| 1 |
kukartse |
1.1 |
#!/usr/bin/env python
|
| 2 |
|
|
# @(#)root/tmva $Id: TMVAnalysis.py,v 1.8 2007/06/20 09:41:24 brun Exp $
|
| 3 |
|
|
# ------------------------------------------------------------------------------ #
|
| 4 |
|
|
# Project : TMVA - a Root-integrated toolkit for multivariate data analysis #
|
| 5 |
|
|
# Package : TMVA #
|
| 6 |
|
|
# Python script: TMVAnalysis.py #
|
| 7 |
|
|
# #
|
| 8 |
|
|
# This python script provides examples for the training and testing of all the #
|
| 9 |
|
|
# TMVA classifiers through PyROOT. Note that the use PyROOT requires that you #
|
| 10 |
|
|
# have a python version > 2.2 installed on your computer. #
|
| 11 |
|
|
# #
|
| 12 |
|
|
# As input data is used a toy-MC sample consisting of four Gaussian-distributed #
|
| 13 |
|
|
# and linearly correlated input variables. #
|
| 14 |
|
|
# #
|
| 15 |
|
|
# The methods to be used can be switched on and off via the prompt command, for #
|
| 16 |
|
|
# example: #
|
| 17 |
|
|
# #
|
| 18 |
|
|
# python TMVAnalysis.py --methods Fisher,Likelihood #
|
| 19 |
|
|
# #
|
| 20 |
|
|
# The output file "TMVA.root" can be analysed with the use of dedicated #
|
| 21 |
|
|
# macros (simply say: root -l <../macros/macro.C>), which can be conveniently #
|
| 22 |
|
|
# invoked through a GUI that will appear at the end of the run of this macro. #
|
| 23 |
|
|
# #
|
| 24 |
|
|
# for help type "python TMVAnalysis.py --help" #
|
| 25 |
|
|
# ------------------------------------------------------------------------------ #
|
| 26 |
|
|
|
| 27 |
|
|
# --------------------------------------------
|
| 28 |
|
|
# standard python import
|
| 29 |
|
|
import sys # exit
|
| 30 |
|
|
import time # time accounting
|
| 31 |
|
|
import getopt # command line parser
|
| 32 |
|
|
|
| 33 |
|
|
# --------------------------------------------
|
| 34 |
|
|
|
| 35 |
|
|
# default settings for command line arguments
|
| 36 |
|
|
DEFAULT_OUTFNAME = "TMVA.root"
|
| 37 |
|
|
DEFAULT_INFNAME = "../examples/data/toy_sigbkg.root"
|
| 38 |
|
|
DEFAULT_TREESIG = "TreeS"
|
| 39 |
|
|
DEFAULT_TREEBKG = "TreeB"
|
| 40 |
|
|
DEFAULT_METHODS = "CutsGA Likelihood LikelihoodPCA PDERS KNN HMatrix Fisher FDA MLP SVM_Gauss BDT RuleFitTMVA"
|
| 41 |
|
|
|
| 42 |
|
|
# print help
|
| 43 |
|
|
def usage():
|
| 44 |
|
|
print " "
|
| 45 |
|
|
print "Usage: python %s [options]" % sys.argv[0]
|
| 46 |
|
|
print " -m | --methods : gives methods to be run (default: all methods)"
|
| 47 |
|
|
print " -i | --inputfile : name of input ROOT file (default: '%s')" % DEFAULT_INFNAME
|
| 48 |
|
|
print " -o | --outputfile : name of output ROOT file containing results (default: '%s')" % DEFAULT_OUTFNAME
|
| 49 |
|
|
print " -t | --inputtrees : input ROOT Trees for signal and background (default: '%s %s')" \
|
| 50 |
|
|
% (DEFAULT_TREESIG, DEFAULT_TREEBKG)
|
| 51 |
|
|
print " -v | --verbose"
|
| 52 |
|
|
print " -? | --usage : print this help message"
|
| 53 |
|
|
print " -h | --help : print this help message"
|
| 54 |
|
|
print " "
|
| 55 |
|
|
|
| 56 |
|
|
# main routine
|
| 57 |
|
|
def main():
|
| 58 |
|
|
|
| 59 |
|
|
try:
|
| 60 |
|
|
# retrive command line options
|
| 61 |
|
|
shortopts = "m:i:t:o:vh?"
