Context Navigation

source: branches/MarsGapdTimeJitter/mjtrain/MJTrainCuts.cc@ 18237

Visit:

Last change on this file since 18237 was 9892, checked in by tbretz, 14 years ago
Fixed the 2D histograms in MJTrainCuts.
File size: 31.5 KB

Line
1	/* ======================================================================== *\
2	!
3	! *
4	! * This file is part of MARS, the MAGIC Analysis and Reconstruction
5	! * Software. It is distributed to you in the hope that it can be a useful
6	! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
7	! * It is distributed WITHOUT ANY WARRANTY.
8	! *
9	! * Permission to use, copy, modify and distribute this software and its
10	! * documentation for any purpose is hereby granted without fee,
11	! * provided that the above copyright notice appear in all copies and
12	! * that both that copyright notice and this permission notice appear
13	! * in supporting documentation. It is provided "as is" without express
14	! * or implied warranty.
15	! *
16	!
17	!
18	! Author(s): Thomas Bretz 08/2010 <mailto:thomas.bretz@epfl.ch>
19	!
20	! Copyright: MAGIC Software Development, 2010
21	!
22	!
23	\* ======================================================================== */
24
25	/////////////////////////////////////////////////////////////////////////////
26	//
27	// MJTrainCuts
28	// =========
29	//
30	// This class is meant as a tool to understand better what a trained
31	// random forest is doing in the multi-dimensional phase space.
32	// Consequently, it can also be used to deduce good one or two dimensional
33	// cuts from the results by mimicing the behaviour of the random forest.
34	//
35	//
36	// Usage
37	// -----
38	//
39	// The instance is created by its default constructor
40	//
41	// MJTrainCuts opt;
42	//
43	// In a first step a random forest must be trained and in a second step its
44	// performance can be evaluated with an independent test sample. The used
45	// samples are defined by two MDataSet objects, one for the on-data (e.g.
46	// gammas) and the other one for the off-data (e.g. protons). SequencesOn
47	// and SequencesOff are used for testing and training respectively.
48	//
49	// MDataSet seton ("myondata.txt");
50	// MDataSet setoff("myoffdata.txt");
51	//
52	// // If you want to use all available events
53	// opt.SetDataSetOn(seton);
54	// opt.SetDataSetOff(setoff);
55	//
56	// // Try to select 10000 and 30000 events for training and testing resp.
57	// // opt.SetDataSetOn(seton, 10000, 30000);
58	// // opt.SetDataSetOff(setoff, 10000, 30000);
59	//
60	// Note that by using several data set in one file (see MDataSet) you can
61	// have everything in a single file.
62	//
63	// The variables which are used for training are now setup as usual
64	//
65	// Int_t p1 = opt.AddParameter("MHillas.fSize");
66	// Int_t p2 = opt.AddParameter("MHillas.GetArea*MGeomCam.fConvMm2Deg^2");
67	// Int_t p3 = opt.AddParameter("MHillasSrc.fDist*MGeomCam.fConvMm2Deg");
68	//
69	// In addition you can now setup a binning for the display of each train
70	// parameter as follows (for details see MBinning)
71	//
72	// opt.AddBinning(p1, MBinning(40, 10, 10000, "", "log"));
73	// opt.AddBinning(p2, MBinning(50, 0, 0.25));
74	// opt.AddBinning(p3, MBinning(50, 0, 2.5));
75	//
76	// Since with increasing number of variables the possibly combinations
77	// increase to fast you have to define which plots you are interested in,
78	// for example:
79	//
80	// opt.AddHist(p3); // A 1D plot dist
81	// opt.AddHist(p1, p2); // A 2D plot area vs. size
82	// opt.AddHist(p3, p2); // A 2D plot dist vs. size
83	//
84	// Also 3D plots are avaiable but they are most probably difficult to
85	// interprete.
86	//
87	// In addition to this you have the usual user interface, i.e. that
88	// - PreCuts
89	// - TrainCuts
90	// - TestCuts
91	// - PreTasks
92	// - PostTasks
93	// - TestTasks
94	// are available. For details see MJOptimizeBase
95	//
96	// void EnableRegression() / void EnableClassification()
97	// Defines whether to use the random forest's regression of
98	// classification method. Classification is the default.
99	//
100	//
101	// The produced plots
102	// ------------------
103	//
104	// The tab with the plots filled will always look like this:
105	//
106	// +--------+--------+
107	// \|1 \|2 \|
108	// +--------+--------\|
109	// \|3 \|4 \|
110	// +--------+ \|
111	// \|5 \| \|
112	// +--------+--------+
113	//
114	// Pad1 and Pad2 contain the weighted event distribution of the test-sample.
115	//
116	// Pad2 and Pad5 conatin a profile of the hadronness distribution of the
117	// test-sample.
118	//
119	// Pad4 contains a profile of the hadronness distribution of on-data and
120	// off-data together of the test-data.
121	//
122	// If the profiles for on-data and off-data are identical the displayed
123	// hadronness is obviously independant of the other (non shown) trainings
124	// variables. Therefore the difference between the two plots show how much
125	// the variables are correlated. The same is true if the prfiles in
126	// pad3 and pad5 don't differe from the profile in pad4.
