sorry for the very lat reply - I’m still not used to the ROOT talk, we mainly use the sourceforge mailing list
for all TMVA related questions. There at least i get a notification (mail) once there’s a question!
But now to your problem. Well, you have 7 variables, imagine how LARGE your possible space of cut values is an you used the MC method (which is just random trial of different cuts) with a tiny number of 2000 ties. Guess, you simply didn’t get any cuts that made any reasonalb sense in this small sample ! For any more in depth debugging I’d certainly need more (e.g. a log output, TMVA version etc.)
Now from the little output log you provided, it might however be another problem all together…
— Interval : maximum lower than minimum
did you look at how the variable distributions look like ? Do they make sense ?? I would almost guess
that one variable is ‘empty’ as the algorithm doesn’t seem to be able to find min/max such that it could
determine the range in which to perform the cut search.
Now, if you really run into trouble, wouldn’t it make sense first to try with 1,2,3 … variables to found out when it fails ? You’d learn much more than just trying one configuration.
And apart from that: try with FitMethod=GA (as recommended in the TMVA Users guide) …
(although that will of course not solve the 'interval min/max problem)