I have a radial basis function that supplies uncertainties (standard deviations) with its predictions, which are numerical values.
This function is computed for a particular point by computing its relative distance to a large set of other reference points in high dimensional space, and compositing a prediction from them.
Over the training set I can compute R to get a correlation between prediction and actual. Weights are assigned to each dimension and optimized to maximize R.
Over a validation set, it seems I'd want to calculate something other than R to measure the model's predictive power, since its predictions are not single values, but ranges.