127	//
128	// In the most simple case - the random forest is only trained with the
129	// variables displayed - all three plots should be identical (apart from the
130	// difference in the distrubution of the three sets).
131	//
132	// The plot in pad4 can now be used to deduce a good classical cut in the
133	// displayed variables.
134	//
135	//
136	// Example:
137	// --------
138	//
139	// MJTrainCuts opt;
140	// MDataSet seton ("dataset_on.txt");
141	// MDataSet setoff("dataset_off.txt");
142	// opt.SetDataSetOn(seton);
143	// opt.SetDataSetOff(setmix);
144	// Int_t p00 = opt.AddParameter("MHillas.fSize");
145	// Int_t p01 = opt.AddParameter("MHillas.GetArea*MGeomCam.fConvMm2Deg^2");
146	// opt.AddHist(p00, p01); // Area vs Size
147	// MStatusDisplay *d = new MStatusDisplay;
148	// opt.SetDisplay(d);
149	// opt.Process("rf-cuts.root");
150	//
151	//
152	// Random Numbers:
153	// ---------------
154	// Use:
155	// if(gRandom)
156	// delete gRandom;
157	// gRandom = new TRandom3();
158	// in advance to change the random number generator.
159	//
160	////////////////////////////////////////////////////////////////////////////
161	#include "MJTrainCuts.h"
162
163	#include <TGraph.h>
164	#include <TMarker.h>
165	#include <TCanvas.h>
166	#include <TPRegexp.h>
167	#include <TStopwatch.h>
168
169	#include "MHMatrix.h"
170
171	#include "MLog.h"
172	#include "MLogManip.h"
173
174	// tools
175	#include "MMath.h"
176	#include "MBinning.h"
177	#include "MTFillMatrix.h"
178	#include "MStatusDisplay.h"
179
180	// eventloop
181	#include "MParList.h"
182	#include "MTaskList.h"
183	#include "MEvtLoop.h"
184
185	// tasks
186	#include "MReadMarsFile.h"
187	#include "MContinue.h"
188	#include "MFillH.h"
189	#include "MRanForestCalc.h"
190
191	// container
192	#include "MParameters.h"
193
194	// histograms
195	#include "MHn.h"
196	#include "MHHadronness.h"
197
198	// filter
199	#include "MFEventSelector.h"
200	#include "MFilterList.h"
201
202	using namespace std;
203
204	class HistSet1D : public TObject
205	{
206	protected:
207	UInt_t fNx;
208
209	virtual void AddHist(MHn &h, const char rx, const char , const char *) const
210	{
211	h.AddHist(rx);
212	}
213	virtual void AddProf(MHn &h, const char rx, const char , const char *) const
214	{
215	h.AddHist(rx, "MHadronness.fVal", MH3::kProfileSpread);
216	}
217	virtual void SetupName(MHn &h, const char *name) const
218	{
219	h.InitName(Form("%s%d;%d", name, fNx, fNx));
220	}
221	virtual void SetupHist(MHn &h, const char name, const char title) const
222	{
223	SetupName(h, name);
224
225	h.SetAutoRange(kFALSE, kFALSE, kFALSE);
226	h.InitTitle(title);
227	}
228
229	void CreateHist(MHn &h, const char rx, const char ry=0, const char *rz=0) const
230	{
231	h.SetLayout(MHn::kComplex);
232	h.SetBit(MHn::kDoNotReset);
233
234	AddHist(h, rx, ry, rz);
235	SetupHist(h, "DistOn", "Distribution of on-data");
236	h.SetWeight("Type.fVal");
237
238	AddHist(h, rx, ry, rz);
239	SetupHist(h, "DistOff", "Distribution of off-data");
240	h.SetWeight("1-Type.fVal");
241
242	AddProf(h, rx, ry, rz);
243	SetupHist(h, "HadOn", "Hadronness profile for on-data");
244	h.SetWeight("Type.fVal");
245
246	AddProf(h, rx, ry, rz);
247	SetupHist(h, "Had", "Hadronness profile for all events");
248
249	AddProf(h, rx, ry, rz);
250	SetupHist(h, "HadOff", "Hadronness profile for off-data");
251	h.SetWeight("1-Type.fVal");
252	}
253
254
255	public:
256	HistSet1D(UInt_t nx) : fNx(nx) { }
257
258	virtual MHn *GetHistN(const TList &rules) const
259	{
260	if (!rules.At(fNx))
261	return 0;
262
263	MHn *h = new MHn(Form("%d", fNx));
264
265	CreateHist(*h, rules.At(fNx)->GetName());
266
267	return h;
268	}
269
270	virtual Bool_t CheckBinning(const TObjArray &binnings) const
271	{
272	return binnings.FindObject(Form("Binning%d", fNx));
273	}
274	};
275
276	class HistSet2D : public HistSet1D
