VHbb/python/train.py

#!/usr/bin/env python
from optparse import OptionParser
import sys
import pickle
import ROOT 
ROOT.gROOT.SetBatch(True)
from array import array
#warnings.filterwarnings( action='ignore', category=RuntimeWarning, message='creating converter.*' )
#usage: ./train run gui

#CONFIGURE
argv = sys.argv
parser = OptionParser()
parser.add_option("-v", "--verbose", action="store_true", dest="verbose", default=False,
                          help="Verbose mode.")
parser.add_option("-T", "--training", dest="training", default="",
                      help="Training")
parser.add_option("-C", "--config", dest="config", default=[], action="append",
                      help="configuration file")
(opts, args) = parser.parse_args(argv)
if opts.config =="":
        opts.config = "config"

#Import after configure to get help message
from myutils import BetterConfigParser, mvainfo, ParseInfo, TreeCache

#load config
config = BetterConfigParser()
config.read(opts.config)
anaTag = config.get("Analysis","tag")
run=opts.training
gui=opts.verbose

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

#get locations:
MVAdir=config.get('Directories','vhbbpath')+'/data/'
samplesinfo=config.get('Directories','samplesinfo')

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

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

VHbbNameSpace=config.get('VHbbNameSpace','library')
ROOT.gSystem.Load(VHbbNameSpace)

#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 = MVAdir+factoryname+'_'+MVAname+'.root'
#locations
path=config.get('Directories','SYSout')

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(' ')    

#Infofile
info = ParseInfo(samplesinfo,path)

#Workdir
workdir=ROOT.gDirectory.GetPath()


TrainCut='%s & EventForTraining==1'%TCut
EvalCut='%s & EventForTraining==0'%TCut
cuts = [TrainCut,EvalCut] 


samples = []
samples = info.get_samples(signals+backgrounds)

tc = TreeCache(cuts,samples,path,config)

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

print '\n\t>>> READING EVENTS <<<\n'

signal_samples = info.get_samples(signals)
background_samples = info.get_samples(backgrounds)

#TRAIN trees
Tbackgrounds = []
TbScales = []
Tsignals = []
TsScales = []
#EVAL trees
Ebackgrounds = []
EbScales = []
Esignals = []
EsScales = []

#load trees
for job in signal_samples:
    print '\tREADING IN %s AS SIG'%job.name
    Tsignal = tc.get_tree(job,TrainCut)
    ROOT.gDirectory.Cd(workdir)
    TsScale = tc.get_scale(job,config)*global_rescale    
    Tsignals.append(Tsignal)
    TsScales.append(TsScale)
    Esignal = tc.get_tree(job,EvalCut)
    Esignals.append(Esignal)
    EsScales.append(TsScale)
    print '\t\t\tTraining %s events'%Tsignal.GetEntries()
    print '\t\t\tEval %s events'%Esignal.GetEntries()
for job in background_samples:
    print '\tREADING IN %s AS BKG'%job.name
    Tbackground = tc.get_tree(job,TrainCut)
    ROOT.gDirectory.Cd(workdir)
    TbScale = tc.get_scale(job,config)*global_rescale
    Tbackgrounds.append(Tbackground)
    TbScales.append(TbScale)
    Ebackground = tc.get_tree(job,EvalCut)
    ROOT.gDirectory.Cd(workdir)
    Ebackgrounds.append(Ebackground)
    EbScales.append(TbScale)
    print '\t\t\tTraining %s events'%Tbackground.GetEntries()
    print '\t\t\tEval %s events'%Ebackground.GetEntries()
            

