Logo ROOT  

Reference Guide
 
Loading...
Searching...
No Matches
TMVA_CNN_Classification.C File Reference
Tutorials » TMVA tutorials

Detailed Description

View in nbviewer Open in SWAN
TMVA Classification Example Using a Convolutional Neural Network

This is an example of using a CNN in TMVA. We do classification using a toy image data set that is generated when running the example macro

Running with nthreads = 4
DataSetInfo : [dataset] : Added class "Signal"
: Add Tree sig_tree of type Signal with 1000 events
DataSetInfo : [dataset] : Added class "Background"
: Add Tree bkg_tree of type Background with 1000 events
Factory : Booking method: ␛[1mBDT␛[0m
:
: Rebuilding Dataset dataset
: Building event vectors for type 2 Signal
: Dataset[dataset] : create input formulas for tree sig_tree
: Using variable vars[0] from array expression vars of size 256
: Building event vectors for type 2 Background
: Dataset[dataset] : create input formulas for tree bkg_tree
: Using variable vars[0] from array expression vars of size 256
DataSetFactory : [dataset] : Number of events in input trees
:
:
: Number of training and testing events
: ---------------------------------------------------------------------------
: Signal -- training events : 800
: Signal -- testing events : 200
: Signal -- training and testing events: 1000
: Background -- training events : 800
: Background -- testing events : 200
: Background -- training and testing events: 1000
:
Factory : Booking method: ␛[1mTMVA_DNN_CPU␛[0m
:
: Parsing option string:
: ... "!H:V:ErrorStrategy=CROSSENTROPY:VarTransform=None:WeightInitialization=XAVIER:Layout=DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,DENSE|1|LINEAR:TrainingStrategy=LearningRate=1e-3,Momentum=0.9,Repetitions=1,ConvergenceSteps=5,BatchSize=100,TestRepetitions=1,MaxEpochs=10,WeightDecay=1e-4,Regularization=None,Optimizer=ADAM,DropConfig=0.0+0.0+0.0+0."
: The following options are set:
: - By User:
: <none>
: - Default:
: Boost_num: "0" [Number of times the classifier will be boosted]
: Parsing option string:
: ... "!H:V:ErrorStrategy=CROSSENTROPY:VarTransform=None:WeightInitialization=XAVIER:Layout=DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,DENSE|1|LINEAR:TrainingStrategy=LearningRate=1e-3,Momentum=0.9,Repetitions=1,ConvergenceSteps=5,BatchSize=100,TestRepetitions=1,MaxEpochs=10,WeightDecay=1e-4,Regularization=None,Optimizer=ADAM,DropConfig=0.0+0.0+0.0+0."
: The following options are set:
: - By User:
: V: "True" [Verbose output (short form of "VerbosityLevel" below - overrides the latter one)]
: VarTransform: "None" [List of variable transformations performed before training, e.g., "D_Background,P_Signal,G,N_AllClasses" for: "Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)"]
: H: "False" [Print method-specific help message]
: Layout: "DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,DENSE|1|LINEAR" [Layout of the network.]
: ErrorStrategy: "CROSSENTROPY" [Loss function: Mean squared error (regression) or cross entropy (binary classification).]
: WeightInitialization: "XAVIER" [Weight initialization strategy]
: TrainingStrategy: "LearningRate=1e-3,Momentum=0.9,Repetitions=1,ConvergenceSteps=5,BatchSize=100,TestRepetitions=1,MaxEpochs=10,WeightDecay=1e-4,Regularization=None,Optimizer=ADAM,DropConfig=0.0+0.0+0.0+0." [Defines the training strategies.]
: - Default:
: VerbosityLevel: "Default" [Verbosity level]
: CreateMVAPdfs: "False" [Create PDFs for classifier outputs (signal and background)]
: IgnoreNegWeightsInTraining: "False" [Events with negative weights are ignored in the training (but are included for testing and performance evaluation)]
: InputLayout: "0|0|0" [The Layout of the input]
: BatchLayout: "0|0|0" [The Layout of the batch]
: RandomSeed: "0" [Random seed used for weight initialization and batch shuffling]
: ValidationSize: "20%" [Part of the training data to use for validation. Specify as 0.2 or 20% to use a fifth of the data set as validation set. Specify as 100 to use exactly 100 events. (Default: 20%)]
: Architecture: "CPU" [Which architecture to perform the training on.]
: Multi-core CPU backend not enabled. For better performances, make sure you have a BLAS implementation and it was successfully detected by CMake as well that the imt CMake flag is set.