277	{
278	protected:
279	UInt_t fNy;
280
281	void AddHist(MHn &h, const char rx, const char ry, const char *) const
282	{
283	h.AddHist(rx, ry);
284	}
285	void AddProf(MHn &h, const char rx, const char ry, const char *) const
286	{
287	h.AddHist(rx, ry, "MHadronness.fVal", MH3::kProfileSpread);
288	}
289	void SetupName(MHn &h, const char *name) const
290	{
291	h.InitName(Form("%s%d:%d;%d;%d", name, fNx, fNy, fNx, fNy));
292	}
293
294	void SetupHist(MHn &h, const char name, const char title) const
295	{
296	HistSet1D::SetupHist(h, name, title);
297	h.SetDrawOption("colz");
298	}
299
300	public:
301	HistSet2D(UInt_t nx, UInt_t ny) : HistSet1D(nx), fNy(ny) { }
302
303	MHn *GetHistN(const TList &rules) const
304	{
305	if (!rules.At(fNx) \|\| !rules.At(fNy))
306	return 0;
307
308	MHn *h = new MHn(Form("%d:%d", fNx, fNy));
309
310	CreateHist(*h, rules.At(fNx)->GetName(), rules.At(fNy)->GetName());
311
312	return h;
313	}
314	Bool_t CheckBinning(const TObjArray &binnings) const
315	{
316	return HistSet1D::CheckBinning(binnings) && binnings.FindObject(Form("Binning%d", fNy));
317	}
318	};
319
320
321	class HistSet3D : public HistSet2D
322	{
323	private:
324	UInt_t fNz;
325
326	void AddHist(MHn &h, const char rx, const char ry, const char *rz) const
327	{
328	h.AddHist(rx, ry, rz);
329	}
330	void AddProf(MHn &h, const char rx, const char ry, const char *rz) const
331	{
332	h.AddHist(rx, ry, rz, "MHadronness.fVal");
333	}
334	void SetupName(MHn &h, const char *name) const
335	{
336	h.InitName(Form("%s%d:%d:%d;%d;%d;%d", name, fNx, fNy, fNz, fNx, fNy, fNz));
337	}
338
339	public:
340	HistSet3D(UInt_t nx, UInt_t ny, UInt_t nz) : HistSet2D(nx, ny), fNz(nz) { }
341
342	MHn *GetHistN(const TList &rules) const
343	{
344	if (!rules.At(fNx) \|\| !rules.At(fNy) \|\| !rules.At(fNz))
345	return 0;
346
347	MHn *h = new MHn(Form("%d:%d:%d", fNx, fNy, fNz));
348
349	CreateHist(*h,
350	rules.At(fNx)->GetName(),
351	rules.At(fNy)->GetName(),
352	rules.At(fNy)->GetName());
353
354	return h;
355	}
356	Bool_t CheckBinning(const TObjArray &binnings) const
357	{
358	return HistSet2D::CheckBinning(binnings) && binnings.FindObject(Form("Binning%d", fNz));
359	}
360	};
361
362	// ---------------------------------------------------------------------------------------
363
364	void MJTrainCuts::AddHist(UInt_t nx)
365	{
366	fHists.Add(new HistSet1D(nx));
367	}
368
369	void MJTrainCuts::AddHist(UInt_t nx, UInt_t ny)
370	{
371	fHists.Add(new HistSet2D(nx, ny));
372	}
373
374	void MJTrainCuts::AddHist(UInt_t nx, UInt_t ny, UInt_t nz)
375	{
376	fHists.Add(new HistSet3D(nx, ny, nz));
377	}
378
379	void MJTrainCuts::AddBinning(UInt_t n, const MBinning &bins)
380	{
381	const char *name = Form("Binning%d", n);
382
383	TObject *o = fBinnings.FindObject(name);
384	if (o)
385	{
386	delete fBinnings.Remove(o);
387	*fLog << warn << "WARNING - Binning for parameter " << n << " (" << name << ") already exists... replaced." << endl;
388	}
389
390	// FIXME: Check for existence
391	fBinnings.Add(new MBinning(bins, name, bins.GetTitle()));
392	}
393
394	/*
395	void MJTrainCuts::AddBinning(const MBinning &bins)
396	{
397	// FIXME: Check for existence
398	fBinnings.Add(new MBinning(bins, bins.GetName(), bins.GetTitle()));
399	}
400	*/
401
402	// ---------------------------------------------------------------------------------------
403
404	// --------------------------------------------------------------------------
405	//
406	void MJTrainCuts::DisplayResult(MH3 &h31, MH3 &h32, Float_t ontime)
407	{
408	TH2D &g = (TH2D&)h32.GetHist();
409	TH2D &h = (TH2D&)h31.GetHist();
410
411	h.SetMarkerColor(kRed);
412	g.SetMarkerColor(kBlue);
413
414	TH2D res1(g);
415	TH2D res2(g);
416
417	h.SetTitle("Hadronness-Distribution vs. Size");
418	res1.SetTitle("Significance Li/Ma");
419	res1.SetXTitle("Size [phe]");
420	res1.SetYTitle("Hadronness");
