Glitch in TMVAMultiClass canvas drawing

TMVA
1

Dear experts,

I am trying TMVAMultiClass on my laptop with ROOT 6.17/01. Training goes fine, and when I attempt to use TMVAMultiClassGui for the usual graphical stuff, I notice something weird with the canvases drawn for the classifier output distributions (buttons 4a, 4b only. Others are fine).

When the canvas is first drawn, it looks fine
Screenshot 2020-02-18 10.17.16.PDF (138.7 KB)

But, when I attempt to do any mouse manipulation with the canvas, like resizing the canvas, or even clicking on it, a bunch of components on it disappear
Screenshot 2020-02-18 10.27.42.PDF (139.4 KB)

I don’t see such issues reported elsewhere on the forum. Is it possible that it is a peculiarity with my ROOT build? ( I should mention that I see the same issue with ROOT 6.16.00 on my university cluster)

I am unsure of how to provide a “test script” for this problem. The “TMVA.root” file generated by the training is 118 MB in size, so I doubt I can attach it to this post. If any of the experts reading this have access to CERN lxplus, I have stored the relevant file at /afs/cern.ch/user/a/avenkate/public/forTMVA/TMVA_Multiclass.root

I would be happy to provide any additional info from my side.

Cheers,
Arvind.

2

Hi,
I can access yout output file (TMVA_Multiclass.root), but the file does not contained the ouput score histograms, needed for the 4a and 4b plots.
Running the GUI I get this message
--- Found directory for method: BDTG::BDTG: mva distribution not available (this is normal for Cut classifier)

Lorenzo

3

Hi Lorenzo,

Thanks for taking a look.

That is very weird. I just ran the GUI on the exact same file, and buttons 4a and 4b don’t give me that error. Are you running TMVAGui by any chance? The correct GUI to use is TMVAMultiClassGui (TMVA::TMVAMultiClassGui(“TMVA_Multiclass.root”) for this instance.

When I run TMVAGui on the file, I see the same error that you do.

Arvind.

4

Hi,

Sorry, you are right. I can reproduce now the problem using TMVAMultiClassGui and we are going to investigating it.
Lorenzo

5

I am looking at it.
Thanks for reporting

6

The fix has been found. A PR will come soon.

7

See https://github.com/root-project/root/pull/5055