: Will use anyway the CPU architecture but with slower performance
Factory : ␛[1mTrain all methods␛[0m
Factory : Train method: BDT for Classification
:
BDT : #events: (reweighted) sig: 800 bkg: 800
: #events: (unweighted) sig: 800 bkg: 800
: Training 200 Decision Trees ... patience please
: Elapsed time for training with 1600 events: 0.697 sec
BDT : [dataset] : Evaluation of BDT on training sample (1600 events)
: Elapsed time for evaluation of 1600 events: 0.00709 sec
: Creating xml weight file: ␛[0;36mdataset/weights/TMVA_CNN_Classification_BDT.weights.xml␛[0m
: Creating standalone class: ␛[0;36mdataset/weights/TMVA_CNN_Classification_BDT.class.C␛[0m
: TMVA_CNN_ClassificationOutput.root:/dataset/Method_BDT/BDT
Factory : Training finished
:
Factory : Train method: TMVA_DNN_CPU for Classification
:
: Start of deep neural network training on single thread CPU (without ROOT-MT support)
:
: ***** Deep Learning Network *****
DEEP NEURAL NETWORK: Depth = 8 Input = ( 1, 1, 256 ) Batch size = 100 Loss function = C
Layer 0 DENSE Layer: ( Input = 256 , Width = 100 ) Output = ( 1 , 100 , 100 ) Activation Function = Relu
Layer 1 BATCH NORM Layer: Input/Output = ( 100 , 100 , 1 ) Norm dim = 100 axis = -1
Layer 2 DENSE Layer: ( Input = 100 , Width = 100 ) Output = ( 1 , 100 , 100 ) Activation Function = Relu
Layer 3 BATCH NORM Layer: Input/Output = ( 100 , 100 , 1 ) Norm dim = 100 axis = -1
Layer 4 DENSE Layer: ( Input = 100 , Width = 100 ) Output = ( 1 , 100 , 100 ) Activation Function = Relu
Layer 5 BATCH NORM Layer: Input/Output = ( 100 , 100 , 1 ) Norm dim = 100 axis = -1
Layer 6 DENSE Layer: ( Input = 100 , Width = 100 ) Output = ( 1 , 100 , 100 ) Activation Function = Relu
Layer 7 DENSE Layer: ( Input = 100 , Width = 1 ) Output = ( 1 , 100 , 1 ) Activation Function = Identity
: Using 1280 events for training and 320 for testing
: Compute initial loss on the validation data
: Training phase 1 of 1: Optimizer ADAM (beta1=0.9,beta2=0.999,eps=1e-07) Learning rate = 0.001 regularization 0 minimum error = 201.811
: --------------------------------------------------------------
: Epoch | Train Err. Val. Err. t(s)/epoch t(s)/Loss nEvents/s Conv. Steps
: --------------------------------------------------------------
: Start epoch iteration ...
: 1 Minimum Test error found - save the configuration
: 1 | 0.87816 0.837052 0.139795 0.0118014 9375.49 0
: 2 Minimum Test error found - save the configuration
: 2 | 0.625459 0.722168 0.139271 0.011734 9409 0
: 3 Minimum Test error found - save the configuration
: 3 | 0.548452 0.715515 0.139181 0.0117254 9415.05 0
: 4 Minimum Test error found - save the configuration
: 4 | 0.467966 0.675566 0.139211 0.0117547 9415 0
: 5 Minimum Test error found - save the configuration
: 5 | 0.421829 0.650581 0.139429 0.0116859 9393.85 0
: 6 Minimum Test error found - save the configuration
: 6 | 0.350826 0.649535 0.141258 0.0116725 9260.32 0
: 7 Minimum Test error found - save the configuration
: 7 | 0.307219 0.628235 0.139095 0.0117198 9421.01 0
: 8 | 0.269934 0.678071 0.138469 0.0113004 9436.28 1
: 9 | 0.229846 0.634002 0.139554 0.0113405 9359.37 2
: 10 | 0.195903 0.630256 0.138481 0.0113461 9438.78 3
:
: Elapsed time for training with 1600 events: 1.42 sec
: Evaluate deep neural network on CPU using batches with size = 100
:
TMVA_DNN_CPU : [dataset] : Evaluation of TMVA_DNN_CPU on training sample (1600 events)
: Elapsed time for evaluation of 1600 events: 0.0599 sec
: Creating xml weight file: ␛[0;36mdataset/weights/TMVA_CNN_Classification_TMVA_DNN_CPU.weights.xml␛[0m
: Creating standalone class: ␛[0;36mdataset/weights/TMVA_CNN_Classification_TMVA_DNN_CPU.class.C␛[0m
Factory : Training finished
:
: Ranking input variables (method specific)...
BDT : Ranking result (top variable is best ranked)
: --------------------------------------
: Rank : Variable : Variable Importance
: --------------------------------------
: 1 : vars : 1.321e-02
: 2 : vars : 1.044e-02
: 3 : vars : 9.523e-03
: 4 : vars : 9.083e-03
: 5 : vars : 8.948e-03
: 6 : vars : 8.896e-03
: 7 : vars : 8.787e-03
: 8 : vars : 8.577e-03
: 9 : vars : 8.556e-03
: 10 : vars : 8.303e-03
: 11 : vars : 8.300e-03
: 12 : vars : 8.264e-03
: 13 : vars : 8.011e-03
: 14 : vars : 7.982e-03
: 15 : vars : 7.963e-03
: 16 : vars : 7.711e-03
: 17 : vars : 7.609e-03
: 18 : vars : 7.598e-03
: 19 : vars : 7.517e-03
: 20 : vars : 7.482e-03
: 21 : vars : 7.466e-03
: 22 : vars : 7.456e-03
: 23 : vars : 7.408e-03
: 24 : vars : 7.070e-03
: 25 : vars : 6.969e-03
: 26 : vars : 6.960e-03
