The idea is that many usages of the TMVA Reader boil down to adding a branch to an existing ntuple. This comes with a perceived overhead to the user that all input variables need to be declared “again” (after they were already defined in the training of the classifiers). Eventually, TMVA checks if the right variables have been added by means of what’s stored in the xml file. Instead of verifying that the user defined the right variables, one can as well just read the variables from the xml file and set up the reader.
The usage comes with as little human input as possible
or
The tool is at the moment a proof of principle and available here:
https://github.com/pseyfert/tmva-branch-adder