Hi @vagudets ,
I have made two updates on the downscaling module.
1-) First, I have added the fcst_only argument under the Downscaling section in the recipe. When the fcst data is provided and this argument is set to TRUE, the module will only downscale the fcst data. If the fcst data is provided and this argument is set to FALSE, the module will downscale both fcst and hcst data. The default value is FALSE. As we discussed separately with you and @nperez , we can revisit this argument when the cross-validation update is available in the future.
2-) The second added argument is probs_cat. This argument is useful for the logistic regression function. The LogisticReg.R function provides categorical (i.e., probabilistic) outcomes instead of numerical ones. The default probs_cat value for the LogisticReg.R function is c(1/3, 2/3). This means that unless probs_cat is provided, the LogisticReg.R function will give probability estimates for terciles as outcomes. In this case, while the RPSS, which is a tercile-based metric, can be calculated with the output of the LogisticReg.R function, metrics such as BSS10 and BSS90 cannot be calculated. For example, to calculate the BSS90 metric with the data downscaled by the LogisticReg.R function, the data should be downscaled by setting probs_cat to 0.9. Users can take advantage of this feature by entering the probs_cat value under the Downscaling section in the recipe.
If anything is unclear, please let me know.
Warm regards,
Eren