Feature Request: ROC curve averaging

[diegozea]

Would be great to have some ROC curve averaging like ROCR.

[davidavdav]

Thanks for the paper. I think Fawcett's ideas originally inspired us to implement the ROC-CH.

As for averaging, you can do this at the level of scores of course, but this assumes equal sample sizes. A better way is to weight the scores inversely proportional to their relative frequency among the different sets. I've implemented that before in an R package.

Fawcett's Vertical averaging doesn't make any sense to me.

[cossio]

Was this added by any chance?