Logo ROOT   6.30.04
Reference Guide
 All Namespaces Files Pages
mvasMulticlass.cxx
Go to the documentation of this file.
1 #include "TMVA/mvasMulticlass.h"
2 #include "TMVA/Types.h"
3 #include "TLegend.h"
4 #include "TText.h"
5 #include "TH2.h"
6 
7 
8 
9 // this macro plots the resulting MVA distributions (Signal and
10 // Background overlayed) of different MVA methods run in TMVA
11 // (e.g. running TMVAnalysis.C).
12 
13 
14 // input: - Input file (result from TMVA)
15 // - use of TMVA plotting TStyle
16 void TMVA::mvasMulticlass(TString dataset, TString fin , HistType htype , Bool_t useTMVAStyle )
17 {
18  // set style and remove existing canvas'
19  TMVAGlob::Initialize( useTMVAStyle );
20 
21  // switches
22  const Bool_t Save_Images = kTRUE;
23 
24  // checks if file with name "fin" is already open, and if not opens one
25  TFile *file = TMVAGlob::OpenFile(fin);
26 
27  // find directory from input variable transformation and extract class names
28  TIter keys_dir = file->GetDirectory(dataset.Data())->GetListOfKeys();
29  TKey *key;
30  TDirectory *tempdir = 0;
31  while ((key = (TKey *)keys_dir())) {
32  TString name = key->GetName();
33  if (name.BeginsWith("InputVariables_")) {
34  tempdir = (TDirectory *)(file->GetDirectory(dataset.Data())->Get(name));
35  break;
36  }
37  }
38  std::vector<TString> classnames(TMVAGlob::GetClassNames(tempdir));
39 
40  // define Canvas layout here!
41  // Int_t xPad = 1; // no of plots in x
42  // Int_t yPad = 1; // no of plots in y
43  // Int_t noPad = xPad * yPad ;
44  const Int_t width = 600; // size of canvas
45 
46  // this defines how many canvases we need
47  TCanvas *c = 0;
48 
49  // counter variables
50  Int_t countCanvas = 0;
51 
52  // search for the right histograms in full list of keys
53  TIter next(file->GetDirectory(dataset.Data())->GetListOfKeys());
54  key = 0;
55  while ((key = (TKey*)next())) {
56 
57  if (!TString(key->GetName()).BeginsWith("Method_")) continue;
58  if (!gROOT->GetClass(key->GetClassName())->InheritsFrom("TDirectory")) continue;
59 
60  TString methodName;
61  TMVAGlob::GetMethodName(methodName,key);
62 
63  TDirectory* mDir = (TDirectory*)key->ReadObj();
64 
65  TIter keyIt(mDir->GetListOfKeys());
66  TKey *titkey;
67  while ((titkey = (TKey*)keyIt())) {
68 
69  if (!gROOT->GetClass(titkey->GetClassName())->InheritsFrom("TDirectory")) continue;
70 
71  TDirectory *titDir = (TDirectory *)titkey->ReadObj();
72  TString methodTitle;
73  TMVAGlob::GetMethodTitle(methodTitle,titDir);
74 
75  cout << "--- Found directory for method: " << methodName << "::" << methodTitle << endl;
76  TString hname = "MVA_" + methodTitle;
77  for(UInt_t icls = 0; icls < classnames.size(); ++icls){
78  TObjArray hists;
79 
80  std::vector<TString>::iterator classiter = classnames.begin();
81  for(; classiter!=classnames.end(); ++classiter){
82  TString name(hname+"_Test_"+ classnames.at(icls)
83  + "_prob_for_" + *classiter);
84  TH1 *hist = (TH1*)titDir->Get(name);
85  if (hist==0){
86  cout << ":\t mva distribution not available (this is normal for Cut classifier)" << endl;
87  continue;
88  }
89  hists.Add(hist);
90  }
91 
92 
93  // chop off useless stuff
94  ((TH1*)hists.First())->SetTitle( Form("TMVA response for classifier: %s", methodTitle.Data() ));
95 
96  // create new canvas
97  //cout << "Create canvas..." << endl;