: 27 : vars : 6.854e-03
: 28 : vars : 6.723e-03
: 29 : vars : 6.718e-03
: 30 : vars : 6.625e-03
: 31 : vars : 6.605e-03
: 32 : vars : 6.576e-03
: 33 : vars : 6.524e-03
: 34 : vars : 6.391e-03
: 35 : vars : 6.356e-03
: 36 : vars : 6.338e-03
: 37 : vars : 6.332e-03
: 38 : vars : 6.293e-03
: 39 : vars : 6.251e-03
: 40 : vars : 6.239e-03
: 41 : vars : 6.215e-03
: 42 : vars : 6.207e-03
: 43 : vars : 6.197e-03
: 44 : vars : 6.175e-03
: 45 : vars : 6.116e-03
: 46 : vars : 6.091e-03
: 47 : vars : 6.038e-03
: 48 : vars : 6.027e-03
: 49 : vars : 6.015e-03
: 50 : vars : 5.987e-03
: 51 : vars : 5.979e-03
: 52 : vars : 5.962e-03
: 53 : vars : 5.899e-03
: 54 : vars : 5.877e-03
: 55 : vars : 5.869e-03
: 56 : vars : 5.847e-03
: 57 : vars : 5.785e-03
: 58 : vars : 5.684e-03
: 59 : vars : 5.681e-03
: 60 : vars : 5.680e-03
: 61 : vars : 5.625e-03
: 62 : vars : 5.593e-03
: 63 : vars : 5.549e-03
: 64 : vars : 5.523e-03
: 65 : vars : 5.489e-03
: 66 : vars : 5.463e-03
: 67 : vars : 5.426e-03
: 68 : vars : 5.417e-03
: 69 : vars : 5.405e-03
: 70 : vars : 5.353e-03
: 71 : vars : 5.346e-03
: 72 : vars : 5.344e-03
: 73 : vars : 5.272e-03
: 74 : vars : 5.260e-03
: 75 : vars : 5.224e-03
: 76 : vars : 5.191e-03
: 77 : vars : 5.185e-03
: 78 : vars : 5.167e-03
: 79 : vars : 5.161e-03
: 80 : vars : 5.148e-03
: 81 : vars : 5.147e-03
: 82 : vars : 5.139e-03
: 83 : vars : 5.119e-03
: 84 : vars : 5.077e-03
: 85 : vars : 5.077e-03
: 86 : vars : 5.069e-03
: 87 : vars : 5.032e-03
: 88 : vars : 5.016e-03
: 89 : vars : 4.985e-03
: 90 : vars : 4.983e-03
: 91 : vars : 4.956e-03
: 92 : vars : 4.899e-03
: 93 : vars : 4.889e-03
: 94 : vars : 4.869e-03
: 95 : vars : 4.850e-03
: 96 : vars : 4.840e-03
: 97 : vars : 4.828e-03
: 98 : vars : 4.796e-03
: 99 : vars : 4.686e-03
: 100 : vars : 4.679e-03
: 101 : vars : 4.660e-03
: 102 : vars : 4.652e-03
: 103 : vars : 4.650e-03
: 104 : vars : 4.612e-03
: 105 : vars : 4.604e-03
: 106 : vars : 4.530e-03
: 107 : vars : 4.519e-03
: 108 : vars : 4.487e-03
: 109 : vars : 4.462e-03
: 110 : vars : 4.424e-03
: 111 : vars : 4.411e-03
: 112 : vars : 4.398e-03
: 113 : vars : 4.385e-03
: 114 : vars : 4.317e-03
: 115 : vars : 4.314e-03
: 116 : vars : 4.277e-03
: 117 : vars : 4.276e-03
: 118 : vars : 4.231e-03
: 119 : vars : 4.171e-03
: 120 : vars : 4.147e-03
: 121 : vars : 4.140e-03
: 122 : vars : 4.131e-03
: 123 : vars : 4.125e-03
: 124 : vars : 4.071e-03
: 125 : vars : 4.067e-03
: 126 : vars : 4.054e-03
: 127 : vars : 4.016e-03
: 128 : vars : 3.971e-03
: 129 : vars : 3.928e-03
: 130 : vars : 3.880e-03
: 131 : vars : 3.872e-03
: 132 : vars : 3.857e-03
: 133 : vars : 3.847e-03
: 134 : vars : 3.824e-03
: 135 : vars : 3.805e-03
: 136 : vars : 3.790e-03
: 137 : vars : 3.769e-03
: 138 : vars : 3.751e-03
: 139 : vars : 3.749e-03
: 140 : vars : 3.743e-03
: 141 : vars : 3.741e-03
: 142 : vars : 3.727e-03
: 143 : vars : 3.717e-03
: 144 : vars : 3.692e-03
: 145 : vars : 3.681e-03
: 146 : vars : 3.654e-03
: 147 : vars : 3.623e-03
: 148 : vars : 3.544e-03
: 149 : vars : 3.476e-03
: 150 : vars : 3.463e-03
: 151 : vars : 3.413e-03
: 152 : vars : 3.381e-03
: 153 : vars : 3.355e-03
: 154 : vars : 3.351e-03
: 155 : vars : 3.350e-03
: 156 : vars : 3.317e-03
: 157 : vars : 3.301e-03
: 158 : vars : 3.286e-03
: 159 : vars : 3.194e-03
: 160 : vars : 3.152e-03
: 161 : vars : 3.133e-03
: 162 : vars : 3.079e-03
: 163 : vars : 3.065e-03
: 164 : vars : 3.050e-03
: 165 : vars : 3.044e-03
: 166 : vars : 3.024e-03
: 167 : vars : 3.021e-03
: 168 : vars : 3.018e-03
: 169 : vars : 3.005e-03
: 170 : vars : 3.003e-03
: 171 : vars : 2.974e-03
: 172 : vars : 2.943e-03
: 173 : vars : 2.929e-03
: 174 : vars : 2.913e-03
: 175 : vars : 2.893e-03
: 176 : vars : 2.880e-03
: 177 : vars : 2.854e-03
: 178 : vars : 2.817e-03
: 179 : vars : 2.794e-03
: 180 : vars : 2.791e-03
: 181 : vars : 2.777e-03
: 182 : vars : 2.764e-03
: 183 : vars : 2.739e-03
: 184 : vars : 2.730e-03
: 185 : vars : 2.710e-03
: 186 : vars : 2.671e-03
: 187 : vars : 2.608e-03
: 188 : vars : 2.576e-03
: 189 : vars : 2.567e-03
: 190 : vars : 2.463e-03
: 191 : vars : 2.298e-03
: 192 : vars : 2.266e-03
: 193 : vars : 2.260e-03
: 194 : vars : 2.199e-03
: 195 : vars : 2.173e-03
: 196 : vars : 2.144e-03
: 197 : vars : 2.107e-03
: 198 : vars : 2.095e-03
: 199 : vars : 1.987e-03
: 200 : vars : 1.904e-03
: 201 : vars : 1.888e-03
: 202 : vars : 1.880e-03
: 203 : vars : 1.879e-03
: 204 : vars : 1.815e-03
: 205 : vars : 1.810e-03
: 206 : vars : 1.778e-03
: 207 : vars : 1.737e-03
: 208 : vars : 1.703e-03
