Home | News | Documentation | Download

TMVA RNN with Variable Input

Hi, I am fairly new to TMVA but I have explored the tutorials a bit.

I want to ask on what is the proper way to implement RNN with TMVA, such that it can take a variable number of input. In my case, I want to classify between signal and background events, where the number of variables depends on the multiplicity of a physics object. I have gone through the tutorials, but I am still not clear on the proper implementation.

Any guidance would be greatly appreciated!

Thank you.

I guess @moneta can help you .


If you need to provide to the RNN a variable input vector, I would pad the input vector with zero for the missing values.
I would add the zero at the beginning of the sequence as suggested by other studies, for example this paper:

Best regards