| 3 |
|
import sys |
| 4 |
|
import pickle |
| 5 |
|
import ROOT |
| 6 |
+ |
ROOT.gROOT.SetBatch(True) |
| 7 |
|
from array import array |
| 7 |
– |
from myutils import BetterConfigParser, sample, printc, mvainfo, parse_info |
| 8 |
– |
#ToDo: |
| 9 |
– |
from gethistofromtree import getScale |
| 10 |
– |
|
| 8 |
|
#warnings.filterwarnings( action='ignore', category=RuntimeWarning, message='creating converter.*' ) |
| 9 |
|
#usage: ./train run gui |
| 10 |
|
|
| 14 |
– |
|
| 11 |
|
#CONFIGURE |
| 12 |
|
argv = sys.argv |
| 13 |
|
parser = OptionParser() |
| 21 |
|
if opts.config =="": |
| 22 |
|
opts.config = "config" |
| 23 |
|
|
| 24 |
+ |
#Import after configure to get help message |
| 25 |
+ |
from myutils import BetterConfigParser, mvainfo, ParseInfo, TreeCache |
| 26 |
+ |
|
| 27 |
|
#load config |
| 28 |
|
config = BetterConfigParser() |
| 29 |
|
config.read(opts.config) |
| 35 |
|
global_rescale=2. |
| 36 |
|
|
| 37 |
|
#get locations: |
| 38 |
< |
MVAdir=config.get('Directories','vhbbpath')+'/data_test/' |
| 38 |
> |
MVAdir=config.get('Directories','vhbbpath')+'/data/' |
| 39 |
|
samplesinfo=config.get('Directories','samplesinfo') |
| 40 |
|
|
| 41 |
|
#systematics |
| 47 |
|
VHbbNameSpace=config.get('VHbbNameSpace','library') |
| 48 |
|
ROOT.gSystem.Load(VHbbNameSpace) |
| 49 |
|
|
| 51 |
– |
def getTree(job,cut,path,subsample=-1): |
| 52 |
– |
#print path+'/'+job.getpath() |
| 53 |
– |
newinput = ROOT.TFile.Open(path+'/'+job.getpath(),'read') |
| 54 |
– |
output.cd() |
| 55 |
– |
Tree = newinput.Get(job.tree) |
| 56 |
– |
#Tree.SetDirectory(0) |
| 57 |
– |
|
| 58 |
– |
|
| 59 |
– |
if subsample>-1: |
| 60 |
– |
#print 'cut: (%s) & (%s)'%(cut,job.subcuts[subsample]) |
| 61 |
– |
CuttedTree=Tree.CopyTree('(%s) & (%s)'%(cut,job.subcuts[subsample])) |
| 62 |
– |
CuttedTree.SetNameTitle(job.subnames[subsample],job.subnames[subsample]) |
| 63 |
– |
print '\t--> read in %s'%job.subnames[subsample] |
| 64 |
– |
else: |
| 65 |
– |
CuttedTree=Tree.CopyTree(cut) |
| 66 |
– |
CuttedTree.SetNameTitle(job.name,job.name) |
| 67 |
– |
print '\t--> read in %s'%job.name |
| 68 |
– |
newinput.Close() |
| 69 |
– |
|
| 70 |
– |
#CuttedTree.SetDirectory(0) |
| 71 |
– |
return CuttedTree |
| 72 |
– |
|
| 73 |
– |
#def getScale(job,subsample=-1): |
| 74 |
– |
# input = TFile.Open(job.getpath()) |
| 75 |
– |
# CountWithPU = input.Get("CountWithPU") |
| 76 |
– |
# CountWithPU2011B = input.Get("CountWithPU2011B") |
| 77 |
– |
# #print lumi*xsecs[i]/hist.GetBinContent(1) |
| 78 |
– |
# return float(job.lumi)*float(job.xsec)*float(job.sf)/(0.46502*CountWithPU.GetBinContent(1)+0.53498*CountWithPU2011B.GetBinContent(1))*2/float(job.split) |
| 79 |
– |
|
| 50 |
|
#CONFIG |
| 81 |
– |
#suffix for output name |