: 209 : vars : 1.656e-03
: 210 : vars : 1.580e-03
: 211 : vars : 1.519e-03
: 212 : vars : 1.487e-03
: 213 : vars : 1.043e-03
: 214 : vars : 9.647e-04
: 215 : vars : 7.100e-04
: 216 : vars : 6.271e-04
: 217 : vars : 0.000e+00
: 218 : vars : 0.000e+00
: 219 : vars : 0.000e+00
: 220 : vars : 0.000e+00
: 221 : vars : 0.000e+00
: 222 : vars : 0.000e+00
: 223 : vars : 0.000e+00
: 224 : vars : 0.000e+00
: 225 : vars : 0.000e+00
: 226 : vars : 0.000e+00
: 227 : vars : 0.000e+00
: 228 : vars : 0.000e+00
: 229 : vars : 0.000e+00
: 230 : vars : 0.000e+00
: 231 : vars : 0.000e+00
: 232 : vars : 0.000e+00
: 233 : vars : 0.000e+00
: 234 : vars : 0.000e+00
: 235 : vars : 0.000e+00
: 236 : vars : 0.000e+00
: 237 : vars : 0.000e+00
: 238 : vars : 0.000e+00
: 239 : vars : 0.000e+00
: 240 : vars : 0.000e+00
: 241 : vars : 0.000e+00
: 242 : vars : 0.000e+00
: 243 : vars : 0.000e+00
: 244 : vars : 0.000e+00
: 245 : vars : 0.000e+00
: 246 : vars : 0.000e+00
: 247 : vars : 0.000e+00
: 248 : vars : 0.000e+00
: 249 : vars : 0.000e+00
: 250 : vars : 0.000e+00
: 251 : vars : 0.000e+00
: 252 : vars : 0.000e+00
: 253 : vars : 0.000e+00
: 254 : vars : 0.000e+00
: 255 : vars : 0.000e+00
: 256 : vars : 0.000e+00
: --------------------------------------
: No variable ranking supplied by classifier: TMVA_DNN_CPU
TH1.Print Name = TrainingHistory_TMVA_DNN_CPU_trainingError, Entries= 0, Total sum= 4.29559
TH1.Print Name = TrainingHistory_TMVA_DNN_CPU_valError, Entries= 0, Total sum= 6.82098
Factory : === Destroy and recreate all methods via weight files for testing ===
:
: Reading weight file: ␛[0;36mdataset/weights/TMVA_CNN_Classification_BDT.weights.xml␛[0m
: Reading weight file: ␛[0;36mdataset/weights/TMVA_CNN_Classification_TMVA_DNN_CPU.weights.xml␛[0m
Factory : ␛[1mTest all methods␛[0m
Factory : Test method: BDT for Classification performance
:
BDT : [dataset] : Evaluation of BDT on testing sample (400 events)
: Elapsed time for evaluation of 400 events: 0.00197 sec
Factory : Test method: TMVA_DNN_CPU for Classification performance
:
: Evaluate deep neural network on CPU using batches with size = 400
:
TMVA_DNN_CPU : [dataset] : Evaluation of TMVA_DNN_CPU on testing sample (400 events)
: Elapsed time for evaluation of 400 events: 0.0151 sec
Factory : ␛[1mEvaluate all methods␛[0m
Factory : Evaluate classifier: BDT
:
BDT : [dataset] : Loop over test events and fill histograms with classifier response...
:
: Dataset[dataset] : variable plots are not produces ! The number of variables is 256 , it is larger than 200
Factory : Evaluate classifier: TMVA_DNN_CPU
:
TMVA_DNN_CPU : [dataset] : Loop over test events and fill histograms with classifier response...
:
: Evaluate deep neural network on CPU using batches with size = 1000
:
: Dataset[dataset] : variable plots are not produces ! The number of variables is 256 , it is larger than 200
:
: Evaluation results ranked by best signal efficiency and purity (area)
: -------------------------------------------------------------------------------------------------------------------
: DataSet MVA
: Name: Method: ROC-integ
: dataset TMVA_DNN_CPU : 0.730
: dataset BDT : 0.688
: -------------------------------------------------------------------------------------------------------------------
:
: Testing efficiency compared to training efficiency (overtraining check)
: -------------------------------------------------------------------------------------------------------------------
: DataSet MVA Signal efficiency: from test sample (from training sample)
: Name: Method: @B=0.01 @B=0.10 @B=0.30
: -------------------------------------------------------------------------------------------------------------------
: dataset TMVA_DNN_CPU : 0.087 (0.223) 0.305 (0.643) 0.640 (0.837)
: dataset BDT : 0.085 (0.255) 0.348 (0.531) 0.535 (0.759)
: -------------------------------------------------------------------------------------------------------------------
:
Dataset:dataset : Created tree 'TestTree' with 400 events
:
Dataset:dataset : Created tree 'TrainTree' with 1600 events
:
Factory : ␛[1mThank you for using TMVA!␛[0m
: ␛[1mFor citation information, please visit: http://tmva.sf.net/citeTMVA.html␛[0m
/***
# TMVA Classification Example Using a Convolutional Neural Network
**/
/// Helper function to create input images data
/// we create a signal and background 2D histograms from 2d gaussians
/// with a location (means in X and Y) different for each event
/// The difference between signal and background is in the gaussian width.
