)
137 , fNmin(100)
138 , fCutNmin(kTRUE)
139 , fMaxDepth(0)
140 , fKernelStr("None")
141 , fKernel(kNone)
142 , fKernelEstimator(NULL)
143 , fTargetSelectionStr("Mean")
144 , fTargetSelection(kMean)
145 , fFillFoamWithOrigWeights(kFALSE)
146 , fUseYesNoCell(kFALSE)
147 , fDTLogic("None")
148 , fDTSeparation(kFoam)
149 , fPeekMax(kTRUE)
150 , fXmin()
151 , fXmax()
152 , fFoam()
153{
154}
155
156////////////////////////////////////////////////////////////////////////////////
157/// PDEFoam can handle classification with multiple classes and regression
158/// with one or more regression-targets
159
161{
162 if (type == Types::kClassification && numberClasses == 2) return kTRUE;
163 if (type == Types::kMulticlass ) return kTRUE;
164 if (type == Types::kRegression) return kTRUE;
165 return kFALSE;
166}
167
168////////////////////////////////////////////////////////////////////////////////
169/// default initialization called by all constructors
170
172{
173 // init PDEFoam options
174 fSigBgSeparated = kFALSE; // default: unified foam
175 fFrac = 0.001; // fraction of outlier events
176 fDiscrErrCut = -1.; // cut on discriminator error
177 fVolFrac = 1./15.; // range searching box size
178 fnActiveCells = 500; // number of active cells to create
179 fnCells = fnActiveCells*2-1; // total number of cells
180 fnSampl = 2000; // number of sampling points in cell
181 fnBin = 5; // number of bins in edge histogram
182 fEvPerBin = 10000; // number of events per bin
183 fNmin = 100; // minimum number of events in cell
184 fMaxDepth = 0; // cell tree depth (default: unlimited)
185 fFillFoamWithOrigWeights = kFALSE; // fill orig. weights into foam
186 fUseYesNoCell = kFALSE; // return -1 or 1 for bg or signal events
187 fDTLogic = "None"; // decision tree algorithmus
188 fDTSeparation = kFoam; // separation type
189
190 fKernel = kNone; // default: use no kernel
191 fKernelEstimator= NULL; // kernel estimator used during evaluation
192 fTargetSelection= kMean; // default: use mean for target selection (only multi target regression!)
193
194 fCompress = kTRUE; // compress ROOT output file
195 fMultiTargetRegression = kFALSE; // multi-target regression
196
197 DeleteFoams();
198
199 if (fUseYesNoCell)
200 SetSignalReferenceCut( 0.0 ); // MVA output in [-1, 1]
201 else
202 SetSignalReferenceCut( 0.5 ); // MVA output in [0, 1]
203}
204
205////////////////////////////////////////////////////////////////////////////////
206/// Declare MethodPDEFoam options
207
209{
210 DeclareOptionRef( fSigBgSeparated = kFALSE, "SigBgSeparate", "Separate foams for signal and background" );
211 DeclareOptionRef( fFrac = 0.001, "TailCut", "Fraction of outlier events that are excluded from the foam in each dimension" );
212 DeclareOptionRef( fVolFrac = 1./15., "VolFrac", "Size of sampling box, used for density calculation during foam build-up (maximum value: 1.0 is equivalent to volume of entire foam)");
213 DeclareOptionRef( fnActiveCells = 500, "nActiveCells", "Maximum number of active cells to be created by the foam");
214 DeclareOptionRef( fnSampl = 2000, "nSampl", "Number of generated MC events per cell");
215 DeclareOptionRef( fnBin = 5, "nBin", "Number of bins in edge histograms");
216 DeclareOptionRef( fCompress = kTRUE, "Compress", "Compress foam output file");
217 DeclareOptionRef( fMultiTargetRegression = kFALSE, "MultiTargetRegression", "Do regression with multiple targets");
218 DeclareOptionRef( fNmin = 100, "Nmin", "Number of events in cell required to split cell");
219 DeclareOptionRef( fMaxDepth = 0, "MaxDepth", "Maximum depth of cell tree (0=unlimited)");
220 DeclareOptionRef( fFillFoamWithOrigWeights = kFALSE, "FillFoamWithOrigWeights", "Fill foam with original or boost weights");
221 DeclareOptionRef( fUseYesNoCell = kFALSE, "UseYesNoCell", "Return -1 or 1 for bkg or signal like events");
222 DeclareOptionRef( fDTLogic = "None", "DTLogic", "Use decision tree algorithm to split cells");
223 AddPreDefVal(TString("None"));
224 AddPreDefVal(TString("GiniIndex"));
225 AddPreDefVal(TString("MisClassificationError"));
226 AddPreDefVal(TString("CrossEntropy"));
227 AddPreDefVal(TString("GiniIndexWithLaplace"));
228 AddPreDefVal(TString("SdivSqrtSplusB"));
229
230 DeclareOptionRef( fKernelStr = "None", "Kernel", "Kernel type used");
231 AddPreDefVal(TString("None"));
232 AddPreDefVal(TString("Gauss"));
233 AddPreDefVal(TString("LinNeighbors"));
234 DeclareOptionRef( fTargetSelectionStr = "Mean", "TargetSelection", "Target selection method");
235 AddPreDefVal(TString("Mean"));
236 AddPreDefVal(TString("Mpv"));
237}
238
239
240////////////////////////////////////////////////////////////////////////////////
241/// options that are used ONLY for the READER to ensure backward compatibility
242
245 DeclareOptionRef(fCutNmin = kTRUE, "CutNmin", "Requirement for minimal number of events in cell");
246 DeclareOptionRef(fPeekMax = kTRUE, "PeekMax", "Peek cell with max. loss for the next split");
247}
248
249////////////////////////////////////////////////////////////////////////////////
250/// process user options
251
253{
254 if (!(fFrac>=0. && fFrac<=1.)) {
255 Log() << kWARNING << "TailCut not in [0.,1] ==> using 0.001 instead" << Endl;
256 fFrac = 0.001;
257 }
258
259 if (fnActiveCells < 1) {
260 Log() << kWARNING << "invalid number of active cells specified: "
261 << fnActiveCells << "; setting nActiveCells=2" << Endl;
262 fnActiveCells = 2;
263 }
264 fnCells = fnActiveCells*2-1;
265
266 // DT logic is only applicable if a single foam is trained
267 if (fSigBgSeparated && fDTLogic != "None") {
268 Log() << kFATAL << "Decision tree logic works only for a single foam (SigBgSeparate=F)" << Endl;
269 }
270
271 // set separation to use
272 if (fDTLogic == "None")
273 fDTSeparation = kFoam;
274 else if (fDTLogic == "GiniIndex")
275 fDTSeparation = kGiniIndex;
276 else if (fDTLogic == "MisClassificationError")
277 fDTSeparation = kMisClassificationError;
278 else if (fDTLogic == "CrossEntropy")
279 fDTSeparation = kCrossEntropy;
280 else if (fDTLogic == "GiniIndexWithLaplace")
281 fDTSeparation = kGiniIndexWithLaplace;
282 else if (fDTLogic == "SdivSqrtSplusB")
283 fDTSeparation = kSdivSqrtSplusB;
284 else {
285 Log() << kWARNING << "Unknown separation type: " << fDTLogic
286 << ", setting to None" << Endl;
287 fDTLogic = "None";
288 fDTSeparation = kFoam;
289 }
290
291 if (fKernelStr == "None" ) fKernel = kNone;
292 else if (fKernelStr == "Gauss" ) fKernel = kGaus;
293 else if (fKernelStr == "LinNeighbors") fKernel = kLinN;
294
295 if (fTargetSelectionStr == "Mean" ) fTargetSelection = kMean;
296 else fTargetSelection = kMpv;
297 // sanity check: number of targets > 1 and MultiTargetRegression=F
298 // makes no sense --> set MultiTargetRegression=T
299 if (DoRegression() && Data()->GetNTargets() > 1 && !fMultiTargetRegression) {
300 Log() << kWARNING << "Warning: number of targets > 1"
301 << " and MultiTargetRegression=F was set, this makes no sense!"
302 << " --> I'm setting MultiTargetRegression=T" << Endl;
303 fMultiTargetRegression = kTRUE;
304 }
305}
306
307////////////////////////////////////////////////////////////////////////////////
308/// destructor
309
311{
312 DeleteFoams();
313
314 if (fKernelEstimator != NULL)
315 delete fKernelEstimator;
316}
317
318////////////////////////////////////////////////////////////////////////////////
319/// Determine foam range [fXmin, fXmax] for all dimensions, such
320/// that a fraction of 'fFrac' events lie outside the foam.
