TMVA_RNN_Classification.py File Reference

Detailed Description

TMVA Classification Example Using a Recurrent Neural Network

This is an example of using a RNN in TMVA. We do classification using a toy time dependent data set that is generated when running this example macro

Running with nthreads  = 4
--- RNNClassification  : Using input file: time_data_t10_d30.root
DataSetInfo              : [dataset] : Added class "Signal"
                         : Add Tree sgn of type Signal with 2000 events
DataSetInfo              : [dataset] : Added class "Background"
                         : Add Tree bkg of type Background with 2000 events
number of variables is 300
vars_time0[0]
vars_time0[1]
vars_time0[2]
vars_time0[3]
vars_time0[4]
vars_time0[5]
vars_time0[6]
vars_time0[7]
vars_time0[8]
vars_time0[9]
vars_time0[10]
vars_time0[11]
vars_time0[12]
vars_time0[13]
vars_time0[14]
vars_time0[15]
vars_time0[16]
vars_time0[17]
vars_time0[18]
vars_time0[19]
vars_time0[20]
vars_time0[21]
vars_time0[22]
vars_time0[23]
vars_time0[24]
vars_time0[25]
vars_time0[26]
vars_time0[27]
vars_time0[28]
vars_time0[29]
vars_time1[0]
vars_time1[1]
vars_time1[2]
vars_time1[3]
vars_time1[4]
vars_time1[5]
vars_time1[6]
vars_time1[7]
vars_time1[8]
vars_time1[9]
vars_time1[10]
vars_time1[11]
vars_time1[12]
vars_time1[13]
vars_time1[14]
vars_time1[15]
vars_time1[16]
vars_time1[17]
vars_time1[18]
vars_time1[19]
vars_time1[20]
vars_time1[21]
vars_time1[22]
vars_time1[23]
vars_time1[24]
vars_time1[25]
vars_time1[26]
vars_time1[27]
vars_time1[28]
vars_time1[29]
vars_time2[0]
vars_time2[1]
vars_time2[2]
vars_time2[3]
vars_time2[4]
vars_time2[5]
vars_time2[6]
vars_time2[7]
vars_time2[8]
vars_time2[9]
vars_time2[10]
vars_time2[11]
vars_time2[12]
vars_time2[13]
vars_time2[14]
vars_time2[15]
vars_time2[16]
vars_time2[17]
vars_time2[18]
vars_time2[19]
vars_time2[20]
vars_time2[21]
vars_time2[22]
vars_time2[23]
vars_time2[24]
vars_time2[25]
vars_time2[26]
vars_time2[27]
vars_time2[28]
vars_time2[29]
vars_time3[0]
vars_time3[1]
vars_time3[2]
vars_time3[3]
vars_time3[4]
vars_time3[5]
vars_time3[6]
vars_time3[7]
vars_time3[8]
vars_time3[9]
vars_time3[10]
vars_time3[11]
vars_time3[12]
vars_time3[13]
vars_time3[14]
vars_time3[15]
vars_time3[16]
vars_time3[17]
vars_time3[18]
vars_time3[19]
vars_time3[20]
vars_time3[21]
vars_time3[22]
vars_time3[23]
vars_time3[24]
vars_time3[25]
vars_time3[26]
vars_time3[27]
vars_time3[28]
vars_time3[29]
vars_time4[0]
vars_time4[1]
vars_time4[2]
vars_time4[3]
vars_time4[4]
vars_time4[5]
vars_time4[6]
vars_time4[7]
vars_time4[8]
vars_time4[9]
vars_time4[10]
vars_time4[11]
vars_time4[12]
vars_time4[13]
vars_time4[14]
vars_time4[15]
vars_time4[16]
vars_time4[17]
vars_time4[18]
vars_time4[19]
vars_time4[20]
vars_time4[21]
vars_time4[22]
vars_time4[23]
vars_time4[24]
vars_time4[25]
vars_time4[26]
vars_time4[27]
vars_time4[28]
vars_time4[29]
vars_time5[0]
vars_time5[1]
vars_time5[2]
vars_time5[3]
vars_time5[4]
vars_time5[5]
vars_time5[6]
vars_time5[7]
vars_time5[8]
vars_time5[9]
vars_time5[10]
vars_time5[11]
vars_time5[12]
vars_time5[13]
vars_time5[14]
vars_time5[15]
vars_time5[16]
vars_time5[17]
vars_time5[18]
vars_time5[19]
vars_time5[20]
vars_time5[21]
vars_time5[22]
vars_time5[23]
vars_time5[24]
vars_time5[25]
vars_time5[26]
vars_time5[27]
vars_time5[28]
vars_time5[29]
vars_time6[0]
vars_time6[1]
vars_time6[2]
vars_time6[3]
vars_time6[4]
vars_time6[5]
vars_time6[6]
vars_time6[7]
vars_time6[8]
vars_time6[9]
vars_time6[10]
vars_time6[11]
vars_time6[12]
vars_time6[13]
vars_time6[14]
vars_time6[15]
vars_time6[16]
vars_time6[17]
vars_time6[18]
vars_time6[19]
vars_time6[20]
vars_time6[21]
vars_time6[22]
vars_time6[23]
vars_time6[24]
vars_time6[25]
vars_time6[26]
vars_time6[27]
vars_time6[28]
vars_time6[29]
vars_time7[0]
vars_time7[1]
vars_time7[2]
vars_time7[3]
vars_time7[4]
vars_time7[5]
vars_time7[6]
vars_time7[7]
vars_time7[8]
vars_time7[9]
vars_time7[10]
vars_time7[11]
vars_time7[12]
vars_time7[13]
vars_time7[14]
vars_time7[15]
vars_time7[16]
vars_time7[17]
vars_time7[18]
vars_time7[19]
vars_time7[20]
vars_time7[21]
vars_time7[22]
vars_time7[23]
vars_time7[24]
vars_time7[25]
vars_time7[26]
vars_time7[27]
vars_time7[28]
vars_time7[29]
vars_time8[0]
vars_time8[1]
vars_time8[2]
vars_time8[3]
vars_time8[4]
vars_time8[5]
vars_time8[6]
vars_time8[7]
vars_time8[8]
vars_time8[9]
vars_time8[10]
vars_time8[11]
vars_time8[12]
vars_time8[13]
vars_time8[14]
vars_time8[15]
vars_time8[16]
vars_time8[17]
vars_time8[18]
vars_time8[19]
vars_time8[20]
vars_time8[21]
vars_time8[22]
vars_time8[23]
vars_time8[24]
vars_time8[25]
vars_time8[26]
vars_time8[27]
vars_time8[28]
vars_time8[29]
vars_time9[0]
vars_time9[1]
vars_time9[2]
vars_time9[3]
vars_time9[4]
vars_time9[5]
vars_time9[6]
vars_time9[7]
vars_time9[8]
vars_time9[9]
vars_time9[10]
vars_time9[11]
vars_time9[12]
vars_time9[13]
vars_time9[14]
vars_time9[15]
vars_time9[16]
vars_time9[17]
vars_time9[18]
vars_time9[19]
vars_time9[20]
vars_time9[21]
vars_time9[22]
vars_time9[23]
vars_time9[24]
vars_time9[25]
vars_time9[26]
vars_time9[27]
vars_time9[28]
vars_time9[29]
prepared DATA LOADER 
Factory                  : Booking method: ␛[1mTMVA_DNN␛[0m
                         : 
                         : Parsing option string: 
                         : ... "!H:V:ErrorStrategy=CROSSENTROPY:VarTransform=None:WeightInitialization=XAVIER:RandomSeed=0:InputLayout=1|1|300:Layout=DENSE|64|TANH,DENSE|TANH|64,DENSE|TANH|64,LINEAR:TrainingStrategy=LearningRate=1e-3,Momentum=0.0,Repetitions=1,ConvergenceSteps=10,BatchSize=256,TestRepetitions=1,WeightDecay=1e-4,Regularization=None,MaxEpochs=20DropConfig=0.0+0.+0.+0.,Optimizer=ADAM:CPU"
                         : 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:RandomSeed=0:InputLayout=1|1|300:Layout=DENSE|64|TANH,DENSE|TANH|64,DENSE|TANH|64,LINEAR:TrainingStrategy=LearningRate=1e-3,Momentum=0.0,Repetitions=1,ConvergenceSteps=10,BatchSize=256,TestRepetitions=1,WeightDecay=1e-4,Regularization=None,MaxEpochs=20DropConfig=0.0+0.+0.+0.,Optimizer=ADAM:CPU"
                         : 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]
                         :     InputLayout: "1|1|300" [The Layout of the input]
                         :     Layout: "DENSE|64|TANH,DENSE|TANH|64,DENSE|TANH|64,LINEAR" [Layout of the network.]
                         :     ErrorStrategy: "CROSSENTROPY" [Loss function: Mean squared error (regression) or cross entropy (binary classification).]
                         :     WeightInitialization: "XAVIER" [Weight initialization strategy]
                         :     RandomSeed: "0" [Random seed used for weight initialization and batch shuffling]
                         :     Architecture: "CPU" [Which architecture to perform the training on.]
                         :     TrainingStrategy: "LearningRate=1e-3,Momentum=0.0,Repetitions=1,ConvergenceSteps=10,BatchSize=256,TestRepetitions=1,WeightDecay=1e-4,Regularization=None,MaxEpochs=20DropConfig=0.0+0.+0.+0.,Optimizer=ADAM" [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)]
