Convert format of TMVA weight file from xml to root

TMVA
1

Dear Expert,

Is it possible to convert the BDT training output model.xml into some weight file in root format? Since I found that in some case, only weight file in root format is used to configure a BDT.

And it is configured as follows:

//________________________________________________________________________________
MVAUtils::BDT* tauRecTools::configureMVABDT( std::map<TString, float*> &availableVars, const TString& weightFile){
	TFile* fBDT = TFile::Open( weightFile );
	if(!fBDT){
		std::cerr << "ERROR Cannot find tau input BDT file: " << weightFile << std::endl;
		return 0;
	}
	TTree* tBDT = dynamic_cast<TTree*> (fBDT->Get("BDT"));
	if(!tBDT){
		delete fBDT;
		std::cerr << "ERROR Cannot find tau input BDT tree" << std::endl;
		return 0;
	}    

	std::cout << "tauRecTools::configureMVABDT opened file: " << weightFile << std::endl;

	std::vector<float*> vars;

	//parsing of variables done here from TNamed object
	TNamed* n_varList = dynamic_cast<TNamed*> (fBDT->Get("varList"));
	if(!n_varList) {
		std::cerr << "ERROR no Variable List in file : " << weightFile << std::endl;
		delete fBDT;
		return 0;
	}
	std::vector<TString> varList_ar = 	tauRecTools::parseStringMVAUtilsBDT(n_varList->GetTitle());
	for(const TString& str : varList_ar){
		if(str.Length()==0) continue;
		std::map<TString, float*>::iterator itr = availableVars.find(str);
		if(itr==availableVars.end()){
			std::cerr << "ERROR Variable : " << str << " is not available" << std::endl;
			delete fBDT;
			delete n_varList;
			return 0;
		}
		vars.push_back( itr->second );
	}
	
	MVAUtils::BDT* reader = new MVAUtils::BDT(tBDT);
  	reader->SetPointers( vars );
  	
  	delete n_varList;
  	delete fBDT;
  	return reader;

}

Thanks a lot!

Cheers,
Bowen

2

Hi,

This code snippet is not part of TMVA/ROOT. TMVA currently only supports outputting weight files in xml format.

There is probably a function in your framework that can do this, you would have to check with the maintainers/documentation of that framework.

Cheers,
Kim

3

Hi Kim,

Thanks for your reply! It’s good to know the problem is not with TMVA. I’ll ask the maintainer on how it works.

Cheers,
Bowen

1 Like