321
323{
324 fXmin.clear();
325 fXmax.clear();
326 UInt_t kDim = GetNvar(); // == Data()->GetNVariables();
327 UInt_t tDim = Data()->GetNTargets();
328 UInt_t vDim = Data()->GetNVariables();
329 if (fMultiTargetRegression)
330 kDim += tDim;
331
332 Float_t *xmin = new Float_t[kDim];
333 Float_t *xmax = new Float_t[kDim];
334
335 // set default values
336 for (UInt_t dim=0; dim<kDim; dim++) {
337 xmin[dim] = FLT_MAX;
338 xmax[dim] = FLT_MIN;
339 }
340
341 Log() << kDEBUG << "Number of training events: " << Data()->GetNTrainingEvents() << Endl;
342 Int_t nevoutside = (Int_t)((Data()->GetNTrainingEvents())*(fFrac)); // number of events that are outside the range
343 Int_t rangehistbins = 10000; // number of bins in histos
344
345 // loop over all testing signal and BG events and clac minimal and
346 // maximal value of every variable
347 for (Long64_t i=0; i<(GetNEvents()); i++) { // events loop
348 const Event* ev = GetEvent(i);
349 for (UInt_t dim=0; dim<kDim; dim++) { // variables loop
350 Float_t val;
351 if (fMultiTargetRegression) {
352 if (dim < vDim)
353 val = ev->GetValue(dim);
354 else
355 val = ev->GetTarget(dim-vDim);
356 }
357 else
358 val = ev->GetValue(dim);
359
360 if (val<xmin[dim])
361 xmin[dim] = val;
362 if (val>xmax[dim])
363 xmax[dim] = val;
364 }
365 }
366
367 // Create and fill histograms for each dimension (with same events
368 // as before), to determine range based on number of events outside
369 // the range
370 TH1F **range_h = new TH1F*[kDim];
371 for (UInt_t dim=0; dim<kDim; dim++) {
372 range_h[dim] = new TH1F(Form("range%i", dim), "range", rangehistbins, xmin[dim], xmax[dim]);
373 }
374
375 // fill all testing events into histos
376 for (Long64_t i=0; i<GetNEvents(); i++) {
377 const Event* ev = GetEvent(i);
378 for (UInt_t dim=0; dim<kDim; dim++) {
379 if (fMultiTargetRegression) {
380 if (dim < vDim)
381 range_h[dim]->Fill(ev->GetValue(dim));
382 else
383 range_h[dim]->Fill(ev->GetTarget(dim-vDim));
384 }
385 else
386 range_h[dim]->Fill(ev->GetValue(dim));
387 }
388 }
389
390 // calc Xmin, Xmax from Histos
391 for (UInt_t dim=0; dim<kDim; dim++) {
392 for (Int_t i=1; i<(rangehistbins+1); i++) { // loop over bins
393 if (range_h[dim]->Integral(0, i) > nevoutside) { // calc left limit (integral over bins 0..i = nevoutside)
394 xmin[dim]=range_h[dim]->GetBinLowEdge(i);
395 break;
396 }
397 }
398 for (Int_t i=rangehistbins; i>0; i--) { // calc right limit (integral over bins i..max = nevoutside)
399 if (range_h[dim]->Integral(i, (rangehistbins+1)) > nevoutside) {
400 xmax[dim]=range_h[dim]->GetBinLowEdge(i+1);
401 break;
402 }
403 }
404 }
405 // now xmin[] and xmax[] contain upper/lower limits for every dimension
406
407 // copy xmin[], xmax[] values to the class variable
408 fXmin.clear();
409 fXmax.clear();
410 for (UInt_t dim=0; dim<kDim; dim++) {
411 fXmin.push_back(xmin[dim]);
412 fXmax.push_back(xmax[dim]);
413 }
414
415
416 delete[] xmin;
417 delete[] xmax;
418
419 // delete histos
420 for (UInt_t dim=0; dim<kDim; dim++)
421 delete range_h[dim];
422 delete[] range_h;
423
424 return;
425}
426
427////////////////////////////////////////////////////////////////////////////////
428/// Train PDE-Foam depending on the set options
429
431{
432 Log() << kVERBOSE << "Calculate Xmin and Xmax for every dimension" << Endl;
433 CalcXminXmax();
434
435 // delete foams
436 DeleteFoams();
437
438 // start training
439 if (DoRegression()) {
440 if (fMultiTargetRegression)
441 TrainMultiTargetRegression();
442 else
443 TrainMonoTargetRegression();
444 }
445 else {
446 if (DoMulticlass())
447 TrainMultiClassification();
448 else {
449 if (DataInfo().GetNormalization() != "EQUALNUMEVENTS" ) {
450 Log() << kHEADER << "NormMode=" << DataInfo().GetNormalization()
451 << " chosen. Note that only NormMode=EqualNumEvents"
452 << " ensures that Discriminant values correspond to"
453 << " signal probabilities." << Endl;
454 }
455
456 Log() << kDEBUG << "N_sig for training events: " << Data()->GetNEvtSigTrain() << Endl;
457 Log() << kDEBUG << "N_bg for training events: " << Data()->GetNEvtBkgdTrain() << Endl;
458 Log() << kDEBUG << "User normalization: " << DataInfo().GetNormalization().Data() << Endl;
459
460 if (fSigBgSeparated)
461 TrainSeparatedClassification();
462 else
463 TrainUnifiedClassification();
464 }
465 }
466
467 // delete the binary search tree in order to save memory
468 for(UInt_t i=0; i<fFoam.size(); i++) {
469 if(fFoam.at(i))
470 fFoam.at(i)->DeleteBinarySearchTree();
471 }
472 ExitFromTraining();
473}
474
475////////////////////////////////////////////////////////////////////////////////
476/// Creation of 2 separated foams: one for signal events, one for
477/// background events. At the end the foam cells of fFoam[0] will
478/// contain the average number of signal events and fFoam[1] will
479/// contain the average number of background events.
480
482{
483 TString foamcaption[2];
484 foamcaption[0] = "SignalFoam";
485 foamcaption[1] = "BgFoam";
486
487 for(int i=0; i<2; i++) {
488 // create 2 PDEFoams
489 fFoam.push_back( InitFoam(foamcaption[i], kSeparate) );
490
491 Log() << kVERBOSE << "Filling binary search tree of " << foamcaption[i]
492 << " with events" << Endl;
493 // insert event to BinarySearchTree
494 for (Long64_t k=0; k<GetNEvents(); ++k) {
495 const Event* ev = GetEvent(k);
496 if ((i==0 && DataInfo().IsSignal(ev)) || (i==1 && !DataInfo().IsSignal(ev)))
497 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
498 fFoam.back()->FillBinarySearchTree(ev);
499 }
500
501 Log() << kINFO << "Build up " << foamcaption[i] << Endl;
502 fFoam.back()->Create(); // build foam
503
504 Log() << kVERBOSE << "Filling foam cells with events" << Endl;
505 // loop over all events -> fill foam cells
506 for (Long64_t k=0; k<GetNEvents(); ++k) {
507 const Event* ev = GetEvent(k);
508 Float_t weight = fFillFoamWithOrigWeights ? ev->GetOriginalWeight() : ev->GetWeight();
509 if ((i==0 && DataInfo().IsSignal(ev)) || (i==1 && !DataInfo().IsSignal(ev)))
510 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
511 fFoam.back()->FillFoamCells(ev, weight);
512 }
513 }
514}
515
516////////////////////////////////////////////////////////////////////////////////
517/// Create only one unified foam (fFoam[0]) whose cells contain the
518/// average discriminator (N_sig)/(N_sig + N_bg)
519
521{
522 fFoam.push_back( InitFoam("DiscrFoam", kDiscr, fSignalClass) );
523
524 Log() << kVERBOSE << "Filling binary search tree of discriminator foam with events" << Endl;
525 // insert event to BinarySearchTree
526 for (Long64_t k=0; k<GetNEvents(); ++k) {
527 const Event* ev = GetEvent(k);
528 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
529 fFoam.back()->FillBinarySearchTree(ev);
530 }
531
532 Log() << kINFO << "Build up discriminator foam" << Endl;
533 fFoam.back()->Create(); // build foam
534
535 Log() << kVERBOSE << "Filling foam cells with events" << Endl;
536 // loop over all training events -> fill foam cells with N_sig and N_Bg
537 for (Long64_t k=0; k<GetNEvents(); ++k) {
538 const Event* ev = GetEvent(k);
539 Float_t weight = fFillFoamWithOrigWeights ? ev->GetOriginalWeight() : ev->GetWeight();
540 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
541 fFoam.back()->FillFoamCells(ev, weight);
542 }
543
544 Log() << kVERBOSE << "Calculate cell discriminator"<< Endl;
545 // calc discriminator (and it's error) for each cell
546 fFoam.back()->Finalize();
547}
548
549////////////////////////////////////////////////////////////////////////////////
550/// Create one unified foam (see TrainUnifiedClassification()) for
551/// each class, where the cells of foam i (fFoam[i]) contain the
552/// average fraction of events of class i, i.e.
553///
554/// D = number events of class i / total number of events
555
557{
558 for (UInt_t iClass=0; iClass<DataInfo().GetNClasses(); ++iClass) {
559
560 fFoam.push_back( InitFoam(Form("MultiClassFoam%u",iClass), kMultiClass, iClass) );
561
562 Log() << kVERBOSE << "Filling binary search tree of multiclass foam "
563 << iClass << " with events" << Endl;
564 // insert event to BinarySearchTree
565 for (Long64_t k=0; k<GetNEvents(); ++k) {
566 const Event* ev = GetEvent(k);
567 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
568 fFoam.back()->FillBinarySearchTree(ev);
569 }
570
571 Log() << kINFO << "Build up multiclass foam " << iClass << Endl;
572 fFoam.back()->Create(); // build foam
573
574 Log() << kVERBOSE << "Filling foam cells with events" << Endl;
575 // loop over all training events and fill foam cells with signal
576 // and background events
577 for (Long64_t k=0; k<GetNEvents(); ++k) {
578 const Event* ev = GetEvent(k);
579 Float_t weight = fFillFoamWithOrigWeights ? ev->GetOriginalWeight() : ev->GetWeight();
580 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
581 fFoam.back()->FillFoamCells(ev, weight);
582 }
583
584 Log() << kVERBOSE << "Calculate cell discriminator"<< Endl;
585 // calc discriminator (and it's error) for each cell
586 fFoam.back()->Finalize();
587 }
588}
589
590////////////////////////////////////////////////////////////////////////////////
591/// Training one (mono target regression) foam, whose cells contain
592/// the average 0th target. The dimension of the foam = number of
593/// non-targets (= number of variables).
