is there a significance plot which allow to determine the best cut when using the TMVAMulticlass.C?
It exist for the simple TMVAClassification.C. If not how can I determine the best cuts?
The multiclass optimisation problem is much more tricky in the multiclass setting as compared to binary classification because of the increased number of dimensions. The standard approach with a line search becomes infeasible.
For now this optimisation must be done outside of TMVA. Approaches include binning the multidimensional output space and optimising the classification in each bin according to some criterion, and optimisation through genetic algorithms.
ok, thank you for your answer.