Hi,
So I’m trying to reproduce some stuff on old TMVA training. I’ve got the TMVA output file, TMVA.root, but it seems there are no ready made functions to do just that (or I can’t seem to grasp the TMVA API).
I’ve opened the file and it contains the TMVA::DataLoader object I presume populated by the TMVA::Factory and I can browse it fully. I can see all the models with their weights, training epoch history, etc., and even the train and test trees with all entries. However, I cannot for some reason open that data up in something useful and extract the TMVA performance plots back.
For example, I would like to call something like:
void tmva_plot(std::string infilename) {
// don't change the order of the function calls, some don't
// work otherwise (no idea why)
// HistType 3 is overtraining plot
auto myFile = std::unique_ptr<TFile>(TFile::Open(infilename.c_str()));
std::cout << myFile->GetName() << std::endl;
// TMVA::DataLoader("dataset"); // does not work for some reason
TMVA::mvas("dataset", infilename, static_cast<TMVA::HistType>(3));
// TMVA::mvaeffs("dataset", infilename); // currently has a bug
TMVA::efficiencies("dataset", infilename);
TMVA::correlations("dataset", infilename);
TMVA::variables("dataset", infilename);
myFile->Close();
}
which mostly works, but I cannot specify anything about the plot parameters I get. I would like to get the ROC curve comparison between all the 4 MVAs stored and zoom on the histograms in the high region. I tried manually doing this, but the histogram is stored with just the 100 bins, so I would need to re-evaluate the models over the stored events in the test tree, but I’m not sure how to load the models at all or do that. The docs over here ROOT: TMVA Namespace Reference are not very clear at least to me. The User Guide was also not a huge help. Any pointers will be much appreciated. Thanks in advance!