594
596{
597 if (Data()->GetNTargets() != 1) {
598 Log() << kFATAL << "Can't do mono-target regression with "
599 << Data()->GetNTargets() << " targets!" << Endl;
600 }
601
602 Log() << kDEBUG << "MethodPDEFoam: number of Targets: " << Data()->GetNTargets() << Endl;
603
604 fFoam.push_back( InitFoam("MonoTargetRegressionFoam", kMonoTarget) );
605
606 Log() << kVERBOSE << "Filling binary search tree with events" << Endl;
607 // insert event to BinarySearchTree
608 for (Long64_t k=0; k<GetNEvents(); ++k) {
609 const Event* ev = GetEvent(k);
610 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
611 fFoam.back()->FillBinarySearchTree(ev);
612 }
613
614 Log() << kINFO << "Build mono target regression foam" << Endl;
615 fFoam.back()->Create(); // build foam
616
617 Log() << kVERBOSE << "Filling foam cells with events" << Endl;
618 // loop over all events -> fill foam cells with target
619 for (Long64_t k=0; k<GetNEvents(); ++k) {
620 const Event* ev = GetEvent(k);
621 Float_t weight = fFillFoamWithOrigWeights ? ev->GetOriginalWeight() : ev->GetWeight();
622 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
623 fFoam.back()->FillFoamCells(ev, weight);
624 }
625
626 Log() << kVERBOSE << "Calculate average cell targets"<< Endl;
627 // calc weight (and it's error) for each cell
628 fFoam.back()->Finalize();
629}
630
631////////////////////////////////////////////////////////////////////////////////
632/// Training one (multi target regression) foam, whose cells contain
633/// the average event density. The dimension of the foam = number
634/// of non-targets + number of targets.
635
637{
638 Log() << kDEBUG << "Number of variables: " << Data()->GetNVariables() << Endl;
639 Log() << kDEBUG << "Number of Targets: " << Data()->GetNTargets() << Endl;
640 Log() << kDEBUG << "Dimension of foam: " << Data()->GetNVariables()+Data()->GetNTargets() << Endl;
641 if (fKernel==kLinN)
642 Log() << kFATAL << "LinNeighbors kernel currently not supported"
643 << " for multi target regression" << Endl;
644
645 fFoam.push_back( InitFoam("MultiTargetRegressionFoam", kMultiTarget) );
646
647 Log() << kVERBOSE << "Filling binary search tree of multi target regression foam with events"
648 << Endl;
649 // insert event to BinarySearchTree
650 for (Long64_t k=0; k<GetNEvents(); ++k) {
651 Event *ev = new Event(*GetEvent(k));
652 // since in multi-target regression targets are handled like
653 // variables --> remove targets and add them to the event variabels
654 std::vector<Float_t> targets(ev->GetTargets());
655 const UInt_t nVariables = ev->GetValues().size();
656 for (UInt_t i = 0; i < targets.size(); ++i)
657 ev->SetVal(i+nVariables, targets.at(i));
658 ev->GetTargets().clear();
659 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
660 fFoam.back()->FillBinarySearchTree(ev);
661 // since the binary search tree copies the event, one can delete
662 // it
663 delete ev;
664 }
665
666 Log() << kINFO << "Build multi target regression foam" << Endl;
667 fFoam.back()->Create(); // build foam
668
669 Log() << kVERBOSE << "Filling foam cells with events" << Endl;
670 // loop over all events -> fill foam cells with number of events
671 for (Long64_t k=0; k<GetNEvents(); ++k) {
672 Event *ev = new Event(*GetEvent(k));
673 // since in multi-target regression targets are handled like
674 // variables --> remove targets and add them to the event variabels
675 std::vector<Float_t> targets = ev->GetTargets();
676 const UInt_t nVariables = ev->GetValues().size();
677 Float_t weight = fFillFoamWithOrigWeights ? ev->GetOriginalWeight() : ev->GetWeight();
678 for (UInt_t i = 0; i < targets.size(); ++i)
679 ev->SetVal(i+nVariables, targets.at(i));
680 ev->GetTargets().clear();
681 if (!(IgnoreEventsWithNegWeightsInTraining() && ev->GetWeight()<=0))
682 fFoam.back()->FillFoamCells(ev, weight);
683 // since the PDEFoam copies the event, one can delete it
684 delete ev;
685 }
686}
687
688////////////////////////////////////////////////////////////////////////////////
689/// Return Mva-Value.
690///
691/// In case of `fSigBgSeparated==false` (one unified PDEFoam was
692/// trained) the function returns the content of the cell, which
693/// corresponds to the current TMVA::Event, i.e. D =
694/// N_sig/(N_bg+N_sig).
695///
696/// In case of `fSigBgSeparated==true` (two separate PDEFoams were
697/// trained) the function returns
698///
699/// D = Density_sig/(Density_sig+Density_bg)
700///
701/// where 'Density_sig' is the content of the cell in the signal
702/// PDEFoam (fFoam[0]) and 'Density_bg' is the content of the cell
703/// in the background PDEFoam (fFoam[1]).
704///
705/// In both cases the error on the discriminant is stored in 'err'
706/// and 'errUpper'. (Of course err and errUpper must be non-zero
707/// and point to valid address to make this work.)
708
710{
711 const Event* ev = GetEvent();
712 Double_t discr = 0.;
713
714 if (fSigBgSeparated) {
715 std::vector<Float_t> xvec = ev->GetValues();
716
717 Double_t density_sig = 0.; // calc signal event density
718 Double_t density_bg = 0.; // calc background event density
719 density_sig = fFoam.at(0)->GetCellValue(xvec, kValueDensity, fKernelEstimator);
720 density_bg = fFoam.at(1)->GetCellValue(xvec, kValueDensity, fKernelEstimator);
721
722 // calc discriminator (normed!)
723 if ( (density_sig+density_bg) > 0 )
724 discr = density_sig/(density_sig+density_bg);
725 else
726 discr = 0.5; // assume 50% signal probability, if no events found (bad assumption, but can be overruled by cut on error)
727 }
728 else { // Signal and Bg not separated
729 // get discriminator direct from the foam
730 discr = fFoam.at(0)->GetCellValue(ev->GetValues(), kValue, fKernelEstimator);
731 }
732
733 // calculate the error
734 if (err || errUpper) {
735 const Double_t discr_error = CalculateMVAError();
736 if (err != 0) *err = discr_error;
737 if (errUpper != 0) *errUpper = discr_error;
738 }
739
740 if (fUseYesNoCell)
741 return (discr < 0.5 ? -1 : 1);
742 else
743 return discr;
744}
745
746////////////////////////////////////////////////////////////////////////////////
747/// Calculate the error on the Mva value
748///
749/// If `fSigBgSeparated == true` the error is calculated from the
750/// number of events in the signal and background PDEFoam cells.
751///
752/// If `fSigBgSeparated == false`, the error is taken directly from
753/// the PDEFoam cell.
754
756{
757 const Event* ev = GetEvent(); // current event
758 Double_t mvaError = 0.0; // the error on the Mva value
759
760 if (fSigBgSeparated) {
761 const std::vector<Float_t>& xvec = ev->GetValues();
762
763 const Double_t neventsB = fFoam.at(1)->GetCellValue(xvec, kValue, fKernelEstimator);
764 const Double_t neventsS = fFoam.at(0)->GetCellValue(xvec, kValue, fKernelEstimator);
765 const Double_t scaleB = 1.;
766 // estimation of statistical error on counted signal/background events
767 const Double_t errorS = neventsS == 0 ? 1.0 : TMath::Sqrt(neventsS);
768 const Double_t errorB = neventsB == 0 ? 1.0 : TMath::Sqrt(neventsB);
769
770 if ((neventsS > 1e-10) || (neventsB > 1e-10)) {
771 // eq. (5) in paper T.Carli, B.Koblitz 2002
772 mvaError = TMath::Sqrt(Sqr(scaleB * neventsB / Sqr(neventsS + scaleB * neventsB) * errorS) +
773 Sqr(scaleB * neventsS / Sqr(neventsS + scaleB * neventsB) * errorB));
774 } else {
775 mvaError = 1.0;
776 }
777 } else { // Signal and Bg not separated
778 // get discriminator error direct from the foam
779 mvaError = fFoam.at(0)->GetCellValue(ev->GetValues(), kValueError, fKernelEstimator);
780 }
781
782 return mvaError;
783}
784
785////////////////////////////////////////////////////////////////////////////////
786/// Get the multiclass MVA response for the PDEFoam classifier. The
787/// returned MVA values are normalized, i.e. their sum equals 1.
