TMVA kDL documentation

Dear experts,

Where can I find documentation about ROOT.TMVA.Types.kDL which seems to replace ROOT.TMVA.Types.kDNN? In particular, I am eager to know the motivation of this change and what are the novelties?

Cheers,

Nikola

I guess @moneta should know.

Hi,

We have replaced the kDNN with kDL because the previous one could not be extended to support different type of layers such as constitutional or recurrent.
In case of dense layer the way of using is basically the same as the previous method, and one can refer to the previous documentation. We are in the process to update the TMVA Users Guide to documente the new methods. At the moment you can look at some tutorials and the most useful are these notebooks in this TMVA tutorial:


In particular this one :

and this other one for CNN networks:

In the next days these tutorials will be included in the ROOT tutorials.
If you have some specific questions, please let me know

Best regards

Lorenzo

Hi,
Many thanks for you replies.
When do you expect TMVA Users Guide be updated?
Is there a tutorial for RNN?
Cheers,
Nikola

Hi,

I am currently working on RNN tutorial, that now is available for GPU and CPU for simpel RNN., LSTM and GRU cells. It should be in the ROOT repository in the next days. I’ll send you the link as soon is available

For the Users Guide I am expecting to update it also soon, hopefully for the next ROOT, 6.22 release, in about 1 month time

Cheers

Lorenzo

Hi,
The updated TMVA Users Guide is available at

and see Section 8.11 for the Deep Learning module.
Tutorials for CNN and RNN are also available in the ROOT master and 6.22 branches.
For CNN:
https://root.cern.ch/doc/master/TMVA__CNN__Classification_8C.html

For RNN:
https://root.cern.ch/doc/master/TMVA__RNN__Classification_8C.html

Lorenzo

Many thanks Lorenzo!