98  TString ctitle = ((htype == kMVAType) ?
99  Form("TMVA response for class %s %s", classnames.at(icls).Data(),methodTitle.Data()) :
100  Form("TMVA comparison for class %s %s", classnames.at(icls).Data(),methodTitle.Data())) ;
101 
102  c = new TCanvas( Form("canvas%d", countCanvas+1), ctitle,
103  countCanvas*50+200, countCanvas*20, width, (Int_t)width*0.78 );
104 
105  // set the histogram style
106  //cout << "Set histogram style..." << endl;
107  TMVAGlob::SetMultiClassStyle( &hists );
108 
109  // normalise all histograms and find maximum
110  Float_t histmax = -1;
111  for(Int_t i=0; i<hists.GetEntriesFast(); ++i){
112  TMVAGlob::NormalizeHist((TH1*)hists[i] );
113  if(((TH1*)hists[i])->GetMaximum() > histmax)
114  histmax = ((TH1*)hists[i])->GetMaximum();
115  }
116 
117  // frame limits (between 0 and 1 per definition)
118  Float_t xmin = 0;
119  Float_t xmax = 1;
120  Float_t ymin = 0;
121  Float_t maxMult = (htype == kCompareType) ? 1.3 : 1.2;
122  Float_t ymax = histmax*maxMult;
123  // build a frame
124  Int_t nb = 500;
125  TString hFrameName(TString("frame") + methodTitle);
126  TObject *o = gROOT->FindObject(hFrameName);
127  if(o) delete o;
128  TH2F* frame = new TH2F( hFrameName, ((TH1*)hists.First())->GetTitle(),
129  nb, xmin, xmax, nb, ymin, ymax );
130  frame->GetXaxis()->SetTitle( methodTitle + " response for "+classnames.at(icls));
131  frame->GetYaxis()->SetTitle("(1/N) dN^{ }/^{ }dx");
132  TMVAGlob::SetFrameStyle( frame );
133 
134  // eventually: draw the frame
135  frame->Draw();
136 
137  c->GetPad(0)->SetLeftMargin( 0.105 );
138  frame->GetYaxis()->SetTitleOffset( 1.2 );
139 
140  // Draw legend
141  TLegend *legend= new TLegend( c->GetLeftMargin(), 1 - c->GetTopMargin() - 0.12,
142  c->GetLeftMargin() + (htype == kCompareType ? 0.40 : 0.3), 1 - c->GetTopMargin() );
143  legend->SetFillStyle( 1 );
144  classiter = classnames.begin();
145 
146  for(Int_t i=0; i<hists.GetEntriesFast(); ++i, ++classiter){
147  legend->AddEntry(((TH1*)hists[i]),*classiter,"F");
148  }
149 
150  legend->SetBorderSize(1);
151  legend->SetMargin( 0.3 );
152  legend->Draw("same");
153 
154 
155  for(Int_t i=0; i<hists.GetEntriesFast(); ++i){
156 
157  ((TH1*)hists[i])->Draw("histsame");
158  TString ytit = TString("(1/N) ") + ((TH1*)hists[i])->GetYaxis()->GetTitle();
159  ((TH1*)hists[i])->GetYaxis()->SetTitle( ytit ); // histograms are normalised
160 
161  }
162 
163 
164  if (htype == kCompareType) {
165 
166  TObjArray othists;
167  // if overtraining check, load additional histograms
168  classiter = classnames.begin();
169  for(; classiter!=classnames.end(); ++classiter){
170  TString name(hname+"_Train_"+ classnames.at(icls)
171  + "_prob_for_" + *classiter);
172  TH1 *hist = (TH1*)titDir->Get(name);
173  if (hist==0){
174  cout << ":\t comparison histogram for overtraining check not available!" << endl;
175  continue;
176  }
177  othists.Add(hist);
178  }
179 
180  TLegend *legend2= new TLegend( 1 - c->GetRightMargin() - 0.42, 1 - c->GetTopMargin() - 0.12,
181  1 - c->GetRightMargin(), 1 - c->GetTopMargin() );