|
| 62 |
|
|
longopts = ["methods=", "inputfile=", "inputtrees=", "outputfile=", "verbose", "help", "usage"]
|
| 63 |
|
|
opts, args = getopt.getopt( sys.argv[1:], shortopts, longopts )
|
| 64 |
|
|
|
| 65 |
|
|
except getopt.GetoptError:
|
| 66 |
|
|
# print help information and exit:
|
| 67 |
|
|
print "ERROR: unknown options in argument %s" % sys.argv[1:]
|
| 68 |
|
|
usage()
|
| 69 |
|
|
sys.exit(1)
|
| 70 |
|
|
|
| 71 |
|
|
infname = DEFAULT_INFNAME
|
| 72 |
|
|
treeNameSig = DEFAULT_TREESIG
|
| 73 |
|
|
treeNameBkg = DEFAULT_TREEBKG
|
| 74 |
|
|
outfname = DEFAULT_OUTFNAME
|
| 75 |
|
|
methods = DEFAULT_METHODS
|
| 76 |
|
|
verbose = False
|
| 77 |
|
|
for o, a in opts:
|
| 78 |
|
|
if o in ("-?", "-h", "--help", "--usage"):
|
| 79 |
|
|
usage()
|
| 80 |
|
|
sys.exit(0)
|
| 81 |
|
|
elif o in ("-m", "--methods"):
|
| 82 |
|
|
methods = a
|
| 83 |
|
|
elif o in ("-i", "--inputfile"):
|
| 84 |
|
|
infname = a
|
| 85 |
|
|
elif o in ("-o", "--outputfile"):
|
| 86 |
|
|
outfname = a
|
| 87 |
|
|
elif o in ("-t", "--inputtrees"):
|
| 88 |
|
|
a.strip()
|
| 89 |
|
|
trees = a.rsplit( ' ' )
|
| 90 |
|
|
trees.sort()
|
| 91 |
|
|
trees.reverse()
|
| 92 |
|
|
if len(trees)-trees.count('') != 2:
|
| 93 |
|
|
print "ERROR: need to give two trees (each one for signal and background)"
|
| 94 |
|
|
print trees
|
| 95 |
|
|
sys.exit(1)
|
| 96 |
|
|
treeNameSig = trees[0]
|
| 97 |
|
|
treeNameBkg = trees[1]
|
| 98 |
|
|
elif o in ("-v", "--verbose"):
|
| 99 |
|
|
verbose = True
|
| 100 |
|
|
|
| 101 |
|
|
# print methods
|
| 102 |
|
|
mlist = methods.split(',')
|
| 103 |
|
|
print "=== TMVAnalysis: use methods..."