421	res2.SetTitle("Significance-Distribution");
422	res2.SetXTitle("Size-Cut [phe]");
423	res2.SetYTitle("Hadronness-Cut");
424	res1.SetContour(50);
425	res2.SetContour(50);
426
427	const Int_t nx = h.GetNbinsX();
428	const Int_t ny = h.GetNbinsY();
429
430	gROOT->SetSelectedPad(NULL);
431
432
433	Double_t Stot = 0;
434	Double_t Btot = 0;
435
436	Double_t max2 = -1;
437
438	TGraph gr1;
439	TGraph gr2;
440	for (int x=nx-1; x>=0; x--)
441	{
442	TH1 *hx = h.ProjectionY("H_py", x+1, x+1);
443	TH1 *gx = g.ProjectionY("G_py", x+1, x+1);
444
445	Double_t S = 0;
446	Double_t B = 0;
447
448	Double_t max1 = -1;
449	Int_t maxy1 = 0;
450	Int_t maxy2 = 0;
451	for (int y=ny-1; y>=0; y--)
452	{
453	const Float_t s = gx->Integral(1, y+1);
454	const Float_t b = hx->Integral(1, y+1);
455	const Float_t sig1 = MMath::SignificanceLiMa(s+b, b);
456	const Float_t sig2 = MMath::SignificanceLiMa(s+Stot+b+Btot, b+Btot)*TMath::Log10(s+Stot+1);
457	if (sig1>max1)
458	{
459	maxy1 = y;
460	max1 = sig1;
461	}
462	if (sig2>max2)
463	{
464	maxy2 = y;
465	max2 = sig2;
466
467	S=s;
468	B=b;
469	}
470
471	res1.SetBinContent(x+1, y+1, sig1);
472	}
473
474	Stot += S;
475	Btot += B;
476
477	gr1.SetPoint(x, h.GetXaxis()->GetBinCenter(x+1), h.GetYaxis()->GetBinCenter(maxy1+1));
478	gr2.SetPoint(x, h.GetXaxis()->GetBinCenter(x+1), h.GetYaxis()->GetBinCenter(maxy2+1));
479
480	delete hx;
481	delete gx;
482	}
483
484	//cout << "--> " << MMath::SignificanceLiMa(Stot+Btot, Btot) << " ";
485	//cout << Stot << " " << Btot << endl;
486
487
488	Int_t mx1=0;
489	Int_t my1=0;
490	Int_t mx2=0;
491	Int_t my2=0;
492	Int_t s1=0;
493	Int_t b1=0;
494	Int_t s2=0;
495	Int_t b2=0;
496	Double_t sig1=-1;
497	Double_t sig2=-1;
498	for (int x=0; x<nx; x++)
499	{
500	TH1 *hx = h.ProjectionY("H_py", x+1);
501	TH1 *gx = g.ProjectionY("G_py", x+1);
502	for (int y=0; y<ny; y++)
503	{
504	const Float_t s = gx->Integral(1, y+1);
505	const Float_t b = hx->Integral(1, y+1);
506	const Float_t sig = MMath::SignificanceLiMa(s+b, b);
507	res2.SetBinContent(x+1, y+1, sig);
508
509	// Search for top-rightmost maximum
510	if (sig>=sig1)
511	{
512	mx1=x+1;
513	my1=y+1;
514	s1 = TMath::Nint(s);
515	b1 = TMath::Nint(b);
516	sig1=sig;
517	}
518	if (TMath::Log10(s)*sig>=sig2)
519	{
520	mx2=x+1;
521	my2=y+1;
522	s2 = TMath::Nint(s);
523	b2 = TMath::Nint(b);
524	sig2=TMath::Log10(s)*sig;
525	}
526	}
527	delete hx;
528	delete gx;
529	}
530
531	TGraph gr3;
532	TGraph gr4;
533	gr4.SetTitle("Significance Li/Ma vs. Hadronness-cut");
534
535	TH1 *hx = h.ProjectionY("H_py");
536	TH1 *gx = g.ProjectionY("G_py");
537	for (int y=0; y<ny; y++)
538	{
539	const Float_t s = gx->Integral(1, y+1);
540	const Float_t b = hx->Integral(1, y+1);
541	const Float_t sg1 = MMath::SignificanceLiMa(s+b, b);
542	const Float_t sg2 = s<1 ? 0 : MMath::SignificanceLiMa(s+b, b)*TMath::Log10(s);
543
544	gr3.SetPoint(y, h.GetYaxis()->GetBinLowEdge(y+2), sg1);
545	gr4.SetPoint(y, h.GetYaxis()->GetBinLowEdge(y+2), sg2);
546	}
547	delete hx;
548	delete gx;
549
550	if (fDisplay)
551	{
552	TCanvas &c = fDisplay->AddTab("OptCut");
553	c.SetBorderMode(0);
554	c.Divide(2,2);
555
556	gROOT->SetSelectedPad(0);
557	c.cd(1);
558	gPad->SetBorderMode(0);
559	gPad->SetFrameBorderMode(0);
560	gPad->SetLogx();
561	gPad->SetGridx();
562	gPad->SetGridy();
563	h.DrawCopy();
564	g.DrawCopy("same");
565	gr1.SetMarkerStyle(kFullDotMedium);
566	gr1.DrawClone("LP")->SetBit(kCanDelete);
567	gr2.SetLineColor(kBlue);
568	gr2.SetMarkerStyle(kFullDotMedium);
569	gr2.DrawClone("LP")->SetBit(kCanDelete);
570
571	gROOT->SetSelectedPad(0);
572	c.cd(3);
573	gPad->SetBorderMode(0);
574	gPad->SetFrameBorderMode(0);
575	gPad->SetGridx();
576	gPad->SetGridy();
577	gr4.SetMinimum(0);
578	gr4.SetMarkerStyle(kFullDotMedium);
579	gr4.DrawClone("ALP")->SetBit(kCanDelete);
580	gr3.SetLineColor(kBlue);