182  legend2->SetFillStyle( 1 );
183  legend2->SetBorderSize(1);
184 
185  classiter = classnames.begin();
186  for(Int_t i=0; i<othists.GetEntriesFast(); ++i, ++classiter){
187  legend2->AddEntry(((TH1*)othists[i]),*classiter+" (training sample)","P");
188  }
189  legend2->SetMargin( 0.1 );
190  legend2->Draw("same");
191 
192  // normalise all histograms and get maximum
193  for(Int_t i=0; i<othists.GetEntriesFast(); ++i){
194  TMVAGlob::NormalizeHist((TH1*)othists[i] );
195  if(((TH1*)othists[i])->GetMaximum() > histmax)
196  histmax = ((TH1*)othists[i])->GetMaximum();
197  }
198 
199  TMVAGlob::SetMultiClassStyle( &othists );
200  for(Int_t i=0; i<othists.GetEntriesFast(); ++i){
201  Int_t col = ((TH1*)hists[i])->GetLineColor();
202  ((TH1*)othists[i])->SetMarkerSize( 0.7 );
203  ((TH1*)othists[i])->SetMarkerStyle( 20 );
204  ((TH1*)othists[i])->SetMarkerColor( col );
205  ((TH1*)othists[i])->SetLineWidth( 1 );
206  ((TH1*)othists[i])->Draw("e1same");
207  }
208 
209  ymax = histmax*maxMult;
210  frame->GetYaxis()->SetLimits( 0, ymax );
211 
212  // for better visibility, plot thinner lines
213  TMVAGlob::SetMultiClassStyle( &othists );
214  for(Int_t i=0; i<hists.GetEntriesFast(); ++i){
215  ((TH1*)hists[i])->SetLineWidth( 1 );
216  }
217 
218 
219  // perform K-S test
220 
221  cout << "--- Perform Kolmogorov-Smirnov tests" << endl;
222  cout << "--- Goodness of consistency for class " << classnames.at(icls)<< endl;
223  //TString probatext("Kolmogorov-Smirnov test: ");
224  for(Int_t j=0; j<othists.GetEntriesFast(); ++j){
225  Float_t kol = ((TH1*)hists[j])->KolmogorovTest(((TH1*)othists[j]),"X");
226  cout << classnames.at(j) << ": " << kol << endl;
227  //probatext.Append(classnames.at(j)+Form(" %.3f ",kol));
228  }
229 
230 
231 
232  //TText* tt = new TText( 0.12, 0.74, probatext );
233  //tt->SetNDC(); tt->SetTextSize( 0.032 ); tt->AppendPad();
234 
235  }
236 
237 
238  // redraw axes
239  frame->Draw("sameaxis");
240 
241  // text for overflows
242  //Int_t nbin = sig->GetNbinsX();
243  //Double_t dxu = sig->GetBinWidth(0);
244  //Double_t dxo = sig->GetBinWidth(nbin+1);
245  //TString uoflow = Form( "U/O-flow (S,B): (%.1f, %.1f)%% / (%.1f, %.1f)%%",
246  // sig->GetBinContent(0)*dxu*100, bgd->GetBinContent(0)*dxu*100,
247  // sig->GetBinContent(nbin+1)*dxo*100, bgd->GetBinContent(nbin+1)*dxo*100 );
248  //TText* t = new TText( 0.975, 0.115, uoflow );
249  //t->SetNDC();
250  //t->SetTextSize( 0.030 );
251  //t->SetTextAngle( 90 );
252  //t->AppendPad();
253 
254  // update canvas
255  c->Update();
256 
257  // save canvas to file
258 
259  TMVAGlob::plot_logo(1.058);
260  if (Save_Images) {
261  if (htype == kMVAType) TMVAGlob::imgconv( c, Form("%s/plots/mva_%s_%s",dataset.Data(),classnames.at(icls).Data(), methodTitle.Data()) );
262  else if (htype == kCompareType) TMVAGlob::imgconv( c, Form("%s/plots/overtrain_%s_%s",dataset.Data(),classnames.at(icls).Data(), methodTitle.Data()) );
263 
264  }
265  countCanvas++;
266  }
267  }
268  cout << "";
269  }
270 }
271