Clearing fStorage and fHistAlias (from Results class) from Factory Training/Testing

TMVA
1

I am using TMVA.Factory() to train/test/evaluate a DNN in Python. I was wondering how to clear the histograms stored in the associated Results object created by factory.

More specifically, I get issues when iterating over the TrainAllMethods(), TestAllMethods() and EvaluateAllMethods(). In the second iteration, when I try running EvaluateAllMethods(), I get the following error:

FATAL: Alias MVA_S already exists in results storage

and when I try to run TMVA.Results.Delete(), I get the error:

unbound method TMVA::Results::Delete must be called with a TMVA::Results instance as first argument

but I am unable to find where the Results class is instantiated by Factory. I essentially want to use the function defined here for the results produced by Factory training/testing done in python.

2

HI,

I am not sure I have understood the cause of your problem. Could you please post your macro or code reproducing this ?

Thank you

Lorenzo

3

Hi @daniel-sunyou-li
I’m facing the same problem
did you manage to solve it please ?

Magdy

4

Hello.

I can reproduce this problem running again the Training/Test/Evaluation using the same method and same dataset. I will investigate it, but I would like to understand it better what is your use case to run again the training/testing workflow ?

Best regards

Lorenzo

5

Hi,
I have investigate this and the solution for having a second iteration of Train/Test/Evaluate is the following:

  • you need to create a new TMVA::Factory instance, but you can use the previously existing data loader. This is needed to have a new ROOT output file for the new training/test results
  • book again your TMVA method
  • delete the existing Results() if you are re-using the DataLoader

For deleting the Results class you should do as following:

 auto data = dataloader->GetDataSetInfo().GetDataSet();
 data->DeleteAllResults(TMVA::Types::kTraining, TMVA::Types::kClassification);
 data->DeleteAllResults(TMVA::Types::kTesting, TMVA::Types::kClassification);

I hope it helps

Lorenzo