                         :     BatchLayout: "0|0|0" [The Layout of the batch]
                         :     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%)]
                         : 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                  : Booking method: ␛[1mBDTG␛[0m
                         : 
                         : the option NegWeightTreatment=InverseBoostNegWeights does not exist for BoostType=Grad
                         : --> change to new default NegWeightTreatment=Pray
                         : Rebuilding Dataset dataset
                         : Building event vectors for type 2 Signal
                         : Dataset[dataset] :  create input formulas for tree sgn
                         : Using variable vars_time0[0] from array expression vars_time0 of size 30
                         : Using variable vars_time1[0] from array expression vars_time1 of size 30
                         : Using variable vars_time2[0] from array expression vars_time2 of size 30
                         : Using variable vars_time3[0] from array expression vars_time3 of size 30
                         : Using variable vars_time4[0] from array expression vars_time4 of size 30
                         : Using variable vars_time5[0] from array expression vars_time5 of size 30
                         : Using variable vars_time6[0] from array expression vars_time6 of size 30
                         : Using variable vars_time7[0] from array expression vars_time7 of size 30
                         : Using variable vars_time8[0] from array expression vars_time8 of size 30
                         : Using variable vars_time9[0] from array expression vars_time9 of size 30
                         : Building event vectors for type 2 Background
                         : Dataset[dataset] :  create input formulas for tree bkg
                         : Using variable vars_time0[0] from array expression vars_time0 of size 30
                         : Using variable vars_time1[0] from array expression vars_time1 of size 30
                         : Using variable vars_time2[0] from array expression vars_time2 of size 30
                         : Using variable vars_time3[0] from array expression vars_time3 of size 30
                         : Using variable vars_time4[0] from array expression vars_time4 of size 30
                         : Using variable vars_time5[0] from array expression vars_time5 of size 30
                         : Using variable vars_time6[0] from array expression vars_time6 of size 30
                         : Using variable vars_time7[0] from array expression vars_time7 of size 30
                         : Using variable vars_time8[0] from array expression vars_time8 of size 30
                         : Using variable vars_time9[0] from array expression vars_time9 of size 30
DataSetFactory           : [dataset] : Number of events in input trees
                         : 
                         : 
                         : Number of training and testing events
                         : ---------------------------------------------------------------------------
                         : Signal     -- training events            : 1600
                         : Signal     -- testing events             : 400
                         : Signal     -- training and testing events: 2000
                         : Background -- training events            : 1600
                         : Background -- testing events             : 400
                         : Background -- training and testing events: 2000
                         : 
Factory                  : ␛[1mTrain all methods␛[0m
Factory                  : Train method: TMVA_DNN for Classification
                         : 
                         : Start of deep neural network training on single thread CPU (without ROOT-MT support) 
                         : 
                         : *****   Deep Learning Network *****
DEEP NEURAL NETWORK:   Depth = 4  Input = ( 1, 1, 300 )  Batch size = 256  Loss function = C
   Layer 0   DENSE Layer:   ( Input =   300 , Width =    64 )  Output = (  1 ,   256 ,    64 )   Activation Function = Tanh
   Layer 1   DENSE Layer:   ( Input =    64 , Width =    64 )  Output = (  1 ,   256 ,    64 )   Activation Function = Tanh
   Layer 2   DENSE Layer:   ( Input =    64 , Width =    64 )  Output = (  1 ,   256 ,    64 )   Activation Function = Tanh
   Layer 3   DENSE Layer:   ( Input =    64 , Width =     1 )  Output = (  1 ,   256 ,     1 )   Activation Function = Identity
                         : Using 2560 events for training and 640 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 = 0.7656
                         : --------------------------------------------------------------
                         :      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 |      0.74855    0.701101    0.147684   0.0109064     18716.5           0
                         :          2 Minimum Test error found - save the configuration 
                         :          2 |     0.687883    0.687651    0.146471   0.0108741     18879.6           0
                         :          3 Minimum Test error found - save the configuration 
                         :          3 |     0.678123    0.686159    0.146611    0.010857     18857.6           0
                         :          4 Minimum Test error found - save the configuration 
                         :          4 |     0.670926    0.685687    0.146378   0.0108192     18884.8           0
                         :          5 Minimum Test error found - save the configuration 
                         :          5 |     0.674238    0.667773    0.146747   0.0108595     18839.1           0
                         :          6 Minimum Test error found - save the configuration 
                         :          6 |      0.66401    0.665899    0.146182    0.010862     18918.1           0
                         :          7 |     0.662094    0.677601    0.146111   0.0106974     18905.1           1
                         :          8 |     0.661421    0.678497    0.146204   0.0107164     18894.7           2
                         :          9 Minimum Test error found - save the configuration 
                         :          9 |     0.661585    0.661861    0.146376   0.0108258       18886           0
                         :         10 Minimum Test error found - save the configuration 
                         :         10 |     0.653909    0.659879    0.146027   0.0108265     18934.8           0
                         :         11 Minimum Test error found - save the configuration 
                         :         11 |     0.654896    0.650151    0.146239   0.0108578     18909.6           0
                         :         12 Minimum Test error found - save the configuration 
                         :         12 |     0.661368    0.649572    0.146207   0.0109703     18929.8           0