| 82 |
– |
suffix='_newVars_v2' |
| 51 |
|
#factory |
| 52 |
|
factoryname=config.get('factory','factoryname') |
| 53 |
|
factorysettings=config.get('factory','factorysettings') |
| 55 |
|
MVAtype=config.get(run,'MVAtype') |
| 56 |
|
MVAname=run |
| 57 |
|
MVAsettings=config.get(run,'MVAsettings') |
| 58 |
< |
fnameOutput = MVAdir+factoryname+'_'+MVAname+suffix+'.root' |
| 58 |
> |
fnameOutput = MVAdir+factoryname+'_'+MVAname+'.root' |
| 59 |
|
#locations |
| 60 |
|
path=config.get('Directories','SYSout') |
| 61 |
|
|
| 77 |
|
MVA_Vars={} |
| 78 |
|
MVA_Vars['Nominal']=config.get(treeVarSet,'Nominal') |
| 79 |
|
MVA_Vars['Nominal']=MVA_Vars['Nominal'].split(' ') |
| 112 |
– |
#Spectators: |
| 113 |
– |
#spectators=config.get(treeVarSet,'spectators') |
| 114 |
– |
#spectators=spectators.split(' ') |
| 115 |
– |
|
| 116 |
– |
#TRAINING samples |
| 117 |
– |
#infofile = open(samplesinfo,'r') |
| 118 |
– |
#info = pickle.load(infofile) |
| 119 |
– |
#infofile.close() |
| 80 |
|
|
| 81 |
< |
info = parse_info(samplesinfo,path) |
| 81 |
> |
#Infofile |
| 82 |
> |
info = ParseInfo(samplesinfo,path) |
| 83 |
|
|
| 84 |
|
#Workdir |
| 85 |
|
workdir=ROOT.gDirectory.GetPath() |
| 87 |
|
|
| 88 |
|
TrainCut='%s & EventForTraining==1'%TCut |
| 89 |
|
EvalCut='%s & EventForTraining==0'%TCut |
| 90 |
+ |
cuts = [TrainCut,EvalCut] |
| 91 |
|
|
| 130 |
– |
#load TRAIN trees |
| 131 |
– |
Tbackgrounds = [] |
| 132 |
– |
TbScales = [] |
| 133 |
– |
Tsignals = [] |
| 134 |
– |
TsScales = [] |
| 92 |
|
|
| 93 |
+ |
samples = [] |
| 94 |
+ |
samples = info.get_samples(signals+backgrounds) |
| 95 |
|
|
| 96 |
+ |
tc = TreeCache(cuts,samples,path) |
| 97 |
|
|
| 98 |
|
output = ROOT.TFile.Open(fnameOutput, "RECREATE") |
| 99 |
|
|
| 100 |
< |
print '\n\t>>> TRAINING EVENTS <<<\n' |
| 100 |
> |
print '\n\t>>> READING EVENTS <<<\n' |
| 101 |
|
|
| 102 |
< |
for job in info: |
| 103 |
< |
if eval(job.active): |
| 104 |
< |
|
| 105 |
< |
if job.subsamples: |
| 106 |
< |
print '\tREADING IN SUBSAMPLES of %s'%job.name |
| 107 |
< |
for subsample in range(0,len(job.group)): |
| 108 |
< |
if job.subnames[subsample] in signals: |
| 109 |
< |
print '\t- %s as SIG'%job.group[subsample] |
| 110 |
< |
Tsignal = getTree(job,TrainCut,path,subsample) |
| 151 |
< |
ROOT.gDirectory.Cd(workdir) |
| 152 |
< |
TsScale = getScale(job,path,config,global_rescale,subsample) |
| 153 |
< |
Tsignals.append(Tsignal) |
| 154 |
< |
TsScales.append(TsScale) |
| 155 |
< |
print '\t\t\t%s events'%Tsignal.GetEntries() |
| 156 |
< |
elif job.subnames[subsample] in backgrounds: |
| 157 |
< |
print '\t- %s as BKG'%job.group[subsample] |
| 158 |
< |
Tbackground = getTree(job,TrainCut,path,subsample) |
| 159 |
< |
ROOT.gDirectory.Cd(workdir) |