581	gr3.SetMarkerStyle(kFullDotMedium);
582	gr3.DrawClone("LP")->SetBit(kCanDelete);
583
584	c.cd(2);
585	gPad->SetBorderMode(0);
586	gPad->SetFrameBorderMode(0);
587	gPad->SetLogx();
588	gPad->SetGridx();
589	gPad->SetGridy();
590	gPad->AddExec("color", "gStyle->SetPalette(1, 0);");
591	res1.SetMaximum(7);
592	res1.DrawCopy("colz");
593
594	c.cd(4);
595	gPad->SetBorderMode(0);
596	gPad->SetFrameBorderMode(0);
597	gPad->SetLogx();
598	gPad->SetGridx();
599	gPad->SetGridy();
600	gPad->AddExec("color", "gStyle->SetPalette(1, 0);");
601	res2.SetMaximum(res2.GetMaximum()*1.05);
602	res2.DrawCopy("colz");
603
604	// Int_t mx, my, mz;
605	// res2.GetMaximumBin(mx, my, mz);
606
607	TMarker m;
608	m.SetMarkerStyle(kStar);
609	m.DrawMarker(res2.GetXaxis()->GetBinCenter(mx1), res2.GetYaxis()->GetBinCenter(my1));
610	m.SetMarkerStyle(kPlus);
611	m.DrawMarker(res2.GetXaxis()->GetBinCenter(mx2), res2.GetYaxis()->GetBinCenter(my2));
612	}
613
614	if (ontime>0)
615	*fLog << all << "Observation Time: " << TMath::Nint(ontime/60) << "min" << endl;
616	*fLog << "Maximum Significance: " << Form("%.1f", sig1);
617	if (ontime>0)
618	*fLog << Form(" [%.1f/sqrt(h)]", sig1/TMath::Sqrt(ontime/3600));
619	*fLog << endl;
620
621	*fLog << "Significance: S=" << Form("%.1f", sig1) << " E=" << s1 << " B=" << b1 << " h<";
622	*fLog << Form("%.2f", res2.GetYaxis()->GetBinCenter(my1)) << " s>";
623	*fLog << Form("%3d", TMath::Nint(res2.GetXaxis()->GetBinCenter(mx1))) << endl;
624	fLog << "SignificanceLogE: S=" << Form("%.1f", sig2/TMath::Log10(s2)) << " E=" << s2 << " B=" << b2 << " h<";
625	*fLog << Form("%.2f", res2.GetYaxis()->GetBinCenter(my2)) << " s>";
626	*fLog << Form("%3d", TMath::Nint(res2.GetXaxis()->GetBinCenter(mx2))) << endl;
627	*fLog << endl;
628	}
629
630	// --------------------------------------------------------------------------
631	//
632	Bool_t MJTrainCuts::Process(const char *out)
633	{
634	// =========================== Consistency checks ==================================
635	if (!fDataSetOn.IsValid())
636	{
637	*fLog << err << "ERROR - DataSet for on-data invalid!" << endl;
638	return kFALSE;
639	}
640	if (!fDataSetOff.IsValid())
641	{
642	*fLog << err << "ERROR - DataSet for off-data invalid!" << endl;
643	return kFALSE;
644	}
645
646	if (fDataSetOn.IsWobbleMode()!=fDataSetOff.IsWobbleMode())
647	{
648	*fLog << err << "ERROR - On- and Off-DataSet have different observation modes!" << endl;
649	return kFALSE;
650	}
651
652	if (fDataSetOn.IsMonteCarlo()!=fDataSetOff.IsMonteCarlo())
653	{
654	*fLog << err << "ERROR - On- and Off-DataSet have different monte carlo modes!" << endl;
655	return kFALSE;
656	}
657
658	if (!HasWritePermission(out))
659	return kFALSE;
660
661	// Check if needed binning exists
662	TIter NextH(&fHists);
663	TObject *o = 0;
664	while ((o=NextH()))
665	{
666	const HistSet1D hs = static_cast<HistSet1D>(o);
667	if (hs->CheckBinning(fBinnings))
668	continue;
669
670	*fLog << err << "ERROR - Not all needed binnning exist." << endl;
671	return kFALSE;
672	}
673
674	// =========================== Preparation ==================================
675
676	if (fDisplay)
677	fDisplay->SetTitle(out);
678
679	TStopwatch clock;
680	clock.Start();
681
682	// ------------------ Setup reading --------------------
683	MReadMarsFile read1("Events");
684	MReadMarsFile read2("Events");
685	MReadMarsFile read3("Events");
686	MReadMarsFile read4("Events");
687	read1.DisableAutoScheme();
688	read2.DisableAutoScheme();
689	read3.DisableAutoScheme();
690	read4.DisableAutoScheme();
691
692	// Setup four reading tasks with the on- and off-data of the two datasets
693	// Training -- On
694	if (!fDataSetOn.AddFilesOn(read1))
695	return kFALSE;
696	// Testing -- On
697	if (!fDataSetOn.AddFilesOff(read4))
698	return kFALSE;
699	// Training -- Off
700	if (!fDataSetOff.AddFilesOn(read3))
701	return kFALSE;