                         :         13 Minimum Test error found - save the configuration 
                         :         13 |     0.650075    0.643589    0.146258   0.0109251     18916.4           0
                         :         14 |     0.646867    0.643999    0.146208   0.0107007       18892           1
                         :         15 Minimum Test error found - save the configuration 
                         :         15 |     0.639088    0.637761    0.146175   0.0108086     18911.7           0
                         :         16 |     0.648182    0.647835    0.145843   0.0106658     18938.1           1
                         :         17 Minimum Test error found - save the configuration 
                         :         17 |     0.637178    0.627919    0.146155   0.0108707     18923.2           0
                         :         18 |     0.623576    0.643479    0.146011   0.0106509     18912.5           1
                         :         19 Minimum Test error found - save the configuration 
                         :         19 |      0.64489    0.618215    0.145927   0.0108083     18946.3           0
                         :         20 Minimum Test error found - save the configuration 
                         :         20 |     0.625495    0.606319    0.146126   0.0108758     18927.9           0
                         : 
                         : Elapsed time for training with 3200 events: 2.94 sec         
                         : Evaluate deep neural network on CPU using batches with size = 256
                         : 
TMVA_DNN                 : [dataset] : Evaluation of TMVA_DNN on training sample (3200 events)
                         : Elapsed time for evaluation of 3200 events: 0.0732 sec       
                         : Creating xml weight file: ␛[0;36mdataset/weights/TMVAClassification_TMVA_DNN.weights.xml␛[0m
                         : Creating standalone class: ␛[0;36mdataset/weights/TMVAClassification_TMVA_DNN.class.C␛[0m
Factory                  : Training finished
                         : 
Factory                  : Train method: BDTG for Classification
                         : 
                         : 
                         : ␛[1m================================================================␛[0m
                         : ␛[1mH e l p   f o r   M V A   m e t h o d   [ BDTG ] :␛[0m
                         : 
                         : ␛[1m--- Short description:␛[0m
                         : 
                         : Boosted Decision Trees are a collection of individual decision
                         : trees which form a multivariate classifier by (weighted) majority 
                         : vote of the individual trees. Consecutive decision trees are  
                         : trained using the original training data set with re-weighted 
                         : events. By default, the AdaBoost method is employed, which gives 
                         : events that were misclassified in the previous tree a larger 
                         : weight in the training of the following tree.
                         : 
                         : Decision trees are a sequence of binary splits of the data sample
                         : using a single discriminant variable at a time. A test event 
                         : ending up after the sequence of left-right splits in a final 
                         : ("leaf") node is classified as either signal or background
                         : depending on the majority type of training events in that node.
                         : 
                         : ␛[1m--- Performance optimisation:␛[0m
                         : 
                         : By the nature of the binary splits performed on the individual
                         : variables, decision trees do not deal well with linear correlations
                         : between variables (they need to approximate the linear split in
                         : the two dimensional space by a sequence of splits on the two 
                         : variables individually). Hence decorrelation could be useful 
                         : to optimise the BDT performance.
                         : 
                         : ␛[1m--- Performance tuning via configuration options:␛[0m
                         : 
                         : The two most important parameters in the configuration are the  
                         : minimal number of events requested by a leaf node as percentage of the 
                         :    number of training events (option "MinNodeSize"  replacing the actual number 
                         :  of events "nEventsMin" as given in earlier versions
                         : If this number is too large, detailed features 
                         : in the parameter space are hard to be modelled. If it is too small, 
                         : the risk to overtrain rises and boosting seems to be less effective
                         :   typical values from our current experience for best performance  
                         :   are between 0.5(%) and 10(%) 
                         : 
                         : The default minimal number is currently set to 
                         :    max(20, (N_training_events / N_variables^2 / 10)) 
                         : and can be changed by the user.
                         : 
                         : The other crucial parameter, the pruning strength ("PruneStrength"),
                         : is also related to overtraining. It is a regularisation parameter 
                         : that is used when determining after the training which splits 
                         : are considered statistically insignificant and are removed. The
                         : user is advised to carefully watch the BDT screen output for
                         : the comparison between efficiencies obtained on the training and
                         : the independent test sample. They should be equal within statistical
                         : errors, in order to minimize statistical fluctuations in different samples.
                         : 
                         : <Suppress this message by specifying "!H" in the booking option>
                         : ␛[1m================================================================␛[0m
                         : 
BDTG                     : #events: (reweighted) sig: 1600 bkg: 1600
                         : #events: (unweighted) sig: 1600 bkg: 1600
                         : Training 100 Decision Trees ... patience please
                         : Elapsed time for training with 3200 events: 0.763 sec         
BDTG                     : [dataset] : Evaluation of BDTG on training sample (3200 events)
                         : Elapsed time for evaluation of 3200 events: 0.00893 sec       
                         : Creating xml weight file: ␛[0;36mdataset/weights/TMVAClassification_BDTG.weights.xml␛[0m