/// The width for the background gaussian is slightly larger than the signal width by few % values
///
///
void MakeImagesTree(int n, int nh, int nw)
{
// image size (nh x nw)
const int ntot = nh * nw;
const TString fileOutName = TString::Format("images_data_%dx%d.root", nh, nw);
TFile f(fileOutName, "RECREATE");
const int nRndmEvts = 10000; // number of events we use to fill each image
double delta_sigma = 0.1; // 5% difference in the sigma
double pixelNoise = 5;
double sX1 = 3;
double sY1 = 3;
double sX2 = sX1 + delta_sigma;
double sY2 = sY1 - delta_sigma;
TH2D h1("h1", "h1", nh, 0, 10, nw, 0, 10);
TH2D h2("h2", "h2", nh, 0, 10, nw, 0, 10);
TF2 f1("f1", "xygaus");
TF2 f2("f2", "xygaus");
TTree sgn("sig_tree", "signal_tree");
TTree bkg("bkg_tree", "background_tree");
std::vector<float> x1(ntot);
std::vector<float> x2(ntot);
// create signal and background trees with a single branch
// an std::vector<float> of size nh x nw containing the image data
std::vector<float> *px1 = &x1;
std::vector<float> *px2 = &x2;
bkg.Branch("vars", "std::vector<float>", &px1);
sgn.Branch("vars", "std::vector<float>", &px2);
// std::cout << "create tree " << std::endl;
sgn.SetDirectory(&f);
bkg.SetDirectory(&f);
f1.SetParameters(1, 5, sX1, 5, sY1);
f2.SetParameters(1, 5, sX2, 5, sY2);
gRandom->SetSeed(0);
std::cout << "Filling ROOT tree " << std::endl;
for (int i = 0; i < n; ++i) {
if (i % 1000 == 0)
std::cout << "Generating image event ... " << i << std::endl;
h1.Reset();
h2.Reset();
// generate random means in range [3,7] to be not too much on the border
f1.SetParameter(1, gRandom->Uniform(3, 7));
f1.SetParameter(3, gRandom->Uniform(3, 7));
f2.SetParameter(1, gRandom->Uniform(3, 7));
f2.SetParameter(3, gRandom->Uniform(3, 7));
h1.FillRandom("f1", nRndmEvts);
h2.FillRandom("f2", nRndmEvts);
for (int k = 0; k < nh; ++k) {
for (int l = 0; l < nw; ++l) {
int m = k * nw + l;
// add some noise in each bin
x1[m] = h1.GetBinContent(k + 1, l + 1) + gRandom->Gaus(0, pixelNoise);
x2[m] = h2.GetBinContent(k + 1, l + 1) + gRandom->Gaus(0, pixelNoise);
}
}
sgn.Fill();
bkg.Fill();
}
sgn.Write();
bkg.Write();
Info("MakeImagesTree", "Signal and background tree with images data written to the file %s", f.GetName());
sgn.Print();
bkg.Print();
f.Close();
}
/// @brief Run the TMVA CNN Classification example
/// @param nevts : number of signal/background events. Use by default a low value (1000)
/// but increase to at least 5000 to get a good result
/// @param opt : vector of bool with method used (default all on if available). The order is:
/// - TMVA CNN
/// - Keras CNN
/// - TMVA DNN
/// - TMVA BDT
/// - PyTorch CNN
void TMVA_CNN_Classification(int nevts = 1000, std::vector<bool> opt = {1, 1, 1, 1, 1})
{
int imgSize = 16 * 16;
TString inputFileName = "images_data_16x16.root";
bool fileExist = !gSystem->AccessPathName(inputFileName);
// if file does not exists create it
if (!fileExist) {
MakeImagesTree(nevts, 16, 16);
}
bool useTMVACNN = (opt.size() > 0) ? opt[0] : false;
bool useKerasCNN = (opt.size() > 1) ? opt[1] : false;
bool useTMVADNN = (opt.size() > 2) ? opt[2] : false;
bool useTMVABDT = (opt.size() > 3) ? opt[3] : false;
bool usePyTorchCNN = (opt.size() > 4) ? opt[4] : false;
#ifndef R__HAS_TMVACPU
#ifndef R__HAS_TMVAGPU
Warning("TMVA_CNN_Classification",
"TMVA is not build with GPU or CPU multi-thread support. Cannot use TMVA Deep Learning for CNN");
useTMVACNN = false;
#endif
#endif
bool writeOutputFile = true;
#ifdef R__USE_IMT
int num_threads = 4; // use by default 4 threads if value is not set before
// switch off MT in OpenBLAS to avoid conflict with tbb
gSystem->Setenv("OMP_NUM_THREADS", "1");
// do enable MT running
if (num_threads >= 0) {
ROOT::EnableImplicitMT(num_threads);
}
#endif
TMVA::Tools::Instance();
std::cout << "Running with nthreads = " << ROOT::GetThreadPoolSize() << std::endl;
#ifdef R__HAS_PYMVA
gSystem->Setenv("KERAS_BACKEND", "tensorflow");
// for using Keras
TMVA::PyMethodBase::PyInitialize();
#else
useKerasCNN = false;
usePyTorchCNN = false;
#endif
TFile *outputFile = nullptr;
if (writeOutputFile)
outputFile = TFile::Open("TMVA_CNN_ClassificationOutput.root", "RECREATE");
/***
## Create TMVA Factory
Create the Factory class. Later you can choose the methods
whose performance you'd like to investigate.