702	// Testing -- Off
703	if (!fDataSetOff.AddFilesOff(read2))
704	return kFALSE;
705
706	// ===============================================================================
707	// ====================== Training =========================
708	// ===============================================================================
709
710	// ---------------- Setup RF Matrix ----------------
711	MHMatrix train("Train");
712	train.AddColumns(fRules);
713	// if (fEnableWeights[kTrainOn] \|\| fEnableWeights[kTrainOff])
714	// train.AddColumn("MWeight.fVal");
715	train.AddColumn("MHadronness.fVal");
716
717	// ----------------- Prepare filling Matrix RF ------------------
718
719	// Setup the hadronness container identifying gammas and off-data
720	// and setup a container for the weights
721	MParameterD had("MHadronness");
722	MParameterD wgt("MWeight");
723	MParameterD typ("Type");
724
725	// Add them to the parameter list
726	MParList plistx;
727	plistx.AddToList(this); // take care of fDisplay!
728	plistx.AddToList(&had);
729	plistx.AddToList(&wgt);
730	plistx.AddToList(&typ);
731
732	// Setup the tool class to fill the matrix
733	MTFillMatrix fill;
734	fill.SetLogStream(fLog);
735	fill.SetDisplay(fDisplay);
736	fill.AddPreCuts(fPreCuts);
737	fill.AddPreCuts(fTrainCuts);
738
739	// ----------------- Fill on data into matrix ------------------
740
741	// Setup the tool class to read the gammas and read them
742	fill.SetName("FillOn");
743	fill.SetDestMatrix1(&train, fNum[kTrainOn]);
744	fill.SetReader(&read1);
745	// fill.AddPreTasks(fPreTasksSet[kTrainOn]);
746	fill.AddPreTasks(fPreTasks);
747	// fill.AddPostTasks(fPostTasksSet[kTrainOn]);
748	fill.AddPostTasks(fPostTasks);
749
750	// Set classifier for gammas
751	had.SetVal(0);
752	wgt.SetVal(1);
753	typ.SetVal(0);
754
755	// Fill matrix
756	if (!fill.Process(plistx))
757	return kFALSE;
758
759	// Check the number or read events
760	const Int_t numontrn = train.GetNumRows();
761	if (numontrn==0)
762	{
763	*fLog << err << "ERROR - No on-data events available for training... aborting." << endl;
764	return kFALSE;
765	}
766
767	// Remove possible post tasks
768	fill.ClearPreTasks();
769	fill.ClearPostTasks();
770
771	// ----------------- Fill off data into matrix ------------------
772
773	// In case of wobble mode we have to do something special
774	// Setup the tool class to read the background and read them
775	fill.SetName("FillOff");
776	fill.SetDestMatrix1(&train, fNum[kTrainOff]);
777	fill.SetReader(&read3);
778	// fill.AddPreTasks(fPreTasksSet[kTrainOff]);
779	fill.AddPreTasks(fPreTasks);
780	// fill.AddPostTasks(fPostTasksSet[kTrainOff]);
781	fill.AddPostTasks(fPostTasks);
782
783	// Set classifier for background
784	had.SetVal(1);
785	wgt.SetVal(1);
786	typ.SetVal(1);
787
788	// Fiull matrix
789	if (!fill.Process(plistx))
790	return kFALSE;
791
792	// Check the number or read events
793	const Int_t numofftrn = train.GetNumRows()-numontrn;
794	if (numofftrn==0)
795	{
796	*fLog << err << "ERROR - No off-data available for training... aborting." << endl;
797	return kFALSE;
798	}
799
800	// ------------------------ Train RF --------------------------
801
802	MRanForestCalc rf("TrainSeparation", fTitle);
803	rf.SetNumTrees(fNumTrees);
804	rf.SetNdSize(fNdSize);
805	rf.SetNumTry(fNumTry);
806	rf.SetNumObsoleteVariables(1);
807	// rf.SetLastDataColumnHasWeights(fEnableWeights[kTrainOn] \|\| fEnableWeights[kTrainOff]);
808	rf.SetDebug(fDebug>1);
809	rf.SetDisplay(fDisplay);
810	rf.SetLogStream(fLog);
811	rf.SetFileName(out);
812	rf.SetNameOutput("MHadronness");
813
814	// Train the random forest either by classification or regression
815	if (!rf.Train(train, fUseRegression))
816	return kFALSE;
817
818	// ----------------- Print result of training ------------------
819
820	// Output information about what was going on so far.