                         : Creating standalone class: ␛[0;36mdataset/weights/TMVAClassification_BDTG.class.C␛[0m
                         : data_RNN_CPU.root:/dataset/Method_BDT/BDTG
Factory                  : Training finished
                         : 
                         : Ranking input variables (method specific)...
                         : No variable ranking supplied by classifier: TMVA_DNN
BDTG                     : Ranking result (top variable is best ranked)
                         : --------------------------------------------
                         : Rank : Variable   : Variable Importance
                         : --------------------------------------------
                         :    1 : vars_time8 : 2.113e-02
                         :    2 : vars_time8 : 2.031e-02
                         :    3 : vars_time9 : 1.993e-02
                         :    4 : vars_time7 : 1.947e-02
                         :    5 : vars_time9 : 1.904e-02
                         :    6 : vars_time7 : 1.887e-02
                         :    7 : vars_time8 : 1.868e-02
                         :    8 : vars_time6 : 1.795e-02
                         :    9 : vars_time7 : 1.782e-02
                         :   10 : vars_time0 : 1.759e-02
                         :   11 : vars_time7 : 1.716e-02
                         :   12 : vars_time9 : 1.693e-02
                         :   13 : vars_time9 : 1.623e-02
                         :   14 : vars_time8 : 1.585e-02
                         :   15 : vars_time8 : 1.575e-02
                         :   16 : vars_time7 : 1.517e-02
                         :   17 : vars_time9 : 1.466e-02
                         :   18 : vars_time6 : 1.463e-02
                         :   19 : vars_time7 : 1.427e-02
                         :   20 : vars_time7 : 1.340e-02
                         :   21 : vars_time7 : 1.292e-02
                         :   22 : vars_time6 : 1.260e-02
                         :   23 : vars_time9 : 1.240e-02
                         :   24 : vars_time7 : 1.196e-02
                         :   25 : vars_time0 : 1.169e-02
                         :   26 : vars_time5 : 1.153e-02
                         :   27 : vars_time8 : 1.138e-02
                         :   28 : vars_time5 : 1.136e-02
                         :   29 : vars_time9 : 1.123e-02
                         :   30 : vars_time6 : 1.097e-02
                         :   31 : vars_time8 : 1.088e-02
                         :   32 : vars_time4 : 1.053e-02
                         :   33 : vars_time8 : 1.047e-02
                         :   34 : vars_time9 : 1.035e-02
                         :   35 : vars_time8 : 1.015e-02
                         :   36 : vars_time0 : 9.862e-03
                         :   37 : vars_time9 : 9.791e-03
                         :   38 : vars_time5 : 9.668e-03
                         :   39 : vars_time6 : 9.559e-03
                         :   40 : vars_time8 : 9.537e-03
                         :   41 : vars_time9 : 9.334e-03
                         :   42 : vars_time5 : 9.297e-03
                         :   43 : vars_time9 : 9.221e-03
                         :   44 : vars_time5 : 9.204e-03
                         :   45 : vars_time6 : 9.193e-03
                         :   46 : vars_time6 : 9.048e-03
                         :   47 : vars_time6 : 8.997e-03
                         :   48 : vars_time8 : 8.845e-03
                         :   49 : vars_time5 : 8.785e-03
                         :   50 : vars_time9 : 8.631e-03
                         :   51 : vars_time0 : 8.490e-03
                         :   52 : vars_time6 : 8.480e-03
                         :   53 : vars_time4 : 8.447e-03
                         :   54 : vars_time8 : 8.394e-03
                         :   55 : vars_time7 : 8.352e-03
                         :   56 : vars_time1 : 8.067e-03
                         :   57 : vars_time8 : 8.042e-03
                         :   58 : vars_time8 : 8.020e-03
                         :   59 : vars_time7 : 7.783e-03
                         :   60 : vars_time8 : 7.715e-03
                         :   61 : vars_time1 : 7.691e-03
                         :   62 : vars_time6 : 7.555e-03
                         :   63 : vars_time7 : 7.337e-03
                         :   64 : vars_time9 : 7.151e-03
                         :   65 : vars_time3 : 6.949e-03
                         :   66 : vars_time7 : 6.821e-03
                         :   67 : vars_time0 : 6.677e-03
                         :   68 : vars_time9 : 6.427e-03
                         :   69 : vars_time9 : 6.409e-03
                         :   70 : vars_time1 : 6.399e-03
                         :   71 : vars_time6 : 6.358e-03
                         :   72 : vars_time8 : 6.346e-03
                         :   73 : vars_time4 : 6.336e-03
                         :   74 : vars_time3 : 6.254e-03
                         :   75 : vars_time2 : 6.088e-03
                         :   76 : vars_time9 : 6.057e-03
                         :   77 : vars_time7 : 6.041e-03
                         :   78 : vars_time0 : 5.930e-03
                         :   79 : vars_time5 : 5.827e-03
                         :   80 : vars_time6 : 5.790e-03
                         :   81 : vars_time1 : 5.679e-03
                         :   82 : vars_time4 : 5.609e-03
                         :   83 : vars_time2 : 5.575e-03
                         :   84 : vars_time6 : 5.540e-03
                         :   85 : vars_time5 : 5.483e-03
                         :   86 : vars_time0 : 5.446e-03
                         :   87 : vars_time4 : 5.412e-03
                         :   88 : vars_time7 : 5.350e-03
                         :   89 : vars_time0 : 5.309e-03
                         :   90 : vars_time8 : 5.184e-03
                         :   91 : vars_time7 : 5.129e-03
                         :   92 : vars_time8 : 4.651e-03
                         :   93 : vars_time0 : 4.613e-03
                         :   94 : vars_time0 : 4.400e-03
                         :   95 : vars_time7 : 4.355e-03
                         :   96 : vars_time0 : 4.290e-03
                         :   97 : vars_time1 : 4.284e-03
                         :   98 : vars_time3 : 4.229e-03
                         :   99 : vars_time2 : 4.153e-03
                         :  100 : vars_time1 : 3.825e-03
                         :  101 : vars_time5 : 3.686e-03
                         :  102 : vars_time4 : 3.685e-03
                         :  103 : vars_time2 : 3.636e-03
                         :  104 : vars_time4 : 3.526e-03
                         :  105 : vars_time6 : 3.396e-03
                         :  106 : vars_time9 : 3.347e-03
                         :  107 : vars_time5 : 3.101e-03
                         :  108 : vars_time7 : 2.871e-03
                         :  109 : vars_time4 : 2.804e-03
                         :  110 : vars_time3 : 2.755e-03
                         :  111 : vars_time6 : 2.228e-03
                         :  112 : vars_time0 : 0.000e+00
