tmva103_Application.C File Reference

Detailed Description

This tutorial illustrates how you can conveniently apply BDTs in C++ using the fast tree inference engine offered by TMVA.

Supported workflows are event-by-event inference, batch inference and pipelines with RDataFrame.

 
using namespace TMVA::Experimental;
 
void tmva103_Application()
{
   const char* model_filename = "tmva101.json";
 
   if (gSystem->AccessPathName(model_filename)) {
      Info("tmva103_Application.C", "%s does not exist", model_filename);
      return;
   }
 
   // Load BDT model from the XGBoost JSON written by tmva101_Training.py
   RBDT bdt = RBDT::LoadXGBoost(model_filename);
 
   // Apply model on a single input
   auto y1 = bdt.Compute({1.0, 2.0, 3.0, 4.0});
 
   std::cout << "Apply model on a single input vector: " << y1[0] << std::endl;
 
   // Apply model on a batch of inputs given as a flat, row-major array:
   // 2 events with 4 variables each
   float data[8] = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0};
   auto y2 = bdt.Compute(data, 4);
 
   std::cout << "Apply model on a batch of inputs:";
   for (const auto &output : y2)
      std::cout << " " << output;
   std::cout << std::endl;
 
   // Apply model as part of an RDataFrame workflow
   ROOT::RDataFrame df("Events", "root://eospublic.cern.ch//eos/root-eos/cms_opendata_2012_nanoaod/SMHiggsToZZTo4L.root");
   auto df2 = df.Filter("nMuon >= 2")
                .Filter("nElectron >= 2")
                .Define("Muon_pt_1", "Muon_pt[0]")
                .Define("Muon_pt_2", "Muon_pt[1]")
                .Define("Electron_pt_1", "Electron_pt[0]")
                .Define("Electron_pt_2", "Electron_pt[1]")
                .Define("y",
                        Compute<4, float>(bdt),
                        {"Muon_pt_1", "Muon_pt_2", "Electron_pt_1", "Electron_pt_2"});
 
   std::cout << "Mean response on the signal sample: " << *df2.Mean("y") << std::endl;
}

Apply model on a single input vector: 0.000386276
Apply model on a batch of inputs: 0.000386276 0.158641
Mean response on the signal sample: 0.64075

Date

December 2018

Author

Stefan Wunsch

Definition in file tmva103_Application.C.