821	*fLog << all;
822	fLog->Separator("The forest was trained with...");
823
824	*fLog << "Training method:" << endl;
825	fLog << " " << (fUseRegression?"regression":"classification") << endl;
826	/*
827	if (fEnableWeights[kTrainOn])
828	fLog << " weights for on-data" << endl;
829	if (fEnableWeights[kTrainOff])
830	fLog << " weights for off-data" << endl;
831	*/
832	*fLog << endl;
833	*fLog << "Events used for training:" << endl;
834	fLog << " Gammas: " << numontrn << endl;
835	fLog << " Background: " << numofftrn << endl;
836	*fLog << endl;
837	*fLog << "Gamma/Background ratio:" << endl;
838	fLog << " Requested: " << (float)fNum[kTrainOn]/fNum[kTrainOff] << endl;
839	fLog << " Result: " << (float)numontrn/numofftrn << endl;
840	*fLog << endl;
841	*fLog << "Run-Time: " << Form("%.1f", clock.RealTime()/60) << "min (CPU: ";
842	*fLog << Form("%.1f", clock.CpuTime()/60) << "min)" << endl;
843	*fLog << endl;
844	*fLog << "Output file name: " << out << endl;
845
846	// ===============================================================================
847	// ====================== Testing =========================
848	// ===============================================================================
849	fLog->Separator("Test");
850
851	clock.Continue();
852
853	// ---------------------- Prepare eventloop off-data ---------------------
854
855	// Setup parlist and tasklist for testing
856	MParList plist;
857	MTaskList tlist;
858	plist.AddToList(this); // Take care of display
859	plist.AddToList(&tlist);
860
861	// MMcEvt mcevt;
862	// plist.AddToList(&mcevt);
863
864	plist.AddToList(&wgt);
865	plist.AddToList(&typ);
866
867	// ----- Setup histograms -----
868	MBinning binsy(50, 0 , 1, "BinningMH3Y", "lin");
869	MBinning binsx(40, 10, 100000, "BinningMH3X", "log");
870
871	plist.AddToList(&binsx);
872	plist.AddToList(&binsy);
873
874	MH3 h31("MHillas.fSize", "MHadronness.fVal");
875	MH3 h32("MHillas.fSize", "MHadronness.fVal");
876	MH3 h40("MMcEvt.fEnergy", "MHadronness.fVal");
877	h31.SetTitle("Background probability vs. Size:Size [phe]:Hadronness h");
878	h32.SetTitle("Background probability vs. Size:Size [phe]:Hadronness h");
879	h40.SetTitle("Background probability vs. Energy:Energy [GeV]:Hadronness h");
880
881	plist.AddToList(&fBinnings);
882
883	MHHadronness hist;
884
885	// ----- Setup tasks -----
886	MFillH fillh0(&hist, "", "FillHadronness");
887	MFillH fillh1(&h31, "", "FillHadVsSize");
888	MFillH fillh2(&h32, "", "FillHadVsSize");
889	MFillH fillh4(&h40, "", "FillHadVsEnergy");
890	fillh0.SetWeight("MWeight");
891	fillh1.SetWeight("MWeight");
892	fillh2.SetWeight("MWeight");
893	fillh4.SetWeight("MWeight");
894	fillh1.SetDrawOption("colz profy");
895	fillh2.SetDrawOption("colz profy");
896	fillh4.SetDrawOption("colz profy");
897	fillh1.SetNameTab("HadSzOff");
898	fillh2.SetNameTab("HadSzOn");
899	fillh4.SetNameTab("HadEnOn");
900	fillh0.SetBit(MFillH::kDoNotDisplay);
901
902	// ----- Setup filter -----
903	MFilterList precuts;
904	precuts.AddToList(fPreCuts);
905	precuts.AddToList(fTestCuts);
906
907	MContinue cont0(&precuts);
908	cont0.SetName("PreCuts");
909	cont0.SetInverted();