                         :  113 : vars_time0 : 0.000e+00
                         :  114 : vars_time0 : 0.000e+00
                         :  115 : vars_time0 : 0.000e+00
                         :  116 : vars_time0 : 0.000e+00
                         :  117 : vars_time0 : 0.000e+00
                         :  118 : vars_time0 : 0.000e+00
                         :  119 : vars_time0 : 0.000e+00
                         :  120 : vars_time0 : 0.000e+00
                         :  121 : vars_time0 : 0.000e+00
                         :  122 : vars_time0 : 0.000e+00
                         :  123 : vars_time0 : 0.000e+00
                         :  124 : vars_time0 : 0.000e+00
                         :  125 : vars_time0 : 0.000e+00
                         :  126 : vars_time0 : 0.000e+00
                         :  127 : vars_time0 : 0.000e+00
                         :  128 : vars_time0 : 0.000e+00
                         :  129 : vars_time0 : 0.000e+00
                         :  130 : vars_time0 : 0.000e+00
                         :  131 : vars_time1 : 0.000e+00
                         :  132 : vars_time1 : 0.000e+00
                         :  133 : vars_time1 : 0.000e+00
                         :  134 : vars_time1 : 0.000e+00
                         :  135 : vars_time1 : 0.000e+00
                         :  136 : vars_time1 : 0.000e+00
                         :  137 : vars_time1 : 0.000e+00
                         :  138 : vars_time1 : 0.000e+00
                         :  139 : vars_time1 : 0.000e+00
                         :  140 : vars_time1 : 0.000e+00
                         :  141 : vars_time1 : 0.000e+00
                         :  142 : vars_time1 : 0.000e+00
                         :  143 : vars_time1 : 0.000e+00
                         :  144 : vars_time1 : 0.000e+00
                         :  145 : vars_time1 : 0.000e+00
                         :  146 : vars_time1 : 0.000e+00
                         :  147 : vars_time1 : 0.000e+00
                         :  148 : vars_time1 : 0.000e+00
                         :  149 : vars_time1 : 0.000e+00
                         :  150 : vars_time1 : 0.000e+00
                         :  151 : vars_time1 : 0.000e+00
                         :  152 : vars_time1 : 0.000e+00
                         :  153 : vars_time1 : 0.000e+00
                         :  154 : vars_time1 : 0.000e+00
                         :  155 : vars_time2 : 0.000e+00
                         :  156 : vars_time2 : 0.000e+00
                         :  157 : vars_time2 : 0.000e+00
                         :  158 : vars_time2 : 0.000e+00
                         :  159 : vars_time2 : 0.000e+00
                         :  160 : vars_time2 : 0.000e+00
                         :  161 : vars_time2 : 0.000e+00
                         :  162 : vars_time2 : 0.000e+00
                         :  163 : vars_time2 : 0.000e+00
                         :  164 : vars_time2 : 0.000e+00
                         :  165 : vars_time2 : 0.000e+00
                         :  166 : vars_time2 : 0.000e+00
                         :  167 : vars_time2 : 0.000e+00
                         :  168 : vars_time2 : 0.000e+00
                         :  169 : vars_time2 : 0.000e+00
                         :  170 : vars_time2 : 0.000e+00
                         :  171 : vars_time2 : 0.000e+00
                         :  172 : vars_time2 : 0.000e+00
                         :  173 : vars_time2 : 0.000e+00
                         :  174 : vars_time2 : 0.000e+00
                         :  175 : vars_time2 : 0.000e+00
                         :  176 : vars_time2 : 0.000e+00
                         :  177 : vars_time2 : 0.000e+00
                         :  178 : vars_time2 : 0.000e+00
                         :  179 : vars_time2 : 0.000e+00
                         :  180 : vars_time2 : 0.000e+00
                         :  181 : vars_time3 : 0.000e+00
                         :  182 : vars_time3 : 0.000e+00
                         :  183 : vars_time3 : 0.000e+00
                         :  184 : vars_time3 : 0.000e+00
                         :  185 : vars_time3 : 0.000e+00
                         :  186 : vars_time3 : 0.000e+00
                         :  187 : vars_time3 : 0.000e+00
                         :  188 : vars_time3 : 0.000e+00
                         :  189 : vars_time3 : 0.000e+00
                         :  190 : vars_time3 : 0.000e+00
                         :  191 : vars_time3 : 0.000e+00
                         :  192 : vars_time3 : 0.000e+00
                         :  193 : vars_time3 : 0.000e+00
                         :  194 : vars_time3 : 0.000e+00
                         :  195 : vars_time3 : 0.000e+00
                         :  196 : vars_time3 : 0.000e+00
                         :  197 : vars_time3 : 0.000e+00
                         :  198 : vars_time3 : 0.000e+00
                         :  199 : vars_time3 : 0.000e+00
                         :  200 : vars_time3 : 0.000e+00
                         :  201 : vars_time3 : 0.000e+00
                         :  202 : vars_time3 : 0.000e+00
                         :  203 : vars_time3 : 0.000e+00
                         :  204 : vars_time3 : 0.000e+00
                         :  205 : vars_time3 : 0.000e+00
                         :  206 : vars_time3 : 0.000e+00
                         :  207 : vars_time4 : 0.000e+00
                         :  208 : vars_time4 : 0.000e+00
                         :  209 : vars_time4 : 0.000e+00
                         :  210 : vars_time4 : 0.000e+00
                         :  211 : vars_time4 : 0.000e+00
                         :  212 : vars_time4 : 0.000e+00
                         :  213 : vars_time4 : 0.000e+00
                         :  214 : vars_time4 : 0.000e+00
                         :  215 : vars_time4 : 0.000e+00
                         :  216 : vars_time4 : 0.000e+00
                         :  217 : vars_time4 : 0.000e+00
                         :  218 : vars_time4 : 0.000e+00
                         :  219 : vars_time4 : 0.000e+00
                         :  220 : vars_time4 : 0.000e+00
                         :  221 : vars_time4 : 0.000e+00
                         :  222 : vars_time4 : 0.000e+00
                         :  223 : vars_time4 : 0.000e+00
                         :  224 : vars_time4 : 0.000e+00
                         :  225 : vars_time4 : 0.000e+00
                         :  226 : vars_time4 : 0.000e+00
                         :  227 : vars_time4 : 0.000e+00
                         :  228 : vars_time4 : 0.000e+00
                         :  229 : vars_time5 : 0.000e+00
                         :  230 : vars_time5 : 0.000e+00
                         :  231 : vars_time5 : 0.000e+00
                         :  232 : vars_time5 : 0.000e+00
                         :  233 : vars_time5 : 0.000e+00
                         :  234 : vars_time5 : 0.000e+00
                         :  235 : vars_time5 : 0.000e+00
                         :  236 : vars_time5 : 0.000e+00
                         :  237 : vars_time5 : 0.000e+00
                         :  238 : vars_time5 : 0.000e+00
                         :  239 : vars_time5 : 0.000e+00
                         :  240 : vars_time5 : 0.000e+00
                         :  241 : vars_time5 : 0.000e+00
                         :  242 : vars_time5 : 0.000e+00
                         :  243 : vars_time5 : 0.000e+00
                         :  244 : vars_time5 : 0.000e+00
                         :  245 : vars_time5 : 0.000e+00
                         :  246 : vars_time5 : 0.000e+00
                         :  247 : vars_time5 : 0.000e+00
                         :  248 : vars_time5 : 0.000e+00
                         :  249 : vars_time6 : 0.000e+00
                         :  250 : vars_time6 : 0.000e+00
                         :  251 : vars_time6 : 0.000e+00
                         :  252 : vars_time6 : 0.000e+00
                         :  253 : vars_time6 : 0.000e+00
                         :  254 : vars_time6 : 0.000e+00
                         :  255 : vars_time6 : 0.000e+00
                         :  256 : vars_time6 : 0.000e+00
                         :  257 : vars_time6 : 0.000e+00
                         :  258 : vars_time6 : 0.000e+00
                         :  259 : vars_time6 : 0.000e+00
                         :  260 : vars_time6 : 0.000e+00
                         :  261 : vars_time6 : 0.000e+00
                         :  262 : vars_time6 : 0.000e+00
                         :  263 : vars_time6 : 0.000e+00
                         :  264 : vars_time7 : 0.000e+00
                         :  265 : vars_time7 : 0.000e+00
                         :  266 : vars_time7 : 0.000e+00
                         :  267 : vars_time7 : 0.000e+00
                         :  268 : vars_time7 : 0.000e+00
                         :  269 : vars_time7 : 0.000e+00
                         :  270 : vars_time7 : 0.000e+00
                         :  271 : vars_time7 : 0.000e+00
                         :  272 : vars_time7 : 0.000e+00
                         :  273 : vars_time7 : 0.000e+00
                         :  274 : vars_time7 : 0.000e+00
                         :  275 : vars_time7 : 0.000e+00
                         :  276 : vars_time8 : 0.000e+00
                         :  277 : vars_time8 : 0.000e+00
                         :  278 : vars_time8 : 0.000e+00
                         :  279 : vars_time8 : 0.000e+00
                         :  280 : vars_time8 : 0.000e+00
                         :  281 : vars_time8 : 0.000e+00
                         :  282 : vars_time8 : 0.000e+00
                         :  283 : vars_time8 : 0.000e+00
                         :  284 : vars_time8 : 0.000e+00
                         :  285 : vars_time8 : 0.000e+00
                         :  286 : vars_time8 : 0.000e+00
                         :  287 : vars_time8 : 0.000e+00
                         :  288 : vars_time9 : 0.000e+00
                         :  289 : vars_time9 : 0.000e+00
                         :  290 : vars_time9 : 0.000e+00
                         :  291 : vars_time9 : 0.000e+00
                         :  292 : vars_time9 : 0.000e+00
                         :  293 : vars_time9 : 0.000e+00
                         :  294 : vars_time9 : 0.000e+00
                         :  295 : vars_time9 : 0.000e+00
                         :  296 : vars_time9 : 0.000e+00
                         :  297 : vars_time9 : 0.000e+00
                         :  298 : vars_time9 : 0.000e+00
                         :  299 : vars_time9 : 0.000e+00
                         :  300 : vars_time9 : 0.000e+00
                         : --------------------------------------------
TH1.Print Name  = TrainingHistory_TMVA_DNN_trainingError, Entries= 0, Total sum= 13.1944
TH1.Print Name  = TrainingHistory_TMVA_DNN_valError, Entries= 0, Total sum= 13.1409
Factory                  : === Destroy and recreate all methods via weight files for testing ===
                         : 
                         : Reading weight file: ␛[0;36mdataset/weights/TMVAClassification_TMVA_DNN.weights.xml␛[0m
                         : Reading weight file: ␛[0;36mdataset/weights/TMVAClassification_BDTG.weights.xml␛[0m
nthreads  = 4
Factory                  : ␛[1mTest all methods␛[0m
Factory                  : Test method: TMVA_DNN for Classification performance
                         : 
                         : Evaluate deep neural network on CPU using batches with size = 800
                         : 
TMVA_DNN                 : [dataset] : Evaluation of TMVA_DNN on testing sample (800 events)
                         : Elapsed time for evaluation of 800 events: 0.0171 sec       
Factory                  : Test method: BDTG for Classification performance
                         : 
BDTG                     : [dataset] : Evaluation of BDTG on testing sample (800 events)
                         : Elapsed time for evaluation of 800 events: 0.00203 sec       
Factory                  : ␛[1mEvaluate all methods␛[0m
Factory                  : Evaluate classifier: TMVA_DNN
                         : 
TMVA_DNN                 : [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 300 , it is larger than 200
Factory                  : Evaluate classifier: BDTG
                         : 
BDTG                     : [dataset] : Loop over test events and fill histograms with classifier response...
                         : 
                         : Dataset[dataset] :  variable plots are not produces ! The number of variables is 300 , it is larger than 200
                         : 
                         : Evaluation results ranked by best signal efficiency and purity (area)
                         : -------------------------------------------------------------------------------------------------------------------
                         : DataSet       MVA                       
                         : Name:         Method:          ROC-integ
                         : dataset       BDTG           : 0.850
                         : dataset       TMVA_DNN       : 0.685
                         : -------------------------------------------------------------------------------------------------------------------
                         : 
                         : 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              BDTG           : 0.245 (0.340)       0.573 (0.654)      0.847 (0.867)
                         : dataset              TMVA_DNN       : 0.029 (0.035)       0.247 (0.305)      0.582 (0.596)
                         : -------------------------------------------------------------------------------------------------------------------
                         : 
Dataset:dataset          : Created tree 'TestTree' with 800 events
                         : 
Dataset:dataset          : Created tree 'TrainTree' with 3200 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 Recurrent Neural Network
 