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

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

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

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


Revision:	1.20
Committed:	Fri Feb 1 08:59:35 2013 UTC (12 years, 3 months ago) by nmohr
Content type:	text/x-python
Branch:	MAIN
Changes since 1.19:	+1 -1 lines
Log Message:	Persistent caching
#	User	Rev	Content
1	peller	1.1	#!/usr/bin/env python
2	nmohr	1.9	from optparse import OptionParser
3	peller	1.1	import sys
4	peller	1.16	import pickle
5			import ROOT
6	nmohr	1.19	ROOT.gROOT.SetBatch(True)
7	peller	1.16	from array import array
8	peller	1.1	#warnings.filterwarnings( action='ignore', category=RuntimeWarning, message='creating converter.*' )
9			#usage: ./train run gui
10
11			#CONFIGURE
12	nmohr	1.9	argv = sys.argv
13			parser = OptionParser()
14			parser.add_option("-v", "--verbose", action="store_true", dest="verbose", default=False,
15			help="Verbose mode.")
16			parser.add_option("-T", "--training", dest="training", default="",
17			help="Training")
18			parser.add_option("-C", "--config", dest="config", default=[], action="append",
19			help="configuration file")
20			(opts, args) = parser.parse_args(argv)
21			if opts.config =="":
22			opts.config = "config"
23	peller	1.1
24	nmohr	1.19	#Import after configure to get help message
25			from myutils import BetterConfigParser, mvainfo, ParseInfo, TreeCache
26
27	peller	1.1	#load config
28	nmohr	1.6	config = BetterConfigParser()
29	nmohr	1.9	config.read(opts.config)
30			anaTag = config.get("Analysis","tag")
31			run=opts.training
32			gui=opts.verbose
33	peller	1.7
34			#GLOABAL rescale from Train/Test Spliiting:
35			global_rescale=2.
36	peller	1.1
37			#get locations:
38	nmohr	1.19	MVAdir=config.get('Directories','vhbbpath')+'/data/'
39	nmohr	1.13	samplesinfo=config.get('Directories','samplesinfo')
40	peller	1.1
41			#systematics
42			systematics=config.get('systematics','systematics')
43			systematics=systematics.split(' ')
44
45			weightF=config.get('Weights','weightF')
46
47	bortigno	1.14	VHbbNameSpace=config.get('VHbbNameSpace','library')
48			ROOT.gSystem.Load(VHbbNameSpace)
49	peller	1.7
50	peller	1.1	#CONFIG
51			#factory
52			factoryname=config.get('factory','factoryname')
53			factorysettings=config.get('factory','factorysettings')
54			#MVA
55			MVAtype=config.get(run,'MVAtype')
56			MVAname=run
57			MVAsettings=config.get(run,'MVAsettings')
58	nmohr	1.19	fnameOutput = MVAdir+factoryname+'_'+MVAname+'.root'
59	peller	1.1	#locations
60	nmohr	1.17	path=config.get('Directories','SYSout')
61	peller	1.1
62			TCutname=config.get(run, 'treeCut')
63			TCut=config.get('Cuts',TCutname)
64	peller	1.3	#print TCut
65	peller	1.1
66			#signals
67			signals=config.get(run,'signals')
68			signals=signals.split(' ')
69			#backgrounds
70			backgrounds=config.get(run,'backgrounds')
71			backgrounds=backgrounds.split(' ')
72
73			treeVarSet=config.get(run,'treeVarSet')
74
75			#variables
76			#TreeVar Array
77			MVA_Vars={}
78			MVA_Vars['Nominal']=config.get(treeVarSet,'Nominal')
79			MVA_Vars['Nominal']=MVA_Vars['Nominal'].split(' ')
80	bortigno	1.15
81	nmohr	1.19	#Infofile
82			info = ParseInfo(samplesinfo,path)
83	peller	1.1
84			#Workdir
85			workdir=ROOT.gDirectory.GetPath()
86
87
88	nmohr	1.9	TrainCut='%s & EventForTraining==1'%TCut
89			EvalCut='%s & EventForTraining==0'%TCut
90	nmohr	1.19	cuts = [TrainCut,EvalCut]
91	peller	1.1
92
93	nmohr	1.19	samples = []
94			samples = info.get_samples(signals+backgrounds)
95	peller	1.7
96	nmohr	1.20	tc = TreeCache(cuts,samples,path,config)
97	peller	1.7
98			output = ROOT.TFile.Open(fnameOutput, "RECREATE")
99
100	nmohr	1.19	print '\n\t>>> READING EVENTS <<<\n'
101
102			signal_samples = info.get_samples(signals)
103			background_samples = info.get_samples(backgrounds)
104	peller	1.7
105	nmohr	1.19	#TRAIN trees
106			Tbackgrounds = []
107			TbScales = []
108			Tsignals = []
109			TsScales = []
110			#EVAL trees
111	peller	1.1	Ebackgrounds = []
112			EbScales = []
113			Esignals = []
114			EsScales = []
115
116	nmohr	1.19	#load trees
117			for job in signal_samples:
118			print '\tREADING IN %s AS SIG'%job.name
119			Tsignal = tc.get_tree(job,TrainCut)
120			ROOT.gDirectory.Cd(workdir)
121			TsScale = tc.get_scale(job,config)*global_rescale
122			Tsignals.append(Tsignal)
123			TsScales.append(TsScale)
124			Esignal = tc.get_tree(job,EvalCut)
125			Esignals.append(Esignal)
126			EsScales.append(TsScale)
127			print '\t\t\tTraining %s events'%Tsignal.GetEntries()
128			print '\t\t\tEval %s events'%Esignal.GetEntries()
129			for job in background_samples:
130			print '\tREADING IN %s AS BKG'%job.name
131			Tbackground = tc.get_tree(job,TrainCut)
132			ROOT.gDirectory.Cd(workdir)
133			TbScale = tc.get_scale(job,config)*global_rescale
134			Tbackgrounds.append(Tbackground)
135			TbScales.append(TbScale)
136			Ebackground = tc.get_tree(job,EvalCut)
137			ROOT.gDirectory.Cd(workdir)
138			Ebackgrounds.append(Ebackground)
139			EbScales.append(TbScale)
140			print '\t\t\tTraining %s events'%Tbackground.GetEntries()
141			print '\t\t\tEval %s events'%Ebackground.GetEntries()
142
143	peller	1.7
144	peller	1.1	factory = ROOT.TMVA.Factory(factoryname, output, factorysettings)
145
146			#set input trees
147			for i in range(len(Tsignals)):
148			factory.AddSignalTree(Tsignals[i], TsScales[i], ROOT.TMVA.Types.kTraining)
149			factory.AddSignalTree(Esignals[i], EsScales[i], ROOT.TMVA.Types.kTesting)
150
151			for i in range(len(Tbackgrounds)):
152			if (Tbackgrounds[i].GetEntries()>0):
153			factory.AddBackgroundTree(Tbackgrounds[i], TbScales[i], ROOT.TMVA.Types.kTraining)
154
155			if (Ebackgrounds[i].GetEntries()>0):
156			factory.AddBackgroundTree(Ebackgrounds[i], EbScales[i], ROOT.TMVA.Types.kTesting)
157
158			for var in MVA_Vars['Nominal']:
159			factory.AddVariable(var,'D') # add the variables
160
161			#Execute TMVA
162			factory.SetSignalWeightExpression(weightF)
163	peller	1.12	factory.SetBackgroundWeightExpression(weightF)
164	peller	1.1	factory.Verbose()
165			factory.BookMethod(MVAtype,MVAname,MVAsettings)
166			factory.TrainAllMethods()
167			factory.TestAllMethods()
168			factory.EvaluateAllMethods()
169			output.Write()
170
171			#WRITE INFOFILE
172	nmohr	1.19	infofile = open(MVAdir+factoryname+'_'+MVAname+'.info','w')
173	peller	1.1	info=mvainfo(MVAname)
174			info.factoryname=factoryname
175			info.factorysettings=factorysettings
176			info.MVAtype=MVAtype
177			info.MVAsettings=MVAsettings
178	nmohr	1.10	info.weightfilepath=MVAdir
179	peller	1.1	info.path=path
180			info.varset=treeVarSet
181			info.vars=MVA_Vars['Nominal']
182			pickle.dump(info,infofile)
183			infofile.close()
184
185			# open the TMVA Gui
186	nmohr	1.9	if gui == True:
187	peller	1.16	ROOT.gROOT.ProcessLine( ".L myutils/TMVAGui.C")
188	peller	1.1	ROOT.gROOT.ProcessLine( "TMVAGui(\"%s\")" % fnameOutput )
189			ROOT.gApplication.Run()
190
191