| 160 |
< |
TbScale = getScale(job,path,config,global_rescale,subsample) |
| 161 |
< |
Tbackgrounds.append(Tbackground) |
| 162 |
< |
TbScales.append(TbScale) |
| 163 |
< |
print '\t\t\t%s events'%Tbackground.GetEntries() |
| 164 |
< |
|
| 165 |
< |
else: |
| 166 |
< |
if job.name in signals: |
| 167 |
< |
print '\tREADING IN %s AS SIG'%job.name |
| 168 |
< |
Tsignal = getTree(job,TrainCut,path) |
| 169 |
< |
ROOT.gDirectory.Cd(workdir) |
| 170 |
< |
TsScale = getScale(job,path,config,global_rescale) |
| 171 |
< |
Tsignals.append(Tsignal) |
| 172 |
< |
TsScales.append(TsScale) |
| 173 |
< |
print '\t\t\t%s events'%Tsignal.GetEntries() |
| 174 |
< |
elif job.name in backgrounds: |
| 175 |
< |
print '\tREADING IN %s AS BKG'%job.name |
| 176 |
< |
Tbackground = getTree(job,TrainCut,path) |
| 177 |
< |
ROOT.gDirectory.Cd(workdir) |
| 178 |
< |
TbScale = getScale(job,path,config,global_rescale) |
| 179 |
< |
Tbackgrounds.append(Tbackground) |
| 180 |
< |
TbScales.append(TbScale) |
| 181 |
< |
print '\t\t\t%s events'%Tbackground.GetEntries() |
| 182 |
< |
|
| 183 |
< |
|
| 184 |
< |
#load EVALUATE trees |
| 102 |
> |
signal_samples = info.get_samples(signals) |
| 103 |
> |
background_samples = info.get_samples(backgrounds) |
| 104 |
> |
|
| 105 |
> |
#TRAIN trees |
| 106 |
> |
Tbackgrounds = [] |
| 107 |
> |
TbScales = [] |
| 108 |
> |
Tsignals = [] |
| 109 |
> |
TsScales = [] |
| 110 |
> |
#EVAL trees |
| 111 |
|
Ebackgrounds = [] |
| 112 |
|
EbScales = [] |
| 113 |
|
Esignals = [] |
| 114 |
|
EsScales = [] |
| 115 |
|
|
| 116 |
< |
print '\n\t>>> TESTING EVENTS <<<\n' |
| 117 |
< |
|
| 118 |
< |
|
| 119 |
< |
for job in info: |
| 120 |
< |
if eval(job.active): |
| 121 |
< |
|
| 122 |
< |
if job.subsamples: |
| 123 |
< |
print '\tREADING IN SUBSAMPLES of %s'%job.name |
| 124 |
< |
for subsample in range(0,len(job.group)): |
| 125 |
< |
if job.subnames[subsample] in signals: |
| 126 |
< |
print '\t- %s as SIG'%job.group[subsample] |
| 127 |
< |
Esignal = getTree(job,EvalCut,path,subsample) |
| 128 |
< |
ROOT.gDirectory.Cd(workdir) |
| 129 |
< |
EsScale = getScale(job,path,config,global_rescale,subsample) |
| 130 |
< |
Esignals.append(Esignal) |
| 131 |
< |
EsScales.append(EsScale) |
| 132 |
< |
print '\t\t\t%s events'%Esignal.GetEntries() |
| 133 |
< |
elif job.subnames[subsample] in backgrounds: |
| 134 |
< |
print '\t- %s as BKG'%job.group[subsample] |
| 135 |
< |
Ebackground = getTree(job,EvalCut,path,subsample) |
| 136 |
< |
ROOT.gDirectory.Cd(workdir) |
| 137 |
< |
EbScale = getScale(job,path,config,global_rescale,subsample) |
| 138 |
< |
Ebackgrounds.append(Ebackground) |
| 139 |
< |
EbScales.append(EbScale) |
| 140 |
< |
print '\t\t\t%s events'%Ebackground.GetEntries() |
| 141 |
< |
|