910
911	MFEventSelector sel; // FIXME: USING IT (WITH PROB?) in READ will by much faster!!!
912	sel.SetNumSelectEvts(fNum[kTestOff]);
913
914	MContinue contsel(&sel);
915	contsel.SetInverted();
916
917	// ----- Setup tasklist -----
918	tlist.AddToList(&read2); // Reading task
919	tlist.AddToList(&contsel); // event selector
920	// tlist.AddToList(fPreTasksSet[kTestOff]);
921	tlist.AddToList(fPreTasks); // list of pre tasks
922	tlist.AddToList(&cont0); // list of pre cuts and test cuts
923	tlist.AddToList(&rf); // evaluate random forest
924	// tlist.AddToList(fPostTasksSet[kTestOff]);
925	tlist.AddToList(fPostTasks); // list of post tasks
926	tlist.AddToList(&fillh1); // Fill HadSzOff
927
928	TList autodel;
929	autodel.SetOwner();
930
931	TPRegexp regexp("([0-9]:*)+");
932
933	NextH.Reset();
934	while ((o=NextH()))
935	{
936	HistSet1D hs = static_cast<HistSet1D>(o);
937
938	// FIXME: Move to beginning of function
939	// Check if needed binning exists
940	if (!hs->CheckBinning(fBinnings))
941	return kFALSE;
942
943	MHn *hist = hs->GetHistN(fRules);
944	MFillH *fill = new MFillH(hist, "", Form("Fill%s", hist->GetName()));
945
946	fill->SetWeight("MWeight");
947	fill->SetDrawOption("colz");
948	fill->SetNameTab(hist->GetName());
949	fill->SetBit(MFillH::kDoNotDisplay);
950
951	tlist.AddToList(fill);
952
953	autodel.Add(fill);
954	autodel.Add(hist);
955	}
956	tlist.AddToList(&fillh0); // Fill MHHadronness (not displayed in first loop)
957	tlist.AddToList(&fTestTasks); // list of test tasks
958
959	// Enable Acceleration
960	tlist.SetAccelerator(MTask::kAccDontReset\|MTask::kAccDontTime);
961
962	// ---------------------- Run eventloop on background ---------------------
963	MEvtLoop loop;
964	loop.SetDisplay(fDisplay);
965	loop.SetLogStream(fLog);
966	loop.SetParList(&plist);
967	//if (!SetupEnv(loop))
968	// return kFALSE;
969
970	wgt.SetVal(1);
971	typ.SetVal(0);
972	if (!loop.Eventloop())
973	return kFALSE;
974
975	// ---------------------- Prepare eventloop on-data ---------------------
976
977	sel.SetNumSelectEvts(fNum[kTestOn]); // set number of target events
978
979	fillh0.ResetBit(MFillH::kDoNotDisplay); // Switch on display MHHadronness
980
981	TIter NextF(&autodel);
982	while ((o=NextF()))
983	{
984	MFillH fill = dynamic_cast<MFillH>(o);
985	if (fill)
986	fill->ResetBit(MFillH::kDoNotDisplay);
987	}
988
989	// Remove PreTasksOff and PostTasksOff from the list
990	// tlist.RemoveFromList(fPreTasksSet[kTestOff]);
991	// tlist.RemoveFromList(fPostTasksSet[kTestOff]);
992
993	tlist.Replace(&read4); // replace reading off-data by on-data
994
995	// Add the PreTasksOn directly after the reading task
996	// tlist.AddToListAfter(fPreTasksSet[kTestOn], &c1);
997
998	// Add the PostTasksOn after rf
999	// tlist.AddToListAfter(fPostTasksSet[kTestOn], &rf);
1000
1001	tlist.Replace(&fillh2); // Fill HadSzOn instead of HadSzOff
1002	tlist.AddToListAfter(&fillh4, &fillh0); // Filling of HadEnOn
1003
1004	// Enable Acceleration
1005	tlist.SetAccelerator(MTask::kAccDontReset\|MTask::kAccDontTime);
1006
1007	// ---------------------- Run eventloop on-data ---------------------
1008
1009	wgt.SetVal(1);
1010	typ.SetVal(1);
1011	if (!loop.Eventloop())
1012	return kFALSE;
1013
1014	// ---------------------- Print/Display result ---------------------
1015
1016	// Show what was going on in the testing
1017	const Double_t numontst = h32.GetHist().GetEntries();
1018	const Double_t numofftst = h31.GetHist().GetEntries();
1019
1020	*fLog << all;
1021	fLog->Separator("The forest was tested with...");
1022	*fLog << "Test method:" << endl;
1023	fLog << " Random Forest: " << out << endl;
1024	/*
1025	if (fEnableWeights[kTestOn])
1026	fLog << " weights for on-data" << endl;
1027	if (fEnableWeights[kTestOff])
1028	fLog << " weights for off-data" << endl;
1029	*/
1030	*fLog << endl;
1031	*fLog << "Events used for test:" << endl;
1032	fLog << " Gammas: " << numontst << endl;
1033	fLog << " Background: " << numofftst << endl;
1034	*fLog << endl;
1035	*fLog << "Gamma/Background ratio:" << endl;
1036	fLog << " Requested: " << (float)fNum[kTestOn]/fNum[kTestOff] << endl;
1037	fLog << " Result: " << (float)numontst/numofftst << endl;
1038	*fLog << endl;
1039
1040	// Display the result plots
1041	DisplayResult(h31, h32, -1);
1042	//DisplayResult(h31, h32, ontime);
1043
1044	*fLog << "Total Run-Time: " << Form("%.1f", clock.RealTime()/60) << "min (CPU: ";
1045	*fLog << Form("%.1f", clock.CpuTime()/60) << "min)" << endl;
1046	fLog->Separator();
1047
1048	// ----------------- Write result ------------------
1049
1050	fDataSetOn.SetName("DataSetOn");
1051	fDataSetOff.SetName("DataSetOff");
1052
1053	// Write the display
1054	TObjArray arr;
1055	arr.Add(const_cast<MDataSet*>(&fDataSetOn));
1056	arr.Add(const_cast<MDataSet*>(&fDataSetOff));
1057	if (fDisplay)
1058	arr.Add(fDisplay);
1059
1060	SetPathOut(out);
1061	return WriteContainer(arr, 0, "UPDATE");
1062	}
1063

Note: See TracBrowser for help on using the repository browser.

Download in other formats: