Since I use from time to time a binning for numeric values to apply the scorer it would be helpful if an export of the confusion matrix as a table would be possible. This would give me the chance to calculate something like a "light" kappa statistics, i.e. missing the target class by one class is not as bad as missing it by two or more classes because of the numerical background of the classes.
Further more I would appreciate a real kappa statistics since a wrong decision with a low probability is not as bad as a wrong decision with a high probability.