| 142 |
< |
else: |
| 217 |
< |
if job.name in signals: |
| 218 |
< |
print '\tREADING IN %s AS SIG'%job.name |
| 219 |
< |
Esignal = getTree(job,EvalCut,path) |
| 220 |
< |
ROOT.gDirectory.Cd(workdir) |
| 221 |
< |
EsScale = getScale(job,path,config,global_rescale) |
| 222 |
< |
Esignals.append(Esignal) |
| 223 |
< |
EsScales.append(EsScale) |
| 224 |
< |
print '\t\t\t%s events'%Esignal.GetEntries() |
| 225 |
< |
elif job.name in backgrounds: |
| 226 |
< |
print '\tREADING IN %s AS BKG'%job.name |
| 227 |
< |
Ebackground = getTree(job,EvalCut,path) |
| 228 |
< |
ROOT.gDirectory.Cd(workdir) |
| 229 |
< |
EbScale = getScale(job,path,config,global_rescale) |
| 230 |
< |
Ebackgrounds.append(Ebackground) |
| 231 |
< |
EbScales.append(EbScale) |
| 232 |
< |
print '\t\t\t%s events'%Ebackground.GetEntries() |
| 233 |
< |
|
| 116 |
> |
#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 |
|
|
| 235 |
– |
#output = ROOT.TFile.Open(fnameOutput, "RECREATE") |
| 144 |
|
factory = ROOT.TMVA.Factory(factoryname, output, factorysettings) |
| 145 |
|
|
| 146 |
|
#set input trees |
| 147 |
|
for i in range(len(Tsignals)): |
| 240 |
– |
|
| 241 |
– |
#print 'Number of SIG entries: %s'%Tsignals[i].GetEntries() |
| 148 |
|
factory.AddSignalTree(Tsignals[i], TsScales[i], ROOT.TMVA.Types.kTraining) |
| 243 |
– |
#print 'Number of SIG entries: %s'%Esignals[i].GetEntries() |
| 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): |
| 248 |
– |
#print 'Number of BKG entries: %s'%Tbackgrounds[i].GetEntries() |
| 153 |
|
factory.AddBackgroundTree(Tbackgrounds[i], TbScales[i], ROOT.TMVA.Types.kTraining) |
| 154 |
|
|
| 155 |
|
if (Ebackgrounds[i].GetEntries()>0): |
| 252 |
– |
#print 'Number of BKG entries: %s'%Ebackgrounds[i].GetEntries() |
| 156 |
|
factory.AddBackgroundTree(Ebackgrounds[i], EbScales[i], ROOT.TMVA.Types.kTesting) |
| 157 |
|
|
| 255 |
– |
|
| 158 |
|
for var in MVA_Vars['Nominal']: |
| 159 |
|
factory.AddVariable(var,'D') # add the variables |
| 258 |
– |
#for var in spectators: |
| 259 |
– |
# factory.AddSpectator(var,'D') #add specators |
| 160 |
|
|
| 161 |
|
#Execute TMVA |
| 162 |
|
factory.SetSignalWeightExpression(weightF) |
| 169 |
|
output.Write() |
| 170 |
|
|
| 171 |
|
#WRITE INFOFILE |
| 172 |
< |
infofile = open(MVAdir+factoryname+'_'+MVAname+suffix+'.info','w') |
| 172 |
> |
infofile = open(MVAdir+factoryname+'_'+MVAname+'.info','w') |
| 173 |
|
info=mvainfo(MVAname) |
| 174 |
|
info.factoryname=factoryname |
| 175 |
|
info.factorysettings=factorysettings |
| 179 |
|
info.path=path |
| 180 |
|
info.varset=treeVarSet |
| 181 |
|
info.vars=MVA_Vars['Nominal'] |
| 282 |
– |
#info.spectators=spectators |
| 182 |
|
pickle.dump(info,infofile) |
| 183 |
|
infofile.close() |
| 184 |
|
|