788
789const std::vector<Float_t>& TMVA::MethodPDEFoam::GetMulticlassValues()
790{
791 const TMVA::Event *ev = GetEvent();
792 std::vector<Float_t> xvec = ev->GetValues();
793
794 if (fMulticlassReturnVal == NULL)
795 fMulticlassReturnVal = new std::vector<Float_t>();
796 fMulticlassReturnVal->clear();
797 fMulticlassReturnVal->reserve(DataInfo().GetNClasses());
798
799 std::vector<Float_t> temp; // temp class. values
800 UInt_t nClasses = DataInfo().GetNClasses();
801 temp.reserve(nClasses);
802 for (UInt_t iClass = 0; iClass < nClasses; ++iClass) {
803 temp.push_back(fFoam.at(iClass)->GetCellValue(xvec, kValue, fKernelEstimator));
804 }
805
806 for (UInt_t iClass = 0; iClass < nClasses; ++iClass) {
807 Float_t norm = 0.0; // normalization
808 for (UInt_t j = 0; j < nClasses; ++j) {
809 if (iClass != j)
810 norm += exp(temp[j] - temp[iClass]);
811 }
812 fMulticlassReturnVal->push_back(1.0 / (1.0 + norm));
813 }
814
815 return *fMulticlassReturnVal;
816}
817
818////////////////////////////////////////////////////////////////////////////////
819/// Compute ranking of input variables from the number of cuts made
820/// in each PDEFoam dimension. The PDEFoam dimension (the variable)
821/// for which the most cuts were done is ranked highest.
822
824{
825 // create the ranking object
826 fRanking = new Ranking(GetName(), "Variable Importance");
827 std::vector<Float_t> importance(GetNvar(), 0);
828
829 // determine variable importances
830 for (UInt_t ifoam = 0; ifoam < fFoam.size(); ++ifoam) {
831 // get the number of cuts made in every dimension of foam
832 PDEFoamCell *root_cell = fFoam.at(ifoam)->GetRootCell();
833 std::vector<UInt_t> nCuts(fFoam.at(ifoam)->GetTotDim(), 0);
834 GetNCuts(root_cell, nCuts);
835
836 // fill the importance vector (ignoring the target dimensions in
837 // case of a multi-target regression foam)
838 UInt_t sumOfCuts = 0;
839 std::vector<Float_t> tmp_importance;
840 for (UInt_t ivar = 0; ivar < GetNvar(); ++ivar) {
841 sumOfCuts += nCuts.at(ivar);
842 tmp_importance.push_back( nCuts.at(ivar) );
843 }
844 // normalization of the variable importances of this foam: the
845 // sum of all variable importances equals 1 for this foam
846 for (UInt_t ivar = 0; ivar < GetNvar(); ++ivar) {
847 if (sumOfCuts > 0)
848 tmp_importance.at(ivar) /= sumOfCuts;
849 else
850 tmp_importance.at(ivar) = 0;
851 }
852 // the overall variable importance is the average over all foams
853 for (UInt_t ivar = 0; ivar < GetNvar(); ++ivar) {
854 importance.at(ivar) += tmp_importance.at(ivar) / fFoam.size();
855 }
856 }
857
858 // fill ranking vector
859 for (UInt_t ivar = 0; ivar < GetNvar(); ++ivar) {
860 fRanking->AddRank(Rank(GetInputLabel(ivar), importance.at(ivar)));
861 }
862
863 return fRanking;
864}
865
866////////////////////////////////////////////////////////////////////////////////
867/// Fill in 'nCuts' the number of cuts made in every foam dimension,
868/// starting at the root cell 'cell'.
869///
870/// Parameters:
871///
872/// - cell - root cell to start the counting from
873///
874/// - nCuts - the number of cuts are saved in this vector
875
876void TMVA::MethodPDEFoam::GetNCuts(PDEFoamCell *cell, std::vector<UInt_t> &nCuts)
877{
878 if (cell == NULL || cell->GetStat() == 1) // cell is active
879 return;
880
881 nCuts.at(cell->GetBest())++;
882
883 if (cell->GetDau0() != NULL)
884 GetNCuts(cell->GetDau0(), nCuts);
885 if (cell->GetDau1() != NULL)
886 GetNCuts(cell->GetDau1(), nCuts);
887}
888
889////////////////////////////////////////////////////////////////////////////////
890/// Set Xmin, Xmax for every dimension in the given pdefoam object
891
893{
894 if (!pdefoam){
895 Log() << kFATAL << "Null pointer given!" << Endl;
896 return;
897 }
898
899 UInt_t num_vars = GetNvar();
900 if (fMultiTargetRegression)
901 num_vars += Data()->GetNTargets();
902
903 for (UInt_t idim=0; idim<num_vars; idim++) { // set upper/ lower limit in foam
904 Log()<< kDEBUG << "foam: SetXmin[dim="<<idim<<"]: " << fXmin.at(idim) << Endl;
905 Log()<< kDEBUG << "foam: SetXmax[dim="<<idim<<"]: " << fXmax.at(idim) << Endl;
906 pdefoam->SetXmin(idim, fXmin.at(idim));
907 pdefoam->SetXmax(idim, fXmax.at(idim));
908 }
909}
910
911////////////////////////////////////////////////////////////////////////////////
912/// Create a new PDEFoam, set the PDEFoam options (nCells, nBin,
913/// Xmin, Xmax, etc.) and initialize the PDEFoam by calling
914/// pdefoam->Initialize().
915///
916/// Parameters:
917///
918/// - foamcaption - name of PDEFoam object
919///
920/// - ft - type of PDEFoam
921///
922/// Candidates are:
923/// - kSeparate - creates TMVA::PDEFoamEvent
924/// - kDiscr - creates TMVA::PDEFoamDiscriminant
925/// - kMonoTarget - creates TMVA::PDEFoamTarget
926/// - kMultiTarget - creates TMVA::MultiTarget
927/// - kMultiClass - creates TMVA::PDEFoamDiscriminant
928///
929/// If 'fDTSeparation != kFoam' then a TMVA::PDEFoamDecisionTree
930/// is created (the separation type depends on fDTSeparation).
931///
932/// - cls - marked event class (optional, default value = 0)
933
935{
936 // number of foam dimensions
937 Int_t dim = 1;
938 if (ft == kMultiTarget)
939 // dimension of foam = number of targets + non-targets
940 dim = Data()->GetNTargets() + Data()->GetNVariables();
941 else
942 dim = GetNvar();
943
944 // calculate range-searching box
945 std::vector<Double_t> box;
946 for (Int_t idim = 0; idim < dim; ++idim) {
947 box.push_back((fXmax.at(idim) - fXmin.at(idim))* fVolFrac);
948 }
949
950 // create PDEFoam and PDEFoamDensityBase
951 PDEFoam *pdefoam = NULL;
952 PDEFoamDensityBase *density = NULL;
953 if (fDTSeparation == kFoam) {
954 // use PDEFoam algorithm
955 switch (ft) {
956 case kSeparate:
957 pdefoam = new PDEFoamEvent(foamcaption);
958 density = new PDEFoamEventDensity(box);
959 break;
960 case kMultiTarget:
961 pdefoam = new PDEFoamMultiTarget(foamcaption, fTargetSelection);
962 density = new PDEFoamEventDensity(box);
963 break;
964 case kDiscr:
965 case kMultiClass:
966 pdefoam = new PDEFoamDiscriminant(foamcaption, cls);
967 density = new PDEFoamDiscriminantDensity(box, cls);
968 break;
969 case kMonoTarget:
970 pdefoam = new PDEFoamTarget(foamcaption, 0);
971 density = new PDEFoamTargetDensity(box, 0);
972 break;
973 default:
974 Log() << kFATAL << "Unknown PDEFoam type!" << Endl;
975 break;
976 }
977 } else {
978 // create a decision tree like PDEFoam
979
980 // create separation type class, which is owned by
981 // PDEFoamDecisionTree (i.e. PDEFoamDecisionTree will delete it)
982 SeparationBase *sepType = NULL;
983 switch (fDTSeparation) {
984 case kGiniIndex:
985 sepType = new GiniIndex();
986 break;
987 case kMisClassificationError:
988 sepType = new MisClassificationError();
989 break;
990 case kCrossEntropy:
991 sepType = new CrossEntropy();
992 break;
993 case kGiniIndexWithLaplace:
994 sepType = new GiniIndexWithLaplace();
995 break;
996 case kSdivSqrtSplusB:
997 sepType = new SdivSqrtSplusB();
998 break;
999 default:
1000 Log() << kFATAL << "Separation type " << fDTSeparation
1001 << " currently not supported" << Endl;
1002 break;
1003 }
1004 switch (ft) {
1005 case kDiscr:
1006 case kMultiClass:
1007 pdefoam = new PDEFoamDecisionTree(foamcaption, sepType, cls);
1008 density = new PDEFoamDecisionTreeDensity(box, cls);
1009 break;
1010 default:
1011 Log() << kFATAL << "Decision tree cell split algorithm is only"
1012 << " available for (multi) classification with a single"
1013 << " PDE-Foam (SigBgSeparate=F)" << Endl;
1014 break;
1015 }
1016 }
1017
1018 if (pdefoam) pdefoam->SetDensity(density);
1019 else Log() << kFATAL << "PDEFoam pointer not set, exiting.." << Endl;
1020
1021 // create pdefoam kernel
1022 fKernelEstimator = CreatePDEFoamKernel();
1023
1024 // set fLogger attributes
1025 pdefoam->Log().SetMinType(this->Log().GetMinType());
1026
1027 // set PDEFoam parameters
1028 pdefoam->SetDim( dim);
1029 pdefoam->SetnCells( fnCells); // optional
1030 pdefoam->SetnSampl( fnSampl); // optional
1031 pdefoam->SetnBin( fnBin); // optional
1032 pdefoam->SetEvPerBin( fEvPerBin); // optional
1033
1034 // cuts
1035 pdefoam->SetNmin(fNmin);
1036 pdefoam->SetMaxDepth(fMaxDepth); // maximum cell tree depth
1037
1038 // Init PDEFoam
1039 pdefoam->Initialize();
1040
1041 // Set Xmin, Xmax
1042 SetXminXmax(pdefoam);
1043
1044 return pdefoam;
1045}
1046
1047////////////////////////////////////////////////////////////////////////////////
1048/// Return regression values for both multi- and mono-target regression
1049
1050const std::vector<Float_t>& TMVA::MethodPDEFoam::GetRegressionValues()
1051{
1052 if (fRegressionReturnVal == 0) fRegressionReturnVal = new std::vector<Float_t>();
1053 fRegressionReturnVal->clear();
1054 fRegressionReturnVal->reserve(Data()->GetNTargets());
1055
1056 const Event* ev = GetEvent();
1057 std::vector<Float_t> vals = ev->GetValues(); // get array of event variables (non-targets)
1058
1059 if (vals.empty()) {
1060 Log() << kWARNING << "<GetRegressionValues> value vector is empty. " << Endl;
1061 }
1062
1063 if (fMultiTargetRegression) {
1064 // create std::map from event variables
1065 std::map<Int_t, Float_t> xvec;
1066 for (UInt_t i=0; i<vals.size(); ++i)
1067 xvec.insert(std::pair<Int_t, Float_t>(i, vals.at(i)));
1068 // get the targets
1069 std::vector<Float_t> targets = fFoam.at(0)->GetCellValue( xvec, kValue );
1070
1071 // sanity check
1072 if (targets.size() != Data()->GetNTargets())
1073 Log() << kFATAL << "Something wrong with multi-target regression foam: "
1074 << "number of targets does not match the DataSet()" << Endl;
1075 for(UInt_t i=0; i<targets.size(); i++)
1076 fRegressionReturnVal->push_back(targets.at(i));
1077 }
1078 else {
1079 fRegressionReturnVal->push_back(fFoam.at(0)->GetCellValue(vals, kValue, fKernelEstimator));
1080 }
1081
1082 // apply inverse transformation to regression values
1083 Event * evT = new Event(*ev);
1084 for (UInt_t itgt = 0; itgt < Data()->GetNTargets(); itgt++) {
1085 evT->SetTarget(itgt, fRegressionReturnVal->at(itgt) );
1086 }
1087 const Event* evT2 = GetTransformationHandler().InverseTransform( evT );
1088 fRegressionReturnVal->clear();
1089 for (UInt_t itgt = 0; itgt < Data()->GetNTargets(); itgt++) {
1090 fRegressionReturnVal->push_back( evT2->GetTarget(itgt) );
1091 }
1092
1093 delete evT;
1094
1095 return (*fRegressionReturnVal);
1096}
1097
1098////////////////////////////////////////////////////////////////////////////////
1099/// create a pdefoam kernel estimator, depending on the current
1100/// value of fKernel
1101
1103{
1104 switch (fKernel) {
1105 case kNone:
1106 return new PDEFoamKernelTrivial();
1107 case kLinN:
1108 return new PDEFoamKernelLinN();
1109 case kGaus:
1110 return new PDEFoamKernelGauss(fVolFrac/2.0);
1111 default:
1112 Log() << kFATAL << "Kernel: " << fKernel << " not supported!" << Endl;
1113 return NULL;
1114 }
1115 return NULL;
1116}
1117
1118////////////////////////////////////////////////////////////////////////////////
1119/// Deletes all trained foams
1120
1122{
1123 for (UInt_t i=0; i<fFoam.size(); i++)
1124 if (fFoam.at(i)) delete fFoam.at(i);
1125 fFoam.clear();
1126}
1127
1128////////////////////////////////////////////////////////////////////////////////
1129/// reset MethodPDEFoam:
1130///
1131/// - delete all PDEFoams
1132/// - delete the kernel estimator
1133
1135{
1136 DeleteFoams();
1137
1138 if (fKernelEstimator != NULL) {
1139 delete fKernelEstimator;
1140 fKernelEstimator = NULL;
1141 }
1142}
1143
1144////////////////////////////////////////////////////////////////////////////////
1145
1147{}
1148
1149////////////////////////////////////////////////////////////////////////////////
1150/// create XML output of PDEFoam method variables
1151
1153{
1154 void* wght = gTools().AddChild(parent, "Weights");
1155 gTools().AddAttr( wght, "SigBgSeparated", fSigBgSeparated );
1156 gTools().AddAttr( wght, "Frac", fFrac );
1157 gTools().AddAttr( wght, "DiscrErrCut", fDiscrErrCut );
1158 gTools().AddAttr( wght, "VolFrac", fVolFrac );
1159 gTools().AddAttr( wght, "nCells", fnCells );
1160 gTools().AddAttr( wght, "nSampl", fnSampl );
1161 gTools().AddAttr( wght, "nBin", fnBin );
1162 gTools().AddAttr( wght, "EvPerBin", fEvPerBin );
1163 gTools().AddAttr( wght, "Compress", fCompress );
1164 gTools().AddAttr( wght, "DoRegression", DoRegression() );
1165 gTools().AddAttr( wght, "CutNmin", fNmin>0 );
1166 gTools().AddAttr( wght, "Nmin", fNmin );
1167 gTools().AddAttr( wght, "CutRMSmin", false );
1168 gTools().AddAttr( wght, "RMSmin", 0.0 );
1169 gTools().AddAttr( wght, "Kernel", KernelToUInt(fKernel) );
1170 gTools().AddAttr( wght, "TargetSelection", TargetSelectionToUInt(fTargetSelection) );
1171 gTools().AddAttr( wght, "FillFoamWithOrigWeights", fFillFoamWithOrigWeights );
1172 gTools().AddAttr( wght, "UseYesNoCell", fUseYesNoCell );
1173
1174 // save foam borders Xmin[i], Xmax[i]
1175 void *xmin_wrap;
1176 for (UInt_t i=0; i<fXmin.size(); i++){
1177 xmin_wrap = gTools().AddChild( wght, "Xmin" );
1178 gTools().AddAttr( xmin_wrap, "Index", i );
1179 gTools().AddAttr( xmin_wrap, "Value", fXmin.at(i) );
1180 }
1181 void *xmax_wrap;
1182 for (UInt_t i=0; i<fXmax.size(); i++){
1183 xmax_wrap = gTools().AddChild( wght, "Xmax" );
1184 gTools().AddAttr( xmax_wrap, "Index", i );
1185 gTools().AddAttr( xmax_wrap, "Value", fXmax.at(i) );
1186 }
1187
1188 // write foams to xml file
1189 WriteFoamsToFile();
1190}
1191
1192////////////////////////////////////////////////////////////////////////////////
1193/// Write PDEFoams to file
1194
1196{
1197 // fill variable names into foam
1198 FillVariableNamesToFoam();
1199
1200 TString rfname( GetWeightFileName() );
1201
1202 // replace in case of txt weight file
1203 rfname.ReplaceAll( TString(".") + gConfig().GetIONames().fWeightFileExtension + ".txt", ".xml" );
1204
1205 // add foam indicator to distinguish from main weight file
1206 rfname.ReplaceAll( ".xml", "_foams.root" );
1207
1208 TFile *rootFile = 0;
1209 if (fCompress) rootFile = new TFile(rfname, "RECREATE", "foamfile", 9);
1210 else rootFile = new TFile(rfname, "RECREATE");
1211
1212 // write the foams
1213 for (UInt_t i=0; i<fFoam.size(); ++i) {
1214 Log() << "writing foam " << fFoam.at(i)->GetFoamName().Data()
1215 << " to file" << Endl;
1216 fFoam.at(i)->Write(fFoam.at(i)->GetFoamName().Data());
1217 }
1218
1219 rootFile->Close();
1220 Log() << kINFO << "Foams written to file: "
1221 << gTools().Color("lightblue") << rfname << gTools().Color("reset") << Endl;
1222}
1223
1224////////////////////////////////////////////////////////////////////////////////
1225/// read options and internal parameters
1226
1228{
1229 istr >> fSigBgSeparated; // Separate Sig and Bg, or not
1230 istr >> fFrac; // Fraction used for calc of Xmin, Xmax
1231 istr >> fDiscrErrCut; // cut on discriminant error
1232 istr >> fVolFrac; // volume fraction (used for density calculation during buildup)
1233 istr >> fnCells; // Number of Cells (500)
1234 istr >> fnSampl; // Number of MC events per cell in build-up (1000)
1235 istr >> fnBin; // Number of bins in build-up (100)
1236 istr >> fEvPerBin; // Maximum events (equiv.) per bin in build-up (1000)
1237 istr >> fCompress; // compress output file
1238
1239 Bool_t regr;
1240 istr >> regr; // regression foam
1241 SetAnalysisType( (regr ? Types::kRegression : Types::kClassification ) );
1242
1243 Bool_t CutNmin, CutRMSmin; // dummy for backwards compatible.
1244 Float_t RMSmin; // dummy for backwards compatible.
1245 istr >> CutNmin; // cut on minimal number of events in cell
1246 istr >> fNmin;
1247 istr >> CutRMSmin; // cut on minimal RMS in cell
1248 istr >> RMSmin;
1249
1250 UInt_t ker = 0;
1251 istr >> ker; // used kernel for GetMvaValue()
1252 fKernel = UIntToKernel(ker);
1253
1254 UInt_t ts = 0;
1255 istr >> ts; // used method for target selection
1256 fTargetSelection = UIntToTargetSelection(ts);
1257
1258 istr >> fFillFoamWithOrigWeights; // fill foam with original event weights
1259 istr >> fUseYesNoCell; // return -1 or 1 for bg or signal event
1260
1261 // clear old range and prepare new range
1262 fXmin.clear();
1263 fXmax.clear();
1264 UInt_t kDim = GetNvar();
1265 if (fMultiTargetRegression)
1266 kDim += Data()->GetNTargets();
1267 fXmin.assign(kDim, 0);
1268 fXmax.assign(kDim, 0);
1269
1270 // read range
1271 for (UInt_t i=0; i<kDim; i++)
1272 istr >> fXmin.at(i);
1273 for (UInt_t i=0; i<kDim; i++)
1274 istr >> fXmax.at(i);
1275
1276 // read pure foams from file
1277 ReadFoamsFromFile();
1278}
1279
1280////////////////////////////////////////////////////////////////////////////////
1281/// read PDEFoam variables from xml weight file
1282
1284{
1285 gTools().ReadAttr( wghtnode, "SigBgSeparated", fSigBgSeparated );
1286 gTools().ReadAttr( wghtnode, "Frac", fFrac );
1287 gTools().ReadAttr( wghtnode, "DiscrErrCut", fDiscrErrCut );
1288 gTools().ReadAttr( wghtnode, "VolFrac", fVolFrac );
1289 gTools().ReadAttr( wghtnode, "nCells", fnCells );
1290 gTools().ReadAttr( wghtnode, "nSampl", fnSampl );
1291 gTools().ReadAttr( wghtnode, "nBin", fnBin );
1292 gTools().ReadAttr( wghtnode, "EvPerBin", fEvPerBin );
1293 gTools().ReadAttr( wghtnode, "Compress", fCompress );
1294 Bool_t regr; // dummy for backwards compatible.
1295 gTools().ReadAttr( wghtnode, "DoRegression", regr );
1296 Bool_t CutNmin; // dummy for backwards compatible.
1297 gTools().ReadAttr( wghtnode, "CutNmin", CutNmin );
1298 gTools().ReadAttr( wghtnode, "Nmin", fNmin );
1299 Bool_t CutRMSmin; // dummy for backwards compatible.
1300 Float_t RMSmin; // dummy for backwards compatible.
1301 gTools().ReadAttr( wghtnode, "CutRMSmin", CutRMSmin );
1302 gTools().ReadAttr( wghtnode, "RMSmin", RMSmin );
1303 UInt_t ker = 0;
1304 gTools().ReadAttr( wghtnode, "Kernel", ker );
1305 fKernel = UIntToKernel(ker);
1306 UInt_t ts = 0;
1307 gTools().ReadAttr( wghtnode, "TargetSelection", ts );
1308 fTargetSelection = UIntToTargetSelection(ts);
1309 if (gTools().HasAttr(wghtnode, "FillFoamWithOrigWeights"))
1310 gTools().ReadAttr( wghtnode, "FillFoamWithOrigWeights", fFillFoamWithOrigWeights );
1311 if (gTools().HasAttr(wghtnode, "UseYesNoCell"))
1312 gTools().ReadAttr( wghtnode, "UseYesNoCell", fUseYesNoCell );
1313
1314 // clear old range [Xmin, Xmax] and prepare new range for reading
1315 fXmin.clear();
1316 fXmax.clear();
1317 UInt_t kDim = GetNvar();
1318 if (fMultiTargetRegression)
1319 kDim += Data()->GetNTargets();
1320 fXmin.assign(kDim, 0);
1321 fXmax.assign(kDim, 0);
1322
1323 // read foam range
1324 void *xmin_wrap = gTools().GetChild( wghtnode );
1325 for (UInt_t counter=0; counter<kDim; counter++) {
1326 UInt_t i=0;
1327 gTools().ReadAttr( xmin_wrap , "Index", i );
1328 if (i>=kDim)
1329 Log() << kFATAL << "dimension index out of range:" << i << Endl;
1330 gTools().ReadAttr( xmin_wrap , "Value", fXmin.at(i) );
1331 xmin_wrap = gTools().GetNextChild( xmin_wrap );
1332 }
1333
1334 void *xmax_wrap = xmin_wrap;
1335 for (UInt_t counter=0; counter<kDim; counter++) {
1336 UInt_t i=0;
1337 gTools().ReadAttr( xmax_wrap , "Index", i );
1338 if (i>=kDim)
1339 Log() << kFATAL << "dimension index out of range:" << i << Endl;
1340 gTools().ReadAttr( xmax_wrap , "Value", fXmax.at(i) );
1341 xmax_wrap = gTools().GetNextChild( xmax_wrap );
1342 }
1343
1344 // if foams exist, delete them
1345 DeleteFoams();
1346
1347 // read pure foams from file
1348 ReadFoamsFromFile();
1349
1350 // recreate the pdefoam kernel estimator
1351 if (fKernelEstimator != NULL)
1352 delete fKernelEstimator;
1353 fKernelEstimator = CreatePDEFoamKernel();
1354}
1355
1356////////////////////////////////////////////////////////////////////////////////
1357/// Reads a foam with name 'foamname' from file, and returns a clone
1358/// of the foam. The given ROOT file must be open. (The ROOT file
1359/// will not be closed in this function.)
1360///
1361/// Parameters:
1362///
1363/// - file - an open ROOT file
1364///
1365/// - foamname - name of foam to load from the file
1366///
1367/// Returns:
1368///
1369/// If a foam with name 'foamname' exists in the file, then it is
1370/// read from the file, cloned and returned. If a foam with name
1371/// 'foamname' does not exist in the file or the clone operation
1372/// does not succeed, then NULL is returned.
1373
1375{
1376 if (file == NULL) {
1377 Log() << kWARNING << "<ReadClonedFoamFromFile>: NULL pointer given" << Endl;
1378 return NULL;
1379 }
1380
1381 // try to load the foam from the file
1382 PDEFoam *foam = (PDEFoam*) file->Get(foamname);
1383 if (foam == NULL) {
1384 return NULL;
1385 }
1386 // try to clone the foam
1387 foam = (PDEFoam*) foam->Clone();
1388 if (foam == NULL) {
1389 Log() << kWARNING << "<ReadClonedFoamFromFile>: " << foamname
1390 << " could not be cloned!" << Endl;
1391 return NULL;
1392 }
1393
1394 return foam;
1395}
1396
1397////////////////////////////////////////////////////////////////////////////////
1398/// read foams from file
1399
1401{
1402 TString rfname( GetWeightFileName() );
1403
1404 // replace in case of txt weight file
1405 rfname.ReplaceAll( TString(".") + gConfig().GetIONames().fWeightFileExtension + ".txt", ".xml" );
1406
1407 // add foam indicator to distinguish from main weight file
1408 rfname.ReplaceAll( ".xml", "_foams.root" );
1409
1410 Log() << kINFO << "Read foams from file: " << gTools().Color("lightblue")
1411 << rfname << gTools().Color("reset") << Endl;
1412 TFile *rootFile = new TFile( rfname, "READ" );
1413 if (rootFile->IsZombie()) Log() << kFATAL << "Cannot open file \"" << rfname << "\"" << Endl;
1414
1415 // read foams from file
1416 if (DoRegression()) {
1417 if (fMultiTargetRegression)
1418 fFoam.push_back(ReadClonedFoamFromFile(rootFile, "MultiTargetRegressionFoam"));
1419 else
1420 fFoam.push_back(ReadClonedFoamFromFile(rootFile, "MonoTargetRegressionFoam"));
1421 } else {
1422 if (fSigBgSeparated) {
1423 fFoam.push_back(ReadClonedFoamFromFile(rootFile, "SignalFoam"));
1424 fFoam.push_back(ReadClonedFoamFromFile(rootFile, "BgFoam"));
1425 } else {
1426 // try to load discriminator foam
1427 PDEFoam *foam = ReadClonedFoamFromFile(rootFile, "DiscrFoam");
1428 if (foam != NULL)
1429 fFoam.push_back(foam);
1430 else {
1431 // load multiclass foams
1432 for (UInt_t iClass=0; iClass<DataInfo().GetNClasses(); ++iClass) {
1433 fFoam.push_back(ReadClonedFoamFromFile(rootFile, Form("MultiClassFoam%u",iClass)));
1434 }
1435 }
1436 }
1437 }
1438
1439 // Close the root file. Note, that the foams are still present in
1440 // memory!
