We have included the possibility of computing the quantiles between the probabilistic categories in cross-validation mode (i.e., leaving out the data of the time step for which the terciles are being calculated). To do so, we have modified .get_probs (where the quantiles are calculated), RPS and RPSS.
@jcos, we have slightly modified some lines in the part of the code that uses weights. We have checked that we obtain the same results as the previous version. Anyway, could you please have a look at it and let us know if anything does not make sense to you? Thanks!!
@nmilders, you can start using these functions in cross-validation mode (setting cross.val = TRUE). Could you please test it with the evaluation tool and report any problems that you may find?
Besides, ROCSS should be modified to also have the possibility of computing the probabilities in cross-validation mode. There is a merge request to include this function. @aho, what do you think is better? Can we include ROCSS in the master branch and then create another merge request to include the cross-validation into this function?
Please let us know if you want to discuss anything.
Best regards,
@nperez and Carlos