|
| 104 |
|
|
for m in mlist:
|
| 105 |
|
|
if m != '':
|
| 106 |
|
|
print "=== <%s>" % m
|
| 107 |
|
|
|
| 108 |
|
|
# import ROOT classes
|
| 109 |
|
|
from ROOT import gSystem, gROOT, gApplication, TFile, TTree, TCut
|
| 110 |
|
|
|
| 111 |
|
|
# logon not automatically loaded through PyROOT (logon loads TMVA library) load also GUI
|
| 112 |
|
|
gROOT.Macro( '../macros/TMVAlogon.C' )
|
| 113 |
|
|
gROOT.LoadMacro( '../macros/TMVAGui.C' )
|
| 114 |
|
|
|
| 115 |
|
|
# import TMVA classes from ROOT
|
| 116 |
|
|
from ROOT import TMVA
|
| 117 |
|
|
|
| 118 |
|
|
# output file
|
| 119 |
|
|
outputFile = TFile( outfname, 'RECREATE' )
|
| 120 |
|
|
|
| 121 |
|
|
# create einstance of factory
|
| 122 |
|
|
factory = TMVA.Factory( "TMVAnalysis", outputFile, "Color" )
|
| 123 |
|
|
|
| 124 |
|
|
# set verbosity
|
| 125 |
|
|
factory.SetVerbose( verbose )
|
| 126 |
|
|
|
| 127 |
|
|
# read input data
|
| 128 |
|
|
if not gSystem.AccessPathName( infname ):
|
| 129 |
|
|
input = TFile( infname )
|
| 130 |
|
|
else:
|
| 131 |
|
|
print "ERROR: could not access data file %s\n" % infname
|
| 132 |
|
|
|
| 133 |
|
|
signal = input.Get( treeNameSig )
|
| 134 |
|
|
background = input.Get( treeNameBkg )
|
| 135 |
|
|
|
| 136 |
|
|
# global event weights (see below for setting event-wise weights)
|
| 137 |
|
|
signalWeight = 1.0
|
| 138 |
|
|
backgroundWeight = 1.0
|
| 139 |
|
|
|
| 140 |
|
|
if not factory.SetInputTrees( signal, background, signalWeight, backgroundWeight ):
|
| 141 |
|
|
print "ERROR: could not set input trees\n"
|
| 142 |
|
|
sys.exit(1)
|
| 143 |
|
|
|
| 144 |
|
|
# Define the input variables that shall be used for the classifier training
|
| 145 |
|
|
# note that you may also use variable expressions, such as: "3*var1/var2*abs(var3)"
|
| 146 |
|
|
# [all types of expressions that can also be parsed by TTree::Draw( "expression" )]
|
| 147 |
|
|
factory.AddVariable("var1+var2", 'F')
|
| 148 |
|
|
factory.AddVariable("var1-var2", 'F')
|
| 149 |
|
|
factory.AddVariable("var3", 'F')
|
| 150 |
|
|
factory.AddVariable("var4", 'F')
|
| 151 |
|
|
|
| 152 |
|
|
# This would set individual event weights (the variables defined in the
|
| 153 |
|
|
# expression need to exist in the original TTree)
|
| 154 |
|
|
# factory->SetWeightExpression("weight1*weight2")
|
| 155 |
|
|
#
|
| 156 |
|
|
# Apply additional cuts on the signal and background sample.
|
| 157 |
|
|
# Assumptions on size of training and testing sample:
|
| 158 |
|
|
# a) equal number of signal and background events is used for training
|
| 159 |
|
|
# b) any numbers of signal and background events are used for testing
|
| 160 |
|
|