The factory is the major TMVA object you have to interact with. Here is the list of parameters you need to pass
- The first argument is the base of the name of all the output
weight files in the directory weight/ that will be created with the
method parameters
- The second argument is the output file for the training results
- The third argument is a string option defining some general configuration for the TMVA session.
For example all TMVA output can be suppressed by removing the "!" (not) in front of the "Silent" argument in the
option string
- note that we disable any pre-transformation of the input variables and we avoid computing correlations between
input variables
***/
TMVA::Factory factory(
"TMVA_CNN_Classification", outputFile,
"!V:ROC:!Silent:Color:AnalysisType=Classification:Transformations=None:!Correlations");
/***
## Declare DataLoader(s)
The next step is to declare the DataLoader class that deals with input variables
Define the input variables that shall be used for the MVA training
note that you may also use variable expressions, which can be parsed by TTree::Draw( "expression" )]
In this case the input data consists of an image of 16x16 pixels. Each single pixel is a branch in a ROOT TTree
**/
TMVA::DataLoader loader("dataset");
/***
## Setup Dataset(s)
Define input data file and signal and background trees
**/
std::unique_ptr<TFile> inputFile{TFile::Open(inputFileName)};
if (!inputFile) {
Error("TMVA_CNN_Classification", "Error opening input file %s - exit", inputFileName.Data());
return;
}
// --- Register the training and test trees
auto signalTree = inputFile->Get<TTree>("sig_tree");
auto backgroundTree = inputFile->Get<TTree>("bkg_tree");
if (!signalTree) {
Error("TMVA_CNN_Classification", "Could not find signal tree in file '%s'", inputFileName.Data());
return;
}
if (!backgroundTree) {
Error("TMVA_CNN_Classification", "Could not find background tree in file '%s'", inputFileName.Data());
return;
}
int nEventsSig = signalTree->GetEntries();
int nEventsBkg = backgroundTree->GetEntries();
// global event weights per tree (see below for setting event-wise weights)
Double_t signalWeight = 1.0;
Double_t backgroundWeight = 1.0;
// You can add an arbitrary number of signal or background trees
loader.AddSignalTree(signalTree, signalWeight);
loader.AddBackgroundTree(backgroundTree, backgroundWeight);
/// add event variables (image)
/// use new method (from ROOT 6.20 to add a variable array for all image data)
loader.AddVariablesArray("vars", imgSize);
// Set individual event weights (the variables must exist in the original TTree)
// for signal : factory->SetSignalWeightExpression ("weight1*weight2");
// for background: factory->SetBackgroundWeightExpression("weight1*weight2");
// loader.SetBackgroundWeightExpression( "weight" );
// Apply additional cuts on the signal and background samples (can be different)
TCut mycuts = ""; // for example: TCut mycuts = "abs(var1)<0.5 && abs(var2-0.5)<1";
TCut mycutb = ""; // for example: TCut mycutb = "abs(var1)<0.5";
// Tell the factory how to use the training and testing events
//
// If no numbers of events are given, half of the events in the tree are used
// for training, and the other half for testing:
// loader.PrepareTrainingAndTestTree( mycut, "SplitMode=random:!V" );
// It is possible also to specify the number of training and testing events,
// note we disable the computation of the correlation matrix of the input variables
int nTrainSig = 0.8 * nEventsSig;
int nTrainBkg = 0.8 * nEventsBkg;
// build the string options for DataLoader::PrepareTrainingAndTestTree
TString prepareOptions = TString::Format(
"nTrain_Signal=%d:nTrain_Background=%d:SplitMode=Random:SplitSeed=100:NormMode=NumEvents:!V:!CalcCorrelations",
nTrainSig, nTrainBkg);
loader.PrepareTrainingAndTestTree(mycuts, mycutb, prepareOptions);
/***
DataSetInfo : [dataset] : Added class "Signal"
: Add Tree sig_tree of type Signal with 10000 events
DataSetInfo : [dataset] : Added class "Background"
: Add Tree bkg_tree of type Background with 10000 events
**/
/****
# Booking Methods
Here we book the TMVA methods. We book a Boosted Decision Tree method (BDT)
**/
// Boosted Decision Trees
if (useTMVABDT) {
factory.BookMethod(&loader, TMVA::Types::kBDT, "BDT",
"!V:NTrees=200:MinNodeSize=2.5%:MaxDepth=2:BoostType=AdaBoost:AdaBoostBeta=0.5:"
"UseBaggedBoost:BaggedSampleFraction=0.5:SeparationType=GiniIndex:nCuts=20");
}
/**
#### Booking Deep Neural Network
Here we book the DNN of TMVA. See the example TMVA_Higgs_Classification.C for a detailed description of the
options
**/
if (useTMVADNN) {
TString layoutString(
"Layout=DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,BNORM,DENSE|100|RELU,DENSE|1|LINEAR");
// Training strategies
// one can catenate several training strings with different parameters (e.g. learning rates or regularizations
// parameters) The training string must be concatenates with the `|` delimiter
TString trainingString1("LearningRate=1e-3,Momentum=0.9,Repetitions=1,"
"ConvergenceSteps=5,BatchSize=100,TestRepetitions=1,"
"MaxEpochs=10,WeightDecay=1e-4,Regularization=None,"
"Optimizer=ADAM,DropConfig=0.0+0.0+0.0+0.");
TString trainingStrategyString("TrainingStrategy=");
trainingStrategyString += trainingString1; // + "|" + trainingString2 + ....