# This is an example of using a RNN in TMVA.
# We do the classification using a toy data set containing a time series of data sample ntimes
# and with dimension ndim that is generated when running the provided function `MakeTimeData (nevents, ntime, ndim)`
 
 
import ROOT
 
num_threads = 4  # use max 4 threads
# do enable MT running
if "imt" in ROOT.gROOT.GetConfigFeatures():
    ROOT.EnableImplicitMT(num_threads)
    # switch off MT in OpenBLAS to avoid conflict with tbb
    ROOT.gSystem.Setenv("OMP_NUM_THREADS", "1")
    print("Running with nthreads  = {}".format(ROOT.GetThreadPoolSize()))
else:
    print("Running in serial mode since ROOT does not support MT")
 
 
TMVA = ROOT.TMVA
TFile = ROOT.TFile
 
import os
import importlib
 
 
TMVA.Tools.Instance()
TMVA.Config.Instance()
 
 
##  Helper function to generate the time data set
##  make some time data but not of fixed length.
##  use a poisson with mu = 5 and truncated at 10
 
 
def MakeTimeData(n, ntime, ndim):
    # ntime = 10;
    # ndim = 30; // number of dim/time
 
    fname = "time_data_t" + str(ntime) + "_d" + str(ndim) + ".root"
    v1 = []
    v2 = []
 
    for i in range(ntime):
        v1.append(ROOT.TH1D("h1_" + str(i), "h1", ndim, 0, 10))
        v2.append(ROOT.TH1D("h2_" + str(i), "h2", ndim, 0, 10))
 
    f1 = ROOT.TF1("f1", "gaus")
    f2 = ROOT.TF1("f2", "gaus")
 
    sgn = ROOT.TTree("sgn", "sgn")
    bkg = ROOT.TTree("bkg", "bkg")
    f = TFile(fname, "RECREATE")
 
    x1 = []
    x2 = []
 
    for i in range(ntime):
        x1.append(ROOT.std.vector["float"](ndim))
        x2.append(ROOT.std.vector["float"](ndim))
 
    for i in range(ntime):
        bkg.Branch("vars_time" + str(i), "std::vector<float>", x1[i])
        sgn.Branch("vars_time" + str(i), "std::vector<float>", x2[i])
 
    sgn.SetDirectory(f)
    bkg.SetDirectory(f)
    ROOT.gRandom.SetSeed(0)
 
    mean1 = ROOT.std.vector["double"](ntime)
    mean2 = ROOT.std.vector["double"](ntime)
    sigma1 = ROOT.std.vector["double"](ntime)
    sigma2 = ROOT.std.vector["double"](ntime)
 
    for j in range(ntime):
        mean1[j] = 5.0 + 0.2 * ROOT.TMath.Sin(ROOT.TMath.Pi() * j / float(ntime))
        mean2[j] = 5.0 + 0.2 * ROOT.TMath.Cos(ROOT.TMath.Pi() * j / float(ntime))
        sigma1[j] = 4 + 0.3 * ROOT.TMath.Sin(ROOT.TMath.Pi() * j / float(ntime))
        sigma2[j] = 4 + 0.3 * ROOT.TMath.Cos(ROOT.TMath.Pi() * j / float(ntime))
 
    for i in range(n):
        if i % 1000 == 0:
            print("Generating  event ... %d", i)
 
        for j in range(ntime):
            h1 = v1[j]
            h2 = v2[j]
            h1.Reset()
            h2.Reset()
 
            f1.SetParameters(1, mean1[j], sigma1[j])
            f2.SetParameters(1, mean2[j], sigma2[j])
 
            h1.FillRandom("f1", 1000)
            h2.FillRandom("f2", 1000)
 
            for k in range(ntime):
                # std::cout << j*10+k << "   ";
                x1[j][k] = h1.GetBinContent(k + 1) + ROOT.gRandom.Gaus(0, 10)
                x2[j][k] = h2.GetBinContent(k + 1) + ROOT.gRandom.Gaus(0, 10)
 
        sgn.Fill()
        bkg.Fill()
 
        if n == 1:
            c1 = ROOT.TCanvas()
            c1.Divide(ntime, 2)
            for j in range(ntime):
                c1.cd(j + 1)
                v1[j].Draw()
            for j in range(ntime):
                c1.cd(ntime + j + 1)
                v2[j].Draw()
 
            ROOT.gPad.Update()
 
    if n > 1:
        sgn.Write()
        bkg.Write()
        sgn.Print()
        bkg.Print()
        f.Close()
 
 
## macro for performing a classification using a Recurrent Neural Network
## @param use_type
##    use_type = 0    use Simple RNN network
##    use_type = 1    use LSTM network
##    use_type = 2    use GRU
##    use_type = 3    build 3 different networks with RNN, LSTM and GRU
 
 
use_type = 1
ninput = 30
ntime = 10
batchSize = 100
maxepochs = 10
 
nTotEvts = 2000  # total events to be generated for signal or background
 
useKeras = False
 
useTMVA_RNN = True
useTMVA_DNN = True
useTMVA_BDT = False
 
if ROOT.gSystem.GetFromPipe("root-config --has-tmva-pymva") == "yes":
    useKeras = True
 
 
rnn_types = ["RNN", "LSTM", "GRU"]
use_rnn_type = [1, 1, 1]
 
if 0 <= use_type < 3:
    use_rnn_type = [0, 0, 0]
    use_rnn_type[use_type] = 1
 
useGPU = True  # use GPU for TMVA if available
 
useGPU = "tmva-gpu" in ROOT.gROOT.GetConfigFeatures()
useTMVA_RNN = ("tmva-cpu" in ROOT.gROOT.GetConfigFeatures()) or useGPU
 
if useTMVA_RNN:
    ROOT.Warning(
        "TMVA_RNN_Classification",
        "TMVA is not build with GPU or CPU multi-thread support. Cannot use TMVA Deep Learning for RNN",
    )
 
archString = "GPU" if useGPU else "CPU"
 
writeOutputFile = True
 
rnn_type = "RNN"
 
if "tmva-pymva" in ROOT.gROOT.GetConfigFeatures():
    TMVA.PyMethodBase.PyInitialize()
else:
    useKeras = False
 
 
 
inputFileName = "time_data_t10_d30.root"
 
fileDoesNotExist = ROOT.gSystem.AccessPathName(inputFileName)
 
# if file does not exists create it
if fileDoesNotExist:
    MakeTimeData(nTotEvts, ntime, ninput)
 
 
inputFile = TFile.Open(inputFileName)
if inputFile is None:
    raise ROOT.Error("Error opening input file %s - exit", inputFileName.Data())
 
 
print("--- RNNClassification  : Using input file: {}".format(inputFile.GetName()))
 
# Create a ROOT output file where TMVA will store ntuples, histograms, etc.
outfileName = "data_RNN_" + archString + ".root"
outputFile = None
 
 
if writeOutputFile:
    outputFile = TFile.Open(outfileName, "RECREATE")
 
 
## Declare 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
# weightfiles 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
 
 
# // Creating the factory object
factory = TMVA.Factory(
    "TMVAClassification",
    outputFile,
    V=False,
    Silent=False,
    Color=True,
    DrawProgressBar=True,
    Transformations=None,
    Correlations=False,
    AnalysisType="Classification",
    ModelPersistence=True,
)
dataloader = TMVA.DataLoader("dataset")
 
signalTree = inputFile.Get("sgn")
background = inputFile.Get("bkg")
 
nvar = ninput * ntime
 
## add variables - use new AddVariablesArray function
for i in range(ntime):
    dataloader.AddVariablesArray("vars_time" + str(i), ninput)
 
 
dataloader.AddSignalTree(signalTree, 1.0)
dataloader.AddBackgroundTree(background, 1.0)
 
# check given input
datainfo = dataloader.GetDataSetInfo()
vars = datainfo.GetListOfVariables()
print("number of variables is {}".format(vars.size()))
 
 
for v in vars:
    print(v)
 
nTrainSig = 0.8 * nTotEvts
nTrainBkg = 0.8 * nTotEvts
 
# Apply additional cuts on the signal and background samples (can be different)
mycuts = ""  # for example: TCut mycuts = "abs(var1)<0.5 && abs(var2-0.5)<1";
mycutb = ""
 
# build the string options for DataLoader::PrepareTrainingAndTestTree
dataloader.PrepareTrainingAndTestTree(
    mycuts,
    mycutb,
    nTrain_Signal=nTrainSig,
    nTrain_Background=nTrainBkg,
    SplitMode="Random",
    SplitSeed=100,
    NormMode="NumEvents",
    V=False,
    CalcCorrelations=False,
)
 
print("prepared DATA LOADER ")
 