# c) an explicit syntax can violate a)
|
| 161 |
|
|
# more Documentation with the Factory class
|
| 162 |
|
|
# example for cut: mycut = TCut( "abs(var1)<0.5 && abs(var2-0.5)<1" )
|
| 163 |
|
|
mycut = TCut( "" )
|
| 164 |
|
|
|
| 165 |
|
|
# here, the relevant variables are copied over in new, slim trees that are
|
| 166 |
|
|
# used for TMVA training and testing
|
| 167 |
|
|
# "SplitMode=Random" means that the input events are randomly shuffled before
|
| 168 |
|
|
# splitting them into training and test samples
|
| 169 |
|
|
factory.PrepareTrainingAndTestTree( mycut, "NSigTrain=3000:NBkgTrain=3000:SplitMode=Random:NormMode=NumEvents:!V" )
|
| 170 |
|
|
|
| 171 |
|
|
# and alternative call to use a different number of signal and background training/test event is:
|
| 172 |
|
|
# factory.PrepareTrainingAndTestTree( mycut, "NSigTrain=3000:NBkgTrain=3000:NSigTest=3000:NBkgTest=3000:SplitMode=Random:!V" )
|
| 173 |
|
|
|
| 174 |
|
|
# Cut optimisation
|
| 175 |
|
|
if "Cuts" in mlist:
|
| 176 |
|
|
factory.BookMethod( TMVA.Types.kCuts, "Cuts",
|
| 177 |
|
|
"!H:!V:FitMethod=MC:EffSel:SampleSize=200000:VarProp=FSmart" )
|
| 178 |
|
|
|
| 179 |
|
|
# Cut optimisation using decorrelated input variables
|
| 180 |
|
|
if "CutsD" in mlist:
|
| 181 |
|
|
factory.BookMethod( TMVA.Types.kCuts, "CutsD",
|
| 182 |
|
|
"!H:!V:FitMethod=MC:EffSel:SampleSize=200000:VarProp=FSmart:VarTransform=Decorrelate" )
|
| 183 |
|
|
|
| 184 |
|
|
# Cut optimisation with a Genetic Algorithm
|
| 185 |
|
|
if "CutsGA" in mlist:
|
| 186 |
|
|
factory.BookMethod( TMVA.Types.kCuts, "CutsGA",
|
| 187 |
|
|
"!H:!V:FitMethod=GA:EffSel:Steps=30:Cycles=3:PopSize=100:SC_steps=10:SC_rate=5:SC_factor=0.95:VarProp=FSmart" )
|
| 188 |
|
|
|
| 189 |
|
|
# Likelihood
|
| 190 |
|
|
if "Likelihood" in mlist:
|
| 191 |
|
|
factory.BookMethod( TMVA.Types.kLikelihood, "Likelihood",
|
| 192 |
|
|
"!H:!V:!TransformOutput:PDFInterpol=Spline2:NSmoothSig[0]=100:NSmoothBkg[0]=10:NSmoothBkg[1]=100:NSmooth=10:NAvEvtPerBin=50" )
|
| 193 |
|
|
|
| 194 |
|
|
# test the decorrelated likelihood
|
| 195 |
|
|
if "LikelihoodD" in mlist:
|
| 196 |
|
|
factory.BookMethod( TMVA.Types.kLikelihood, "LikelihoodD",
|
| 197 |
|
|
"!H:!V:!TransformOutput:PDFInterpol=Spline2:NSmoothSig[0]=100:NSmoothBkg[0]=10:NSmooth=5:NAvEvtPerBin=50:VarTransform=Decorrelate" )
|
| 198 |
|
|
|
| 199 |
|
|
if "LikelihoodPCA" in mlist:
|
| 200 |
|
|
factory.BookMethod( TMVA.Types.kLikelihood, "LikelihoodPCA",
|
| 201 |
|
|
"!H:!V:!TransformOutput:PDFInterpol=Spline2:NSmoothSig[0]=100:NSmoothBkg[0]=10:NSmooth=5:NAvEvtPerBin=50:VarTransform=PCA" )
|
| 202 |
|
|
|
| 203 |
|
|
# likelihood method with unbinned kernel estimator
|
