VHbb/python/train.py

#!/usr/bin/env python
from samplesclass import sample
from printcolor import printc
import pickle
import ROOT 
from ROOT import TFile, TTree
import ROOT
from array import array
from BetterConfigParser import BetterConfigParser
import sys
from mvainfos import mvainfo
from gethistofromtree import getScale


#warnings.filterwarnings( action='ignore', category=RuntimeWarning, message='creating converter.*' )


#usage: ./train run gui


#CONFIGURE

#load config
config = BetterConfigParser()
#config.read('./config')
config.read('./config7TeV')


#GLOABAL rescale from Train/Test Spliiting:
global_rescale=2.

#get locations:
Wdir=config.get('Directories','Wdir')

#systematics
systematics=config.get('systematics','systematics')
systematics=systematics.split(' ')

weightF=config.get('Weights','weightF')


def getTree(job,cut,subsample=-1):

    newinput = TFile.Open(job.getpath(),'read')
    output.cd()
    Tree = newinput.Get(job.tree)
    #Tree.SetDirectory(0)

      
    if subsample>-1:
        CuttedTree=Tree.CopyTree('(%s) & (%s)'%(cut,job.subcuts[subsample]))    
        #print '\t--> read in %s'%job.group[subsample]

    else:
        CuttedTree=Tree.CopyTree(cut)
        #print '\t--> read in %s'%job.name


    #CuttedTree.SetDirectory(0)
    return CuttedTree
        
#def getScale(job,subsample=-1):
#    input = TFile.Open(job.getpath())
#    CountWithPU = input.Get("CountWithPU")
#    CountWithPU2011B = input.Get("CountWithPU2011B")
#    #print lumi*xsecs[i]/hist.GetBinContent(1)
#    return float(job.lumi)*float(job.xsec)*float(job.sf)/(0.46502*CountWithPU.GetBinContent(1)+0.53498*CountWithPU2011B.GetBinContent(1))*2/float(job.split)

run=sys.argv[1]
gui=sys.argv[2]


#CONFIG
#factory
factoryname=config.get('factory','factoryname')
factorysettings=config.get('factory','factorysettings')
#MVA
MVAtype=config.get(run,'MVAtype')
MVAname=run
MVAsettings=config.get(run,'MVAsettings')
fnameOutput = Wdir +'/weights/'+factoryname+'_'+MVAname+'.root'
#locations
path=config.get(run,'path')

TCutname=config.get(run, 'treeCut')
TCut=config.get('Cuts',TCutname)
#print TCut

#signals
signals=config.get(run,'signals')
signals=signals.split(' ')
#backgrounds
backgrounds=config.get(run,'backgrounds')
backgrounds=backgrounds.split(' ')

treeVarSet=config.get(run,'treeVarSet')
        
#variables
#TreeVar Array
MVA_Vars={}
MVA_Vars['Nominal']=config.get(treeVarSet,'Nominal')
MVA_Vars['Nominal']=MVA_Vars['Nominal'].split(' ')    
#Spectators:
#spectators=config.get(treeVarSet,'spectators')
#spectators=spectators.split(' ')

#TRAINING samples
infofile = open(path+'/samples.info','r')
info = pickle.load(infofile)
infofile.close()

#Workdir
workdir=ROOT.gDirectory.GetPath()


TrainCut='%s && EventForTraining==1'%TCut
EvalCut='%s && EventForTraining==0'%TCut

#load TRAIN trees
Tbackgrounds = []
TbScales = []
Tsignals = []
TsScales = []


output = ROOT.TFile.Open(fnameOutput, "RECREATE")

print '\n*** TRAINING EVENTS ***\n'

for job in info:
    if eval(job.active):
        if job.name in signals:
            print '\tREADING IN %s AS SIG'%job.name
            Tsignal = getTree(job,TrainCut)
            ROOT.gDirectory.Cd(workdir)
            TsScale = getScale(job,global_rescale)
            Tsignals.append(Tsignal)
            TsScales.append(TsScale)
            
        if job.name in backgrounds:
            if job.subsamples:
                print '\tREADING IN SUBSAMPLES of %s AS BKG'%job.name
                for subsample in range(0,len(job.group)):
                    print '\t- %s'%job.group[subsample]
                    Tbackground = getTree(job,TrainCut,subsample)
                    ROOT.gDirectory.Cd(workdir)
                    TbScale = getScale(job,global_rescale,subsample)
                    Tbackgrounds.append(Tbackground)
                    TbScales.append(TbScale)
            
