We use scaled predictions in tutorial. Prediction scaling is becoming increasingly obsolete: it was used to maximize kappas that were computed on rounded values but in fact it increases MSE. Since it's used in tutorial, many new users assume this is the default approach.
I think we should replace it with raw predictions. Thoughts?
We use scaled predictions in tutorial. Prediction scaling is becoming increasingly obsolete: it was used to maximize kappas that were computed on rounded values but in fact it increases MSE. Since it's used in tutorial, many new users assume this is the default approach.
I think we should replace it with raw predictions. Thoughts?