| 204 |
|
|
if "LikelihoodKDE" in mlist:
|
| 205 |
|
|
factory.BookMethod( TMVA.Types.kLikelihood, "LikelihoodKDE",
|
| 206 |
|
|
"!H:!V:!TransformOutput:PDFInterpol=KDE:KDEtype=Gauss:KDEiter=Nonadaptive:KDEborder=None:NAvEvtPerBin=50" )
|
| 207 |
|
|
|
| 208 |
|
|
# PDE - RS method
|
| 209 |
|
|
if "PDERS" in mlist:
|
| 210 |
|
|
factory.BookMethod( TMVA.Types.kPDERS, "PDERS",
|
| 211 |
|
|
"!H:!V:VolumeRangeMode=Adaptive:KernelEstimator=Gauss:GaussSigma=0.3:NEventsMin=400:NEventsMax=600:InitialScale=0.99" )
|
| 212 |
|
|
|
| 213 |
|
|
if "PDERSD" in mlist:
|
| 214 |
|
|
factory.BookMethod( TMVA.Types.kPDERS, "PDERSD",
|
| 215 |
|
|
"!H:!V:VolumeRangeMode=Adaptive:KernelEstimator=Gauss:GaussSigma=0.3:NEventsMin=400:NEventsMax=600:InitialScale=0.99:VarTransform=Decorrelate" )
|
| 216 |
|
|
|
| 217 |
|
|
if "PDERSPCA" in mlist:
|
| 218 |
|
|
factory.BookMethod( TMVA.Types.kPDERS, "PDERSPCA",
|
| 219 |
|
|
"!H:!V:VolumeRangeMode=Adaptive:KernelEstimator=Gauss:GaussSigma=0.3:NEventsMin=400:NEventsMax=600:InitialScale=0.99:VarTransform=PCA" )
|
| 220 |
|
|
|
| 221 |
|
|
# HMatrix (chi2-squared) method
|
| 222 |
|
|
if "HMatrix" in mlist:
|
| 223 |
|
|
factory.BookMethod( TMVA.Types.kHMatrix, "HMatrix", "!H:!V" )
|
| 224 |
|
|
|
| 225 |
|
|
# Fisher - also creates PDF for MVA output (here as an example, can be used for any other classifier)
|
| 226 |
|
|
if "Fisher" in mlist:
|
| 227 |
|
|
factory.BookMethod( TMVA.Types.kFisher, "Fisher",
|
| 228 |
|
|
"H:!V:!Normalise:CreateMVAPdfs:Fisher:NbinsMVAPdf=50:NsmoothMVAPdf=1" )
|
| 229 |
|
|
|
| 230 |
|
|
# Function discriminant analysis
|
| 231 |
|
|
if "FDA" in mlist:
|
| 232 |
|
|
factory.BookMethod( TMVA.Types.kFDA,"FDA_MT",
|
| 233 |
|
|
"H:!V:Formula=(0)+(1)*x0+(2)*x1+(3)*x2+(4)*x3:ParRanges=(-1,1);(-10,10);(-10,10);(-10,10);(-10,10):FitMethod=MINUIT:ErrorLevel=1:PrintLevel=-1:FitStrategy=2:UseImprove:UseMinos:SetBatch" )
|
| 234 |
|
|
|
| 235 |
|
|
# the new TMVA ANN: MLP (recommended ANN)
|
| 236 |
|
|
if "MLP" in mlist:
|
| 237 |
|
|
factory.BookMethod( TMVA.Types.kMLP, "MLP", "Normalise:H:!V:NCycles=200:HiddenLayers=N+1,N:TestRate=5" )
|
| 238 |
|
|
|
| 239 |
|
|
# CF(Clermont-Ferrand)ANN
|
| 240 |
|
|
if "CFMlpANN" in mlist:
|
| 241 |
|
|
factory.BookMethod( TMVA.Types.kCFMlpANN, "CFMlpANN", "!H:!V:NCycles=500:HiddenLayers=N+1,N" )
|
| 242 |
|
|
|
| 243 |
|
|
# Tmlp(Root)ANN
|
| 244 |
|
|
if "TMlpANN" in mlist:
|
| 245 |
|
|
factory.BookMethod( TMVA.Types.kTMlpANN, "TMlpANN", "!H:!V:NCycles=200:HiddenLayers=N+1,N" )
|
| 246 |
|
|
|
| 247 |
|
|
# Support Vector Machine with varying kernel functions
|
| 248 |
|
|
