Closed hmrtn closed 4 years ago
I would like all of these function names to be prepended with skill_
so they are easy to find when reviewing the list of available functions.
After some googling I am now comfortably convinced that we should usecaret::confusionMatrix()
as our base implementation of the confusion matrix and then build any higher level metrics we need from the components found in object returned by caret::confusionMatrix()
.
As with the move to the Raster package, we should stop rolling our own and just adopt caret as it appears to be full featured and well supported.
The functionality is based largely on the caret
package and reproduces the same results but follows our data model a little more closesly.
ROC and AUROC are important for determining the model skill. In doing so, we will need a consistent method for calculating confusion matrices.