TMVA Application script gives same error regardless of input

Hello,

I’m trying to apply the results of TMVA training script, but I getting the same error over and over again:

: 16-th variable of the event is NaN → return MVA value -999,
: that’s all I can do, please fix or remove this event.

This error just appears on loop with little variance. It’ll occasionally say 7-th instead of 16-th, but other than that it’s always the same. I reran the training with different signal and background files but still got the same error, and changing the file that I apply the training to doesn’t change the result. Surprisingly, running an older training-file combo that once ran successfully still gave the error. I can’t seem to find any previous mention of this error online, does anyone know what they issue may be? I’ve attached the scripts below. I replaced all of the variables that I declared with a single … to save space.

TMVA Training script:

void MVATrain(){

    // Open file to store command line output

    freopen("../Rankings/rankingsSM.txt","w",stdout);

    // Load library, then open input and output files

    TMVA::Tools::Instance();

    TFile *sigFile = TFile::Open("/mnt/c/Users/uther/Dropbox/UROP_ALPs/kpietapipi-sim-sm.root");

    TFile *backFile = TFile::Open("/mnt/c/Users/uther/Dropbox/UROP_ALPs/kpietapipi_500000.root");

    TFile *outFile = TFile::Open("../trainingOutputFiles/TMVAOutputSM.root", "RECREATE");

    // Create TMVA factory

    auto factory = new TMVA::Factory("BXSignalSM", outFile,

    "!V:ROC:!Correlations:!Silent:Color:!DrawProgressBar:AnalysisType=Classification");

    // Get trees from files

    TTree *tSig;

    TTree *tBackFull;

    TTree *tBack;

    sigFile->GetObject("T;l", tSig);

    backFile->GetObject("T;l", tBackFull);

    // Disable unused branches in tree, keep needed ones

    tSig->SetBranchStatus...

    // Copy the background tree

    tBack = tBackFull->CloneTree(tSig->GetEntries());

    // Create DataLoader and add variables for cuts

    TMVA::DataLoader loader("dataset");

    loader.AddVariable...

    // Create cuts and add tree to dataset

    TCut sigCut = "iCand==0";

    TCut backCut = "iCand==0";

    loader.SetInputTrees(tSig, tBack, 1.0, 1.0);

    loader.PrepareTrainingAndTestTree( sigCut, backCut, 

    "nTrain_Signal=0:nTrain_Background=0:SplitMode=Random:NormMode=NumEvents:!V");

    // Book MVA methods to be run 

    factory->BookMethod(&loader,TMVA::Types::kMLP,"NN");//,"H:!V:NeuronType=tanh:VarTransform=N:NCycles=600:HiddenLayers=N+5:TestRate=5:!UseRegulator:TrainingMethod=BFGS:BPMode=batch");    

    factory->BookMethod(&loader,TMVA::Types::kBDT,"BDT");

    factory->BookMethod(&loader,TMVA::Types::kLD,"LinkDis");

    factory->BookMethod(&loader,TMVA::Types::kHMatrix,"LDAHMatrix");

    factory->BookMethod(&loader,TMVA::Types::kFisher,"LDAFisher");

    factory->BookMethod(&loader,TMVA::Types::kKNN,"KNearestN");

    factory->BookMethod(&loader,TMVA::Types::kRuleFit,"PredictLearn");

    factory->BookMethod(&loader,TMVA::Types::kLikelihood,"ProjectLikeliEst");

    factory->BookMethod(&loader,TMVA::Types::kCuts,"RectCutOpt");

    // Train, Test and Evaluate Methods

    factory->TrainAllMethods();

    factory->TestAllMethods();

    factory->EvaluateAllMethods();

    // Close output command line files

    outFile->Close();

    // Open and ROC curve, Draw GUI, then delete the factory

    if(!gROOT->IsBatch()) TMVA::TMVAGui("/mnt/c/Users/uther/Dropbox/UROP_ALPs/Wagura/trainingOutputFiles/TMVAOutputSM.root");

    delete factory;

    fclose(stdout);

}

TMVA Application script:


void MVAApply(){

    // Load the library

    TMVA::Tools::Instance();

    

    // Initialize the reader

    TMVA::Reader *reader = new TMVA::Reader();

    // Create local variables to store updated inputs

    Float_t b_fd(0.);...

    double db_fd(0.);...

    // Add variables used to make cuts

    reader->AddVariable("b_fd ...

    // Book method

    reader->BookMVA("NN","dataset/weights/BXSignalShort_NN.weights.xml");

    // Create output histogram

    TH1F *cutHisto = new TH1F( "NN-Long", "NN-Long", 100, -.5, .5);

    // Open input file

    //TFile *input = TFile::Open("/mnt/c/Users/uther/Dropbox/alps_kpi3pi/kpi3pi.root");

    TFile *input = TFile::Open("/mnt/c/Users/uther/Dropbox/UROP_ALPS/kpietapipi.root");

    // Prepare event tree

    TTree* applyTree;

    input->GetObject("T;78",applyTree);

    applyTree->SetBranchAddress("b_fd...

    // Loop over events

    for (int i=0, n=applyTree->GetEntries(); i<n; i++)

    {

        applyTree->GetEntry(i);

        // Set Values of TMVA variables

        b_fd = db_fd;...
        //Fill histogram

        cutHisto->Fill(reader->EvaluateMVA("NN"));

    }

    // Write and draw the histogram, then close the file and the reader

    TFile *target = new TFile("TMVAAppOmega.root", "RECREATE");

    cutHisto->Write();

    //cutHisto->Draw();

    target->Close();

    delete reader;

}

_ROOT Version: 6.22/00
_Platform: Ubuntu on WSL
Compiler: Not Provided


@swunsch @moneta @sitong can you take a look please?

Hi!

I cannot spot the issue right away, but I see that you use doubles and floats. TMVA uses floats (and you put in floats) but are you sure that you have doubles in the tree? In case the respective variables are in fact not doubles (in the tree), this could cause such behavior.

You could also just print the input variables to see what you put into TMVA. Are there any NaNs?

If this is not the issue, I would love to have a reproducible piece of code so that I can figure out the issue on my setup.

Best
Stefan