            
            else:
                print '\tREADING IN %s AS BKG'%job.name
                Tbackground = getTree(job,TrainCut)
                ROOT.gDirectory.Cd(workdir)
                TbScale = getScale(job,global_rescale)
                Tbackgrounds.append(Tbackground)
                TbScales.append(TbScale)
        
#load EVALUATE trees
Ebackgrounds = []
EbScales = []
Esignals = []
EsScales = []

print '\n*** TESTING EVENTS ***\n'


for job in info:
    if eval(job.active):

        if job.name in signals:
            print '\tREADING IN %s AS SIG'%job.name
            Esignal = getTree(job,EvalCut)
            ROOT.gDirectory.Cd(workdir)
            EsScale = getScale(job,global_rescale)
            Esignals.append(Esignal)
            EsScales.append(EsScale)
            
        if job.name in backgrounds:
            if job.subsamples:
                print '\tREADING IN SUBSAMPLES of %s AS BKG'%job.name
                for subsample in range(0,len(job.group)):
                    print '\t- %s'%job.group[subsample]
                    Ebackground = getTree(job,EvalCut,subsample)
                    ROOT.gDirectory.Cd(workdir)
                    EbScale = getScale(job,global_rescale,subsample)
                    Ebackgrounds.append(Ebackground)
                    EbScales.append(EbScale)
            
            
            else:
                print '\tREADING IN %s AS BKG'%job.name
                Ebackground = getTree(job,EvalCut)
                ROOT.gDirectory.Cd(workdir)
                EbScale = getScale(job,global_rescale)
                Ebackgrounds.append(Ebackground)
                EbScales.append(EbScale)
        

#output = ROOT.TFile.Open(fnameOutput, "RECREATE")
factory = ROOT.TMVA.Factory(factoryname, output, factorysettings)

#set input trees
for i in range(len(Tsignals)):

    factory.AddSignalTree(Tsignals[i], TsScales[i], ROOT.TMVA.Types.kTraining)
    factory.AddSignalTree(Esignals[i], EsScales[i], ROOT.TMVA.Types.kTesting)

for i in range(len(Tbackgrounds)):
    if (Tbackgrounds[i].GetEntries()>0):
        factory.AddBackgroundTree(Tbackgrounds[i], TbScales[i], ROOT.TMVA.Types.kTraining)

    if (Ebackgrounds[i].GetEntries()>0):
        factory.AddBackgroundTree(Ebackgrounds[i], EbScales[i], ROOT.TMVA.Types.kTesting)
        
        
for var in MVA_Vars['Nominal']:
    factory.AddVariable(var,'D') # add the variables
#for var in spectators:
#    factory.AddSpectator(var,'D') #add specators

#Execute TMVA
factory.SetSignalWeightExpression(weightF)
factory.Verbose()
factory.BookMethod(MVAtype,MVAname,MVAsettings)
factory.TrainAllMethods()
factory.TestAllMethods()
factory.EvaluateAllMethods()
output.Write()

#WRITE INFOFILE
infofile = open(Wdir+'/weights/'+factoryname+'_'+MVAname+'.info','w')
info=mvainfo(MVAname)
info.factoryname=factoryname
info.factorysettings=factorysettings
info.MVAtype=MVAtype
info.MVAsettings=MVAsettings
info.weightfilepath=Wdir+'/weights'
info.path=path
info.varset=treeVarSet
info.vars=MVA_Vars['Nominal']
#info.spectators=spectators
pickle.dump(info,infofile)
infofile.close()

# open the TMVA Gui 
if gui == 'gui': 
    ROOT.gROOT.ProcessLine( ".L TMVAGui.C")
    ROOT.gROOT.ProcessLine( "TMVAGui(\"%s\")" % fnameOutput )
    ROOT.gApplication.Run() 


Revision:	1.7
Committed:	Thu Aug 2 16:03:52 2012 UTC (12 years, 9 months ago) by peller
Content type:	text/x-python
Branch:	MAIN
Changes since 1.6:	+103 -40 lines
Log Message:	removed flavour splitting, fixed training readin problem
#	User	Rev	Content
1	peller	1.1	#!/usr/bin/env python
2	peller	1.5	from samplesclass import sample
3	peller	1.1	from printcolor import printc
4			import pickle
5			import ROOT
6			from ROOT import TFile, TTree
7			import ROOT
8			from array import array
9	nmohr	1.6	from BetterConfigParser import BetterConfigParser
10	peller	1.1	import sys
11			from mvainfos import mvainfo
12	peller	1.7	from gethistofromtree import getScale
13
14	peller	1.1
15			#warnings.filterwarnings( action='ignore', category=RuntimeWarning, message='creating converter.*' )
16
17
18			#usage: ./train run gui
19
20
21			#CONFIGURE
22
23			#load config
24	nmohr	1.6	config = BetterConfigParser()
25	peller	1.7	#config.read('./config')
26			config.read('./config7TeV')
27
28
29			#GLOABAL rescale from Train/Test Spliiting:
30			global_rescale=2.
31	peller	1.1
32			#get locations:
33			Wdir=config.get('Directories','Wdir')
34
35			#systematics
36			systematics=config.get('systematics','systematics')
37			systematics=systematics.split(' ')
38
39			weightF=config.get('Weights','weightF')
40
41	peller	1.7
42
43
44			def getTree(job,cut,subsample=-1):
45
46			newinput = TFile.Open(job.getpath(),'read')
47			output.cd()
48			Tree = newinput.Get(job.tree)
49	peller	1.1	#Tree.SetDirectory(0)
50	peller	1.7
51
52			if subsample>-1:
53			CuttedTree=Tree.CopyTree('(%s) & (%s)'%(cut,job.subcuts[subsample]))
54			#print '\t--> read in %s'%job.group[subsample]
55
56			else:
57			CuttedTree=Tree.CopyTree(cut)
58			#print '\t--> read in %s'%job.name
59
60
61			#CuttedTree.SetDirectory(0)
62	peller	1.1	return CuttedTree
63
64	peller	1.7	#def getScale(job,subsample=-1):
65			# input = TFile.Open(job.getpath())
66			# CountWithPU = input.Get("CountWithPU")
67			# CountWithPU2011B = input.Get("CountWithPU2011B")
68			# #print lumi*xsecs[i]/hist.GetBinContent(1)
69			# return float(job.lumi)float(job.xsec)float(job.sf)/(0.46502CountWithPU.GetBinContent(1)+0.53498CountWithPU2011B.GetBinContent(1))*2/float(job.split)
70	peller	1.1
71			run=sys.argv[1]
72			gui=sys.argv[2]
73
74
75			#CONFIG
76			#factory
77			factoryname=config.get('factory','factoryname')
78			factorysettings=config.get('factory','factorysettings')
79			#MVA
80			MVAtype=config.get(run,'MVAtype')
81			MVAname=run
82			MVAsettings=config.get(run,'MVAsettings')
83			fnameOutput = Wdir +'/weights/'+factoryname+'_'+MVAname+'.root'
84			#locations
85			path=config.get(run,'path')
86
87			TCutname=config.get(run, 'treeCut')
88			TCut=config.get('Cuts',TCutname)
89	peller	1.3	#print TCut
90	peller	1.1
91			#signals
92			signals=config.get(run,'signals')
93			signals=signals.split(' ')
94			#backgrounds
95			backgrounds=config.get(run,'backgrounds')
96			backgrounds=backgrounds.split(' ')
97
98			treeVarSet=config.get(run,'treeVarSet')
99
100			#variables
101			#TreeVar Array
102			MVA_Vars={}
103			MVA_Vars['Nominal']=config.get(treeVarSet,'Nominal')
104			MVA_Vars['Nominal']=MVA_Vars['Nominal'].split(' ')
105			#Spectators:
106	peller	1.3	#spectators=config.get(treeVarSet,'spectators')
107			#spectators=spectators.split(' ')
108	peller	1.1
109			#TRAINING samples
110			infofile = open(path+'/samples.info','r')
111			info = pickle.load(infofile)
112			infofile.close()
113
114			#Workdir
115			workdir=ROOT.gDirectory.GetPath()
116
117
118			TrainCut='%s && EventForTraining==1'%TCut
119			EvalCut='%s && EventForTraining==0'%TCut
120
121			#load TRAIN trees
122			Tbackgrounds = []
123			TbScales = []
124			Tsignals = []
125			TsScales = []
126
127	peller	1.7
128
129			output = ROOT.TFile.Open(fnameOutput, "RECREATE")
130
131			print '\n* TRAINING EVENTS *\n'
132
133	peller	1.1	for job in info:
134	peller	1.7	if eval(job.active):
135			if job.name in signals:
136			print '\tREADING IN %s AS SIG'%job.name
137			Tsignal = getTree(job,TrainCut)
138			ROOT.gDirectory.Cd(workdir)
139			TsScale = getScale(job,global_rescale)
140			Tsignals.append(Tsignal)
141			TsScales.append(TsScale)
142
143			if job.name in backgrounds:
144			if job.subsamples:
145			print '\tREADING IN SUBSAMPLES of %s AS BKG'%job.name
146			for subsample in range(0,len(job.group)):
147			print '\t- %s'%job.group[subsample]
148			Tbackground = getTree(job,TrainCut,subsample)
149			ROOT.gDirectory.Cd(workdir)
150			TbScale = getScale(job,global_rescale,subsample)
151			Tbackgrounds.append(Tbackground)
152			TbScales.append(TbScale)
153
154
155			else:
156			print '\tREADING IN %s AS BKG'%job.name
157			Tbackground = getTree(job,TrainCut)
158			ROOT.gDirectory.Cd(workdir)
159			TbScale = getScale(job,global_rescale)
160			Tbackgrounds.append(Tbackground)
161			TbScales.append(TbScale)
162	peller	1.1
163			#load EVALUATE trees
164			Ebackgrounds = []
165			EbScales = []
166			Esignals = []
167			EsScales = []
168
169	peller	1.7	print '\n* TESTING EVENTS *\n'
170
171
172	peller	1.1	for job in info:
173	peller	1.7	if eval(job.active):
174
175			if job.name in signals:
176			print '\tREADING IN %s AS SIG'%job.name
177			Esignal = getTree(job,EvalCut)
178			ROOT.gDirectory.Cd(workdir)
179			EsScale = getScale(job,global_rescale)
180			Esignals.append(Esignal)
181			EsScales.append(EsScale)
182
183			if job.name in backgrounds:
184			if job.subsamples:
185			print '\tREADING IN SUBSAMPLES of %s AS BKG'%job.name
186			for subsample in range(0,len(job.group)):
187			print '\t- %s'%job.group[subsample]
188			Ebackground = getTree(job,EvalCut,subsample)
189			ROOT.gDirectory.Cd(workdir)
190			EbScale = getScale(job,global_rescale,subsample)
191			Ebackgrounds.append(Ebackground)
192			EbScales.append(EbScale)
193
194
195			else:
196			print '\tREADING IN %s AS BKG'%job.name
197			Ebackground = getTree(job,EvalCut)
198			ROOT.gDirectory.Cd(workdir)
199			EbScale = getScale(job,global_rescale)
200			Ebackgrounds.append(Ebackground)
201			EbScales.append(EbScale)
202
203	peller	1.1
204	peller	1.7
205			#output = ROOT.TFile.Open(fnameOutput, "RECREATE")
206	peller	1.1	factory = ROOT.TMVA.Factory(factoryname, output, factorysettings)
207
208			#set input trees
209			for i in range(len(Tsignals)):
210
211			factory.AddSignalTree(Tsignals[i], TsScales[i], ROOT.TMVA.Types.kTraining)
212			factory.AddSignalTree(Esignals[i], EsScales[i], ROOT.TMVA.Types.kTesting)
213
214			for i in range(len(Tbackgrounds)):
215			if (Tbackgrounds[i].GetEntries()>0):
216			factory.AddBackgroundTree(Tbackgrounds[i], TbScales[i], ROOT.TMVA.Types.kTraining)
217
218			if (Ebackgrounds[i].GetEntries()>0):
219			factory.AddBackgroundTree(Ebackgrounds[i], EbScales[i], ROOT.TMVA.Types.kTesting)
220
221
222			for var in MVA_Vars['Nominal']:
223			factory.AddVariable(var,'D') # add the variables
224			#for var in spectators:
225			# factory.AddSpectator(var,'D') #add specators
226
227			#Execute TMVA
228			factory.SetSignalWeightExpression(weightF)
229			factory.Verbose()
230			factory.BookMethod(MVAtype,MVAname,MVAsettings)
231			factory.TrainAllMethods()
232			factory.TestAllMethods()
233			factory.EvaluateAllMethods()
234			output.Write()
235
236			#WRITE INFOFILE
237			infofile = open(Wdir+'/weights/'+factoryname+'_'+MVAname+'.info','w')
238			info=mvainfo(MVAname)
239			info.factoryname=factoryname
240			info.factorysettings=factorysettings
241			info.MVAtype=MVAtype
242			info.MVAsettings=MVAsettings
243			info.weightfilepath=Wdir+'/weights'
244			info.path=path
245			info.varset=treeVarSet
246			info.vars=MVA_Vars['Nominal']
247	peller	1.3	#info.spectators=spectators
248	peller	1.1	pickle.dump(info,infofile)
249			infofile.close()
250
251			# open the TMVA Gui
252			if gui == 'gui':
253			ROOT.gROOT.ProcessLine( ".L TMVAGui.C")
254			ROOT.gROOT.ProcessLine( "TMVAGui(\"%s\")" % fnameOutput )
255			ROOT.gApplication.Run()
256
257