1441 rootFile->Close();
1442 delete rootFile;
1443
1444 for (UInt_t i=0; i<fFoam.size(); ++i) {
1445 if (!fFoam.at(0))
1446 Log() << kFATAL << "Could not load foam!" << Endl;
1447 }
1448}
1449
1450////////////////////////////////////////////////////////////////////////////////
1451/// convert UInt_t to EKernel (used for reading weight files)
1452
1454{
1455 switch(iker) {
1456 case 0: return kNone;
1457 case 1: return kGaus;
1458 case 2: return kLinN;
1459 default:
1460 Log() << kWARNING << "<UIntToKernel>: unknown kernel number: " << iker << Endl;
1461 return kNone;
1462 }
1463 return kNone;
1464}
1465
1466////////////////////////////////////////////////////////////////////////////////
1467/// convert UInt_t to ETargetSelection (used for reading weight files)
1468
1470{
1471 switch(its) {
1472 case 0: return kMean;
1473 case 1: return kMpv;
1474 default:
1475 Log() << kWARNING << "<UIntToTargetSelection>: unknown method TargetSelection: " << its << Endl;
1476 return kMean;
1477 }
1478 return kMean;
1479}
1480
1481////////////////////////////////////////////////////////////////////////////////
1482/// store the variable names in all foams
1483
1485{
1486 for (UInt_t ifoam=0; ifoam<fFoam.size(); ifoam++) {
1487 for (Int_t idim=0; idim<fFoam.at(ifoam)->GetTotDim(); idim++) {
1488 if(fMultiTargetRegression && (UInt_t)idim>=DataInfo().GetNVariables())
1489 fFoam.at(ifoam)->AddVariableName(DataInfo().GetTargetInfo(idim-DataInfo().GetNVariables()).GetExpression().Data());
1490 else
1491 fFoam.at(ifoam)->AddVariableName(DataInfo().GetVariableInfo(idim).GetExpression().Data());
1492 }
1493 }
1494}
1495
1496////////////////////////////////////////////////////////////////////////////////
1497/// write PDEFoam-specific classifier response
1498/// NOT IMPLEMENTED YET!
1499
1500void TMVA::MethodPDEFoam::MakeClassSpecific( std::ostream& /*fout*/, const TString& /*className*/ ) const
1501{
1502}
1503
1504////////////////////////////////////////////////////////////////////////////////
1505/// provide help message
1506
1508{
1509 Log() << Endl;
1510 Log() << gTools().Color("bold") << "--- Short description:" << gTools().Color("reset") << Endl;
1511 Log() << Endl;
1512 Log() << "PDE-Foam is a variation of the PDE-RS method using a self-adapting" << Endl;
1513 Log() << "binning method to divide the multi-dimensional variable space into a" << Endl;
1514 Log() << "finite number of hyper-rectangles (cells). The binning algorithm " << Endl;
1515 Log() << "adjusts the size and position of a predefined number of cells such" << Endl;
1516 Log() << "that the variance of the signal and background densities inside the " << Endl;
1517 Log() << "cells reaches a minimum" << Endl;
1518 Log() << Endl;
1519 Log() << gTools().Color("bold") << "--- Use of booking options:" << gTools().Color("reset") << Endl;
1520 Log() << Endl;
1521 Log() << "The PDEFoam classifier supports two different algorithms: " << Endl;
1522 Log() << Endl;
1523 Log() << " (1) Create one foam, which stores the signal over background" << Endl;
1524 Log() << " probability density. During foam buildup the variance of the" << Endl;
1525 Log() << " discriminant inside the cells is minimised." << Endl;
1526 Log() << Endl;
1527 Log() << " Booking option: SigBgSeparated=F" << Endl;
1528 Log() << Endl;
1529 Log() << " (2) Create two separate foams, one for the signal events and one for" << Endl;
1530 Log() << " background events. During foam buildup the variance of the" << Endl;
1531 Log() << " event density inside the cells is minimised separately for" << Endl;
1532 Log() << " signal and background." << Endl;
1533 Log() << Endl;
1534 Log() << " Booking option: SigBgSeparated=T" << Endl;
1535 Log() << Endl;
1536 Log() << "The following options can be set (the listed values are found to be a" << Endl;
1537 Log() << "good starting point for most applications):" << Endl;
1538 Log() << Endl;
1539 Log() << " SigBgSeparate False Separate Signal and Background" << Endl;
1540 Log() << " TailCut 0.001 Fraction of outlier events that excluded" << Endl;
1541 Log() << " from the foam in each dimension " << Endl;
1542 Log() << " VolFrac 0.0666 Volume fraction (used for density calculation" << Endl;
1543 Log() << " during foam build-up) " << Endl;
1544 Log() << " nActiveCells 500 Maximal number of active cells in final foam " << Endl;
1545 Log() << " nSampl 2000 Number of MC events per cell in foam build-up " << Endl;
1546 Log() << " nBin 5 Number of bins used in foam build-up " << Endl;
1547 Log() << " Nmin 100 Number of events in cell required to split cell" << Endl;
1548 Log() << " Kernel None Kernel type used (possible values are: None," << Endl;
1549 Log() << " Gauss)" << Endl;
1550 Log() << " Compress True Compress foam output file " << Endl;
1551 Log() << Endl;
1552 Log() << " Additional regression options:" << Endl;
1553 Log() << Endl;
1554 Log() << "MultiTargetRegression False Do regression with multiple targets " << Endl;
1555 Log() << " TargetSelection Mean Target selection method (possible values are: " << Endl;
1556 Log() << " Mean, Mpv)" << Endl;
1557 Log() << Endl;
1558 Log() << gTools().Color("bold") << "--- Performance optimisation:" << gTools().Color("reset") << Endl;
1559 Log() << Endl;
1560 Log() << "The performance of the two implementations was found to be similar for" << Endl;
1561 Log() << "most examples studied. For the same number of cells per foam, the two-" << Endl;
1562 Log() << "foam option approximately doubles the amount of computer memory needed" << Endl;
1563 Log() << "during classification. For special cases where the event-density" << Endl;
1564 Log() << "distribution of signal and background events is very different, the" << Endl;
1565 Log() << "two-foam option was found to perform significantly better than the" << Endl;
1566 Log() << "option with only one foam." << Endl;
1567 Log() << Endl;
1568 Log() << "In order to gain better classification performance we recommend to set" << Endl;
1569 Log() << "the parameter \"nActiveCells\" to a high value." << Endl;
1570 Log() << Endl;
1571 Log() << "The parameter \"VolFrac\" specifies the size of the sampling volume" << Endl;
1572 Log() << "during foam buildup and should be tuned in order to achieve optimal" << Endl;
1573 Log() << "performance. A larger box leads to a reduced statistical uncertainty" << Endl;
1574 Log() << "for small training samples and to smoother sampling. A smaller box on" << Endl;
1575 Log() << "the other hand increases the sensitivity to statistical fluctuations" << Endl;
1576 Log() << "in the training samples, but for sufficiently large training samples" << Endl;
1577 Log() << "it will result in a more precise local estimate of the sampled" << Endl;
1578 Log() << "density. In general, higher dimensional problems require larger box" << Endl;
1579 Log() << "sizes, due to the reduced average number of events per box volume. The" << Endl;
1580 Log() << "default value of 0.0666 was optimised for an example with 5" << Endl;
1581 Log() << "observables and training samples of the order of 50000 signal and" << Endl;
1582 Log() << "background events each." << Endl;
1583 Log() << Endl;
1584 Log() << "Furthermore kernel weighting can be activated, which will lead to an" << Endl;
1585 Log() << "additional performance improvement. Note that Gauss weighting will" << Endl;
1586 Log() << "significantly increase the response time of the method. LinNeighbors" << Endl;
1587 Log() << "weighting performs a linear interpolation with direct neighbor cells" << Endl;
1588 Log() << "for each dimension and is much faster than Gauss weighting." << Endl;
1589 Log() << Endl;
1590 Log() << "The classification results were found to be rather insensitive to the" << Endl;
1591 Log() << "values of the parameters \"nSamples\" and \"nBin\"." << Endl;
1592}
#define REGISTER_METHOD(CLASS)
for example
const Handle_t kNone
Definition: GuiTypes.h:87
#define e(i)
Definition: RSha256.hxx:103
int Int_t
Definition: RtypesCore.h:41
unsigned int UInt_t
Definition: RtypesCore.h:42
const Bool_t kFALSE
Definition: RtypesCore.h:88
bool Bool_t
Definition: RtypesCore.h:59
double Double_t
Definition: RtypesCore.h:55
long long Long64_t
Definition: RtypesCore.h:69
float Float_t
Definition: RtypesCore.h:53
const Bool_t kTRUE
Definition: RtypesCore.h:87
#define ClassImp(name)
Definition: Rtypes.h:363
int type
Definition: TGX11.cxx:120
float xmin
Definition: THbookFile.cxx:93
float xmax
Definition: THbookFile.cxx:93
double exp(double)
char * Form(const char *fmt,...)
A ROOT file is a suite of consecutive data records (TKey instances) with a well defined format.
Definition: TFile.h:48
virtual void Close(Option_t *option="")
Close a file.
Definition: TFile.cxx:912
1-D histogram with a float per channel (see TH1 documentation)}
Definition: TH1.h:571
virtual Int_t Fill(Double_t x)
Increment bin with abscissa X by 1.
Definition: TH1.cxx:3251
virtual Double_t GetBinLowEdge(Int_t bin) const
Return bin lower edge for 1D histogram.
Definition: TH1.cxx:8473
Implementation of the CrossEntropy as separation criterion.
Definition: CrossEntropy.h:43
Class that contains all the data information.
Definition: DataSetInfo.h:60
Float_t GetValue(UInt_t ivar) const
return value of i'th variable
Definition: Event.cxx:237
void SetTarget(UInt_t itgt, Float_t value)
set the target value (dimension itgt) to value
Definition: Event.cxx:360
std::vector< Float_t > & GetTargets()
Definition: Event.h:98
Double_t GetOriginalWeight() const
Definition: Event.h:79
Double_t GetWeight() const
return the event weight - depending on whether the flag IgnoreNegWeightsInTraining is or not.
