Is there a method in tmva for unsupervised learning? as in training data that does not have labels.
or are there any progress in this area that can be shared at the moment? I found some slide through googling that this features is still in development.
Hi! Yes there is an auto-encoder currently under development but it is not release ready yet
I could recommend you to use pyROOT to export your root data to python and e.g. numpy arrays. From there it is fairly straight forward to use keras, sklearn or any other ML library.
Thanks for the reply!! I have thought about doing it that way.
I am fairly new to ML. So, I am learning more about machine learning tools so I have been playing with pytorch, keras, predominantly ML python libraries. Just wondered if tmva has it as my experiment data are in root format. Mainly worried about the speed of handling the data between different interfaces.
If speed is an issue for prototyping and experimentation you could always convert the data into numpy format and save it as such. I think pandas has good interfaces for this. This would become cumbersome if you need to rerun the conversion too many times, but the approach has saved many hours from my own trainings