// Build now the full DNN Option string
TString dnnOptions("!H:V:ErrorStrategy=CROSSENTROPY:VarTransform=None:"
"WeightInitialization=XAVIER");
dnnOptions.Append(":");
dnnOptions.Append(layoutString);
dnnOptions.Append(":");
dnnOptions.Append(trainingStrategyString);
TString dnnMethodName = "TMVA_DNN_CPU";
// use GPU if available
#ifdef R__HAS_TMVAGPU
dnnOptions += ":Architecture=GPU";
dnnMethodName = "TMVA_DNN_GPU";
#elif defined(R__HAS_TMVACPU)
dnnOptions += ":Architecture=CPU";
#endif
factory.BookMethod(&loader, TMVA::Types::kDL, dnnMethodName, dnnOptions);
}
/***
### Book Convolutional Neural Network in TMVA
For building a CNN one needs to define
- Input Layout : number of channels (in this case = 1) | image height | image width
- Batch Layout : batch size | number of channels | image size = (height*width)
Then one add Convolutional layers and MaxPool layers.
- For Convolutional layer the option string has to be:
- CONV | number of units | filter height | filter width | stride height | stride width | padding height | paddig
width | activation function
- note in this case we are using a filer 3x3 and padding=1 and stride=1 so we get the output dimension of the
conv layer equal to the input
- note we use after the first convolutional layer a batch normalization layer. This seems to help significantly the
convergence
- For the MaxPool layer:
- MAXPOOL | pool height | pool width | stride height | stride width
The RESHAPE layer is needed to flatten the output before the Dense layer
Note that to run the CNN is required to have CPU or GPU support
***/
if (useTMVACNN) {
TString inputLayoutString("InputLayout=1|16|16");
// Batch Layout
TString layoutString("Layout=CONV|10|3|3|1|1|1|1|RELU,BNORM,CONV|10|3|3|1|1|1|1|RELU,MAXPOOL|2|2|1|1,"
"RESHAPE|FLAT,DENSE|100|RELU,DENSE|1|LINEAR");
// Training strategies.
TString trainingString1("LearningRate=1e-3,Momentum=0.9,Repetitions=1,"
"ConvergenceSteps=5,BatchSize=100,TestRepetitions=1,"
"MaxEpochs=10,WeightDecay=1e-4,Regularization=None,"
"Optimizer=ADAM,DropConfig=0.0+0.0+0.0+0.0");
TString trainingStrategyString("TrainingStrategy=");
trainingStrategyString +=
trainingString1; // + "|" + trainingString2 + "|" + trainingString3; for concatenating more training strings
// Build full CNN Options.
TString cnnOptions("!H:V:ErrorStrategy=CROSSENTROPY:VarTransform=None:"
"WeightInitialization=XAVIER");
cnnOptions.Append(":");
cnnOptions.Append(inputLayoutString);
cnnOptions.Append(":");
cnnOptions.Append(layoutString);
cnnOptions.Append(":");
cnnOptions.Append(trainingStrategyString);
//// New DL (CNN)
TString cnnMethodName = "TMVA_CNN_CPU";
// use GPU if available
#ifdef R__HAS_TMVAGPU
cnnOptions += ":Architecture=GPU";
cnnMethodName = "TMVA_CNN_GPU";
#else
cnnOptions += ":Architecture=CPU";
cnnMethodName = "TMVA_CNN_CPU";
#endif
factory.BookMethod(&loader, TMVA::Types::kDL, cnnMethodName, cnnOptions);
}
/**
### Book Convolutional Neural Network in Keras using a generated model
**/
#ifdef R__HAS_PYMVA
// The next section uses Python packages, execute it only if PyMVA is available
TString tmva_python_exe{TMVA::Python_Executable()};
TString python_exe = tmva_python_exe.IsNull() ? "python" : tmva_python_exe;
if (useKerasCNN) {
Info("TMVA_CNN_Classification", "Building convolutional keras model");
// create python script which can be executed
// create 2 conv2d layer + maxpool + dense
TMacro m;
m.AddLine("import tensorflow");
m.AddLine("from tensorflow.keras.models import Sequential");
m.AddLine("from tensorflow.keras.optimizers import Adam");
m.AddLine(
"from tensorflow.keras.layers import Input, Dense, Dropout, Flatten, Conv2D, MaxPooling2D, Reshape, BatchNormalization");
m.AddLine("");
m.AddLine("model = Sequential() ");
m.AddLine("model.add(Reshape((16, 16, 1), input_shape = (256, )))");
m.AddLine("model.add(Conv2D(10, kernel_size = (3, 3), kernel_initializer = 'glorot_normal',activation = "
"'relu', padding = 'same'))");
m.AddLine("model.add(BatchNormalization())");
m.AddLine("model.add(Conv2D(10, kernel_size = (3, 3), kernel_initializer = 'glorot_normal',activation = "
"'relu', padding = 'same'))");
// m.AddLine("model.add(BatchNormalization())");
m.AddLine("model.add(MaxPooling2D(pool_size = (2, 2), strides = (1,1))) ");
m.AddLine("model.add(Flatten())");
m.AddLine("model.add(Dense(256, activation = 'relu')) ");
m.AddLine("model.add(Dense(2, activation = 'sigmoid')) ");
m.AddLine("model.compile(loss = 'binary_crossentropy', optimizer = Adam(learning_rate = 0.001), weighted_metrics = ['accuracy'])");
m.AddLine("model.save('model_cnn.h5')");
m.AddLine("model.summary()");
m.SaveSource("make_cnn_model.py");
// execute
gSystem->Exec(python_exe + " make_cnn_model.py");
if (gSystem->AccessPathName("model_cnn.h5")) {
Warning("TMVA_CNN_Classification", "Error creating Keras model file - skip using Keras");
} else {
// book PyKeras method only if Keras model could be created