Definition: Event.cxx:382
void SetVal(UInt_t ivar, Float_t val)
set variable ivar to val
Definition: Event.cxx:341
std::vector< Float_t > & GetValues()
Definition: Event.h:89
Float_t GetTarget(UInt_t itgt) const
Definition: Event.h:97
Implementation of the GiniIndex With Laplace correction as separation criterion.
Implementation of the GiniIndex as separation criterion.
Definition: GiniIndex.h:63
Virtual base Class for all MVA method.
Definition: MethodBase.h:109
virtual void DeclareCompatibilityOptions()
options that are used ONLY for the READER to ensure backward compatibility they are hence without any...
Definition: MethodBase.cxx:601
The PDEFoam method is an extension of the PDERS method, which divides the multi-dimensional phase spa...
Definition: MethodPDEFoam.h:67
const Ranking * CreateRanking()
Compute ranking of input variables from the number of cuts made in each PDEFoam dimension.
void Init(void)
default initialization called by all constructors
virtual Bool_t HasAnalysisType(Types::EAnalysisType type, UInt_t numberClasses, UInt_t numberTargets)
PDEFoam can handle classification with multiple classes and regression with one or more regression-ta...
void Train(void)
Train PDE-Foam depending on the set options.
const std::vector< Float_t > & GetMulticlassValues()
Get the multiclass MVA response for the PDEFoam classifier.
Double_t CalculateMVAError()
Calculate the error on the Mva value.
void PrintCoefficients(void)
void TrainMultiClassification()
Create one unified foam (see TrainUnifiedClassification()) for each class, where the cells of foam i ...
void TrainMultiTargetRegression(void)
Training one (multi target regression) foam, whose cells contain the average event density.
void ReadWeightsFromXML(void *wghtnode)
read PDEFoam variables from xml weight file
void DeleteFoams()
Deletes all trained foams.
void ReadWeightsFromStream(std::istream &i)
read options and internal parameters
virtual ~MethodPDEFoam(void)
destructor
void DeclareOptions()
Declare MethodPDEFoam options.
PDEFoam * InitFoam(TString, EFoamType, UInt_t cls=0)
Create a new PDEFoam, set the PDEFoam options (nCells, nBin, Xmin, Xmax, etc.) and initialize the PDE...
virtual const std::vector< Float_t > & GetRegressionValues()
Return regression values for both multi- and mono-target regression.
void FillVariableNamesToFoam() const
store the variable names in all foams
void TrainMonoTargetRegression(void)
Training one (mono target regression) foam, whose cells contain the average 0th target.
void TrainUnifiedClassification(void)
Create only one unified foam (fFoam[0]) whose cells contain the average discriminator (N_sig)/(N_sig ...
void ReadFoamsFromFile()
read foams from file
EKernel UIntToKernel(UInt_t iker)
convert UInt_t to EKernel (used for reading weight files)
PDEFoamKernelBase * CreatePDEFoamKernel()
create a pdefoam kernel estimator, depending on the current value of fKernel
void CalcXminXmax()
Determine foam range [fXmin, fXmax] for all dimensions, such that a fraction of 'fFrac' events lie ou...
void MakeClassSpecific(std::ostream &, const TString &) const
write PDEFoam-specific classifier response NOT IMPLEMENTED YET!
void GetNCuts(PDEFoamCell *cell, std::vector< UInt_t > &nCuts)
Fill in 'nCuts' the number of cuts made in every foam dimension, starting at the root cell 'cell'.
PDEFoam * ReadClonedFoamFromFile(TFile *, const TString &)
Reads a foam with name 'foamname' from file, and returns a clone of the foam.
MethodPDEFoam(const TString &jobName, const TString &methodTitle, DataSetInfo &dsi, const TString &theOption="PDEFoam")
init PDEFoam objects
Double_t GetMvaValue(Double_t *err=0, Double_t *errUpper=0)
Return Mva-Value.
ETargetSelection UIntToTargetSelection(UInt_t its)
convert UInt_t to ETargetSelection (used for reading weight files)
void DeclareCompatibilityOptions()
options that are used ONLY for the READER to ensure backward compatibility
void GetHelpMessage() const
provide help message
void TrainSeparatedClassification(void)
Creation of 2 separated foams: one for signal events, one for background events.
void SetXminXmax(TMVA::PDEFoam *)
Set Xmin, Xmax for every dimension in the given pdefoam object.
virtual void Reset()
reset MethodPDEFoam:
void WriteFoamsToFile() const
Write PDEFoams to file.
void AddWeightsXMLTo(void *parent) const
create XML output of PDEFoam method variables
void ProcessOptions()
process user options
Implementation of the MisClassificationError as separation criterion.
void SetMinType(EMsgType minType)
Definition: MsgLogger.h:72
Int_t GetStat() const
Definition: PDEFoamCell.h:91
PDEFoamCell * GetDau1() const
Definition: PDEFoamCell.h:95
PDEFoamCell * GetDau0() const
Definition: PDEFoamCell.h:94
Int_t GetBest() const
Definition: PDEFoamCell.h:78
This is a concrete implementation of PDEFoam.
This PDEFoam variant acts like a decision tree and stores in every cell the discriminant.
This is an abstract class, which provides an interface for a PDEFoam density estimator.
This is a concrete implementation of PDEFoam.
This PDEFoam variant stores in every cell the discriminant.
This is a concrete implementation of PDEFoam.
This PDEFoam variant stores in every cell the sum of event weights and the sum of the squared event w...
Definition: PDEFoamEvent.h:39
This class is the abstract kernel interface for PDEFoam.
This PDEFoam kernel estimates a cell value for a given event by weighting all cell values with a gaus...
This PDEFoam kernel estimates a cell value for a given event by weighting with cell values of the nea...
This class is a trivial PDEFoam kernel estimator.
This PDEFoam variant is used to estimate multiple targets by creating an event density foam (PDEFoamE...
This is a concrete implementation of PDEFoam.
This PDEFoam variant stores in every cell the average target fTarget (see the Constructor) as well as...
Definition: PDEFoamTarget.h:39
Implementation of PDEFoam.
Definition: PDEFoam.h:77
void SetNmin(UInt_t val)
Definition: PDEFoam.h:203
void SetMaxDepth(UInt_t maxdepth)
Definition: PDEFoam.h:205
void SetDensity(PDEFoamDensityBase *dens)
Definition: PDEFoam.h:192
MsgLogger & Log() const
Definition: PDEFoam.h:238
void SetXmax(Int_t idim, Double_t wmax)
set upper foam bound in dimension idim
Definition: PDEFoam.cxx:277
void SetEvPerBin(Int_t EvPerBin)
Definition: PDEFoam.h:190
void SetnBin(Int_t nBin)
Definition: PDEFoam.h:189
void Initialize()
Definition: PDEFoam.h:171
void SetXmin(Int_t idim, Double_t wmin)
set lower foam bound in dimension idim
Definition: PDEFoam.cxx:266
void SetnCells(Long_t nCells)
Definition: PDEFoam.h:187
void SetnSampl(Long_t nSampl)
Definition: PDEFoam.h:188
void SetDim(Int_t kDim)
Sets dimension of cubical space.
Definition: PDEFoam.cxx:251
Ranking for variables in method (implementation)
Definition: Ranking.h:48
Implementation of the SdivSqrtSplusB as separation criterion.
An interface to calculate the "SeparationGain" for different separation criteria used in various trai...
void * GetNextChild(void *prevchild, const char *childname=0)
XML helpers.
Definition: Tools.cxx:1174
void * AddChild(void *parent, const char *childname, const char *content=0, bool isRootNode=false)
add child node
Definition: Tools.cxx:1136
const TString & Color(const TString &)
human readable color strings
Definition: Tools.cxx:840
void * GetChild(void *parent, const char *childname=0)
get child node
Definition: Tools.cxx:1162
void ReadAttr(void *node, const char *, T &value)
read attribute from xml
Definition: Tools.h:337
void AddAttr(void *node, const char *, const T &value, Int_t precision=16)
add attribute to xml
Definition: Tools.h:355
Singleton class for Global types used by TMVA.
Definition: Types.h:73
EAnalysisType
Definition: Types.h:127
@ kMulticlass
Definition: Types.h:130
@ kClassification
Definition: Types.h:128
@ kRegression
Definition: Types.h:129
virtual Int_t Write(const char *name=0, Int_t option=0, Int_t bufsize=0)
Write this object to the current directory.
Definition: TObject.cxx:785
virtual TObject * Clone(const char *newname="") const
Make a clone of an object using the Streamer facility.
Definition: TObject.cxx:144
R__ALWAYS_INLINE Bool_t IsZombie() const
Definition: TObject.h:134
Basic string class.
Definition: TString.h:131
TString & ReplaceAll(const TString &s1, const TString &s2)
Definition: TString.h:687
void box(Int_t pat, Double_t x1, Double_t y1, Double_t x2, Double_t y2)
Definition: fillpatterns.C:1
std::string GetName(const std::string &scope_name)
Definition: Cppyy.cxx:146
Config & gConfig()
Tools & gTools()
MsgLogger & Endl(MsgLogger &ml)
Definition: MsgLogger.h:158
Double_t Log(Double_t x)
Definition: TMath.h:748
Double_t Sqrt(Double_t x)
Definition: TMath.h:679
Definition: file.py:1