 
## Book TMVA  recurrent models
 
# Book the different types of recurrent models in TMVA  (SimpleRNN, LSTM or GRU)
 
 
if useTMVA_RNN:
    for i in range(3):
        if not use_rnn_type[i]:
            continue
 
        rnn_type = rnn_types[i]
 
        ## Define RNN layer layout
        ##  it should be   LayerType (RNN or LSTM or GRU) |  number of units | number of inputs | time steps | remember output (typically no=0 | return full sequence
        rnnLayout = str(rnn_type) + "|10|" + str(ninput) + "|" + str(ntime) + "|0|1,RESHAPE|FLAT,DENSE|64|TANH,LINEAR"
 
        ## Defining Training strategies. Different training strings can be concatenate. Use however only one
        trainingString1 = "LearningRate=1e-3,Momentum=0.0,Repetitions=1,ConvergenceSteps=5,BatchSize=" + str(batchSize)
        trainingString1 += ",TestRepetitions=1,WeightDecay=1e-2,Regularization=None,MaxEpochs=" + str(maxepochs)
        trainingString1 += "Optimizer=ADAM,DropConfig=0.0+0.+0.+0."
 
        ## define the inputlayout string for RNN
        ## the input data should be organize as   following:
        ##/ input layout for RNN:    time x ndim
        ## add after RNN a reshape layer (needed top flatten the output) and a dense layer with 64 units and a last one
        ## Note the last layer is linear because  when using Crossentropy a Sigmoid is applied already
        ## Define the full RNN Noption string adding the final options for all network
        rnnName = "TMVA_" + str(rnn_type)
        factory.BookMethod(
            dataloader,
            TMVA.Types.kDL,
            rnnName,
            H=False,
            V=True,
            ErrorStrategy="CROSSENTROPY",
            VarTransform=None,
            WeightInitialization="XAVIERUNIFORM",
            ValidationSize=0.2,
            RandomSeed=1234,
            InputLayout=str(ntime) + "|" + str(ninput),
            Layout=rnnLayout,
            TrainingStrategy=trainingString1,
            Architecture=archString
        )
 
 
## Book TMVA  fully connected dense layer  models
if useTMVA_DNN:
    # Method DL with Dense Layer
    # Training strategies.
    trainingString1 = ROOT.TString(
        "LearningRate=1e-3,Momentum=0.0,Repetitions=1,"
        "ConvergenceSteps=10,BatchSize=256,TestRepetitions=1,"
        "WeightDecay=1e-4,Regularization=None,MaxEpochs=20"
        "DropConfig=0.0+0.+0.+0.,Optimizer=ADAM:"
    )  # + "|" + trainingString2
    # General Options.
    trainingString1.Append(archString)
    dnnName = "TMVA_DNN"
    factory.BookMethod(
        dataloader,
        TMVA.Types.kDL,
        dnnName,
        H=False,
        V=True,
        ErrorStrategy="CROSSENTROPY",
        VarTransform=None,
        WeightInitialization="XAVIER",
        RandomSeed=0,
        InputLayout="1|1|" + str(ntime * ninput),
        Layout="DENSE|64|TANH,DENSE|TANH|64,DENSE|TANH|64,LINEAR",
        TrainingStrategy=trainingString1
    )
 
 
## Book Keras recurrent models
 
# Book the different types of recurrent models in Keras  (SimpleRNN, LSTM or GRU)
 
 
if useKeras:
    for i in range(3):
        if use_rnn_type[i]:
            modelName = "model_" + rnn_types[i] + ".h5"
            trainedModelName = "trained_" + modelName
            print("Building recurrent keras model using a", rnn_types[i], "layer")
            # create python script which can be executed
            # create 2 conv2d layer + maxpool + dense
            from tensorflow.keras.models import Sequential
            from tensorflow.keras.optimizers import Adam
 
            # from keras.initializers import TruncatedNormal
            # from keras import initializations
            from tensorflow.keras.layers import Input, Dense, Dropout, Flatten, SimpleRNN, GRU, LSTM, Reshape, BatchNormalization
 
            model = Sequential()
            model.add(Reshape((10, 30), input_shape=(10 * 30,)))
            # add recurrent neural network depending on type / Use option to return the full output
            if rnn_types[i] == "LSTM":
                model.add(LSTM(units=10, return_sequences=True))
            elif rnn_types[i] == "GRU":
                model.add(GRU(units=10, return_sequences=True))
            else:
                model.add(SimpleRNN(units=10, return_sequences=True))
            # m.AddLine("model.add(BatchNormalization())");
            model.add(Flatten())  # needed if returning the full time output sequence
            model.add(Dense(64, activation="tanh"))
            model.add(Dense(2, activation="sigmoid"))
            model.compile(loss="binary_crossentropy", optimizer=Adam(learning_rate=0.001), weighted_metrics=["accuracy"])
            model.save(modelName)
            model.summary()
            print("saved recurrent model", modelName)
 
            if not os.path.exists(modelName):
                useKeras = False
                print("Error creating Keras recurrent model file - Skip using Keras")
            else:
                # book PyKeras method only if Keras model could be created
                print("Booking Keras  model ", rnn_types[i])
                factory.BookMethod(
                    dataloader,
                    TMVA.Types.kPyKeras,
                    "PyKeras_" + rnn_types[i],
                    H=True,
                    V=False,
                    VarTransform=None,
                    FilenameModel=modelName,
                    FilenameTrainedModel="trained_" + modelName,
                    NumEpochs=maxepochs,
                    BatchSize=batchSize,
                    GpuOptions="allow_growth=True",
                )
 
 
# use BDT in case not using Keras or TMVA DL
if not useKeras or not useTMVA_BDT:
    useTMVA_BDT = True
 
 
## Book TMVA BDT
 
 
if useTMVA_BDT:
    factory.BookMethod(
        dataloader,
        TMVA.Types.kBDT,
        "BDTG",
        H=True,
        V=False,
        NTrees=100,
        MinNodeSize="2.5%",
        BoostType="Grad",
        Shrinkage=0.10,
        UseBaggedBoost=True,
        BaggedSampleFraction=0.5,
        nCuts=20,
        MaxDepth=2,
    )
 
 
## Train all methods
factory.TrainAllMethods()
 
print("nthreads  = {}".format(ROOT.GetThreadPoolSize()))
 
# ---- Evaluate all MVAs using the set of test events
factory.TestAllMethods()
 
# ----- Evaluate and compare performance of all configured MVAs
factory.EvaluateAllMethods()
 
# check method
 
# plot ROC curve
c1 = factory.GetROCCurve(dataloader)
c1.Draw()
 
if outputFile:
    outputFile.Close()

Author

Harshal Shende

Definition in file TMVA_RNN_Classification.py.