if "SVM_Gauss" in mlist:
|
| 249 |
|
|
factory.BookMethod( TMVA.Types.kSVM, "SVM_Gauss", "Sigma=2:C=1:Tol=0.001:Kernel=Gauss" )
|
| 250 |
|
|
|
| 251 |
|
|
if "SVM_Poly" in mlist:
|
| 252 |
|
|
factory.BookMethod( TMVA.Types.kSVM, "SVM_Poly", "Order=4:Theta=1:C=0.1:Tol=0.001:Kernel=Polynomial" )
|
| 253 |
|
|
|
| 254 |
|
|
if "SVM_Lin" in mlist:
|
| 255 |
|
|
factory.BookMethod( TMVA.Types.kSVM, "SVM_Lin", "!H:!V:Kernel=Linear:C=1:Tol=0.001" )
|
| 256 |
|
|
|
| 257 |
|
|
# Boosted Decision Trees
|
| 258 |
|
|
if "BDT" in mlist:
|
| 259 |
|
|
factory.BookMethod( TMVA.Types.kBDT, "BDT",
|
| 260 |
|
|
"!V:NTrees=400:BoostType=AdaBoost:SeparationType=GiniIndex:nCuts=20:PruneMethod=CostComplexity:PruneStrength=4.5")
|
| 261 |
|
|
|
| 262 |
|
|
# Decorrelated Boosted Decision Trees
|
| 263 |
|
|
if "BDTD" in mlist:
|
| 264 |
|
|
factory.BookMethod( TMVA.Types.kBDT, "BDTD",
|
| 265 |
|
|
"!H:!V:NTrees=400:BoostType=AdaBoost:SeparationType=GiniIndex:nCuts=20:PruneMethod=CostComplexity:PruneStrength=4.5" )
|
| 266 |
|
|
|
| 267 |
|
|
# Friedman's RuleFit method
|
| 268 |
|
|
if "RuleFitTMVA" in mlist:
|
| 269 |
|
|
factory.BookMethod( TMVA.Types.kRuleFit, "RuleFitTMVA",
|
| 270 |
|
|
"H:!V:RuleFitModule=RFTMVA:Model=ModRuleLinear:MinImp=0.001:RuleMinDist=0.001:NTrees=20:fEventsMin=0.01:fEventsMax=0.5:GDTau=-1.0:GDTauPrec=0.01:GDStep=0.01:GDNSteps=10000:GDErrScale=1.02" )
|
| 271 |
|
|
|
| 272 |
|
|
if "RuleFitJF" in mlist:
|
| 273 |
|
|
factory.BookMethod( TMVA.Types.kRuleFit, "RuleFitJF",
|
| 274 |
|
|
"!V:RuleFitModule=RFFriedman:Model=ModRuleLinear:GDStep=0.01:GDNSteps=10000:GDErrScale=1.1:RFNendnodes=4" )
|
| 275 |
|
|
|
| 276 |
|
|
# ---- Now you can tell the factory to train, test, and evaluate the MVAs.
|
| 277 |
|
|
|
| 278 |
|
|
# Train MVAs
|
| 279 |
|
|
factory.TrainAllMethods()
|
| 280 |
|
|
|
| 281 |
|
|
# Test MVAs
|
| 282 |
|
|
factory.TestAllMethods()
|
| 283 |
|
|
|
| 284 |
|
|
# Evaluate MVAs
|
| 285 |
|
|
factory.EvaluateAllMethods()
|
| 286 |
|
|
|
| 287 |
|
|
# Save the output.
|
| 288 |
|
|
outputFile.Close()
|
| 289 |
|
|
|
| 290 |
|
|
# clean up
|
| 291 |
|
|
factory.IsA().Destructor( factory )
|
| 292 |
|
|
|
| 293 |
|
|
print "=== wrote root file %s\n" % outfname
|
| 294 |
|
|
print "=== TMVAnalysis is done!\n"
|
| 295 |
|
|
|
| 296 |
|
|
# open the GUI for the result macros
|
| 297 |
|
|
gROOT.ProcessLine( "TMVAGui(\"%s\")" % outfname );
|
| 298 |
|
|
|
| 299 |
|
|
# keep the ROOT thread running
|
| 300 |
|
|
gApplication.Run()
|
| 301 |
|
|
|
| 302 |
|
|
# ----------------------------------------------------------
|
| 303 |
|
|
|
| 304 |
|
|
if __name__ == "__main__":
|
| 305 |
|
|
main()
|