Info("TMVA_CNN_Classification", "Booking tf.Keras CNN model");
factory.BookMethod(
&loader, TMVA::Types::kPyKeras, "PyKeras",
"H:!V:VarTransform=None:FilenameModel=model_cnn.h5:tf.keras:"
"FilenameTrainedModel=trained_model_cnn.h5:NumEpochs=10:BatchSize=100:"
"GpuOptions=allow_growth=True"); // needed for RTX NVidia card and to avoid TF allocates all GPU memory
}
}
if (usePyTorchCNN) {
Info("TMVA_CNN_Classification", "Using Convolutional PyTorch Model");
TString pyTorchFileName = gROOT->GetTutorialDir() + TString("/tmva/PyTorch_Generate_CNN_Model.py");
// check that pytorch can be imported and file defining the model and used later when booking the method is
// existing
if (gSystem->Exec(python_exe + " -c 'import torch'") || gSystem->AccessPathName(pyTorchFileName)) {
Warning("TMVA_CNN_Classification", "PyTorch is not installed or model building file is not existing - skip using PyTorch");
} else {
// book PyTorch method only if PyTorch model could be created
Info("TMVA_CNN_Classification", "Booking PyTorch CNN model");
TString methodOpt = "H:!V:VarTransform=None:FilenameModel=PyTorchModelCNN.pt:"
"FilenameTrainedModel=PyTorchTrainedModelCNN.pt:NumEpochs=10:BatchSize=100";
methodOpt += TString(":UserCode=") + pyTorchFileName;
factory.BookMethod(&loader, TMVA::Types::kPyTorch, "PyTorch", methodOpt);
}
}
#endif
//// ## Train Methods
factory.TrainAllMethods();
/// ## Test and Evaluate Methods
factory.TestAllMethods();
factory.EvaluateAllMethods();
/// ## Plot ROC Curve
auto c1 = factory.GetROCCurve(&loader);
c1->Draw();
// close outputfile to save output file
outputFile->Close();
}
f
#define f(i)
Definition RSha256.hxx:104
Double_t
double Double_t
Definition RtypesCore.h:59
Info
void Info(const char *location, const char *msgfmt,...)
Use this function for informational messages.
Definition TError.cxx:218
Error
void Error(const char *location, const char *msgfmt,...)
Use this function in case an error occurred.
Definition TError.cxx:185
Warning
void Warning(const char *location, const char *msgfmt,...)
Use this function in warning situations.
Definition TError.cxx:229
x2
Option_t Option_t TPoint TPoint const char x2
Definition TGWin32VirtualXProxy.cxx:70
x1
Option_t Option_t TPoint TPoint const char x1
Definition TGWin32VirtualXProxy.cxx:70
i
Int_t i
Definition TGraphAsymmErrors.cxx:1306
gROOT
#define gROOT
Definition TROOT.h:414
gRandom
R__EXTERN TRandom * gRandom
Definition TRandom.h:62
gSystem
R__EXTERN TSystem * gSystem
Definition TSystem.h:566
TCut
A specialized string object used for TTree selections.
Definition TCut.h:25
TF2
A 2-Dim function with parameters.
Definition TF2.h:29
TFile
A file, usually with extension .root, that stores data and code in the form of serialized objects in ...
Definition TFile.h:53
TFile::Open
static TFile * Open(const char *name, Option_t *option="", const char *ftitle="", Int_t compress=ROOT::RCompressionSetting::EDefaults::kUseCompiledDefault, Int_t netopt=0)
Create / open a file.
Definition TFile.cxx:4110
TFile::Close
void Close(Option_t *option="") override
Close a file.
Definition TFile.cxx:971
TH2D
2-D histogram with a double per channel (see TH1 documentation)
Definition TH2.h:357
TMVA::Factory
This is the main MVA steering class.
Definition Factory.h:80
TMVA::PyMethodBase::PyInitialize
static void PyInitialize()
Initialize Python interpreter.
Definition PyMethodBase.cxx:153
TMVA::Tools::Instance
static Tools & Instance()
Definition Tools.cxx:71
TMacro
Class supporting a collection of lines with C++ code.
Definition TMacro.h:31
TString
Basic string class.
Definition TString.h:139
TString::Data
const char * Data() const
Definition TString.h:376
TString::IsNull
Bool_t IsNull() const
Definition TString.h:414
TString::Format
static TString Format(const char *fmt,...)
Static method which formats a string using a printf style format descriptor and return a TString.
Definition TString.cxx:2378
TTree
A TTree represents a columnar dataset.
Definition TTree.h:79
TTree::GetEntries
virtual Long64_t GetEntries() const
Definition TTree.h:466
c1
return c1
Definition legend1.C:41
n
const Int_t n
Definition legend1.C:16
h1
TH1F * h1
Definition legend1.C:5
f1
TF1 * f1
Definition legend1.C:11
ROOT::EnableImplicitMT
void EnableImplicitMT(UInt_t numthreads=0)
Enable ROOT's implicit multi-threading for all objects and methods that provide an internal paralleli...
Definition TROOT.cxx:539
ROOT::GetThreadPoolSize
UInt_t GetThreadPoolSize()
Returns the size of ROOT's thread pool.
Definition TROOT.cxx:602
TMVA_CNN_Classification
Definition TMVA_CNN_Classification.py:1
TMVA::Python_Executable
TString Python_Executable()
Function to find current Python executable used by ROOT If "Python3" is installed,...
Definition PyMethodBase.cxx:43
m
TMarker m
Definition textangle.C:8
l
TLine l
Definition textangle.C:4
Author
Lorenzo Moneta

Definition in file TMVA_CNN_Classification.C.