Add auc method for data.frames

[jarauh]

The functions auc and roc basically need the same information to get their work done. However, the interfaces are slightly different: There is a method roc.data.frame, but the corresponding method auc.data.frame is missing.

[xrobin]

Thanks for your suggestion.

Actually I want to do exactly the opposite: remove the auc.formula and auc.default methods of auc (and plot, and ci, and all the other functions). The roc function, and only the roc function, should build a ROC curve.

See #70 for some preparation work in this direction.

Some reasons for that:

• Single-responsibility principle: the auc function calculates the AUC. Not build the ROC curve, not plot the roc curve (yes it currently can do that).
• Avoid the duplication of these .formula and .default methods with pretty much every one of the package function
  • including some inconsistent behaviors with var, sd, plot etc where the .default method does something else altogether.
  • the duplication makes maintenance and especially testing harder.

Basically if you have a data.frame and want to calculate an AUC, I currently believe you should do:

auc(roc(my.data.frame, response, predictor))

That's more explicit, and really if pROC was fully consistent throughout it wouldn't even be harder for the user.

Due to my current lack of time to work on this, and the scope of preparation work needed in #70, deprecating those methods is not going to happen soon though. Not even to mention removing them in a pROC 2.0 version.

Unless you have a strong argument in favor of this proposal, I'd close it as a wontfix.

[jarauh]

Well, I guess no strong argument here. My strongest argument is consistency, but it seems that you have it on your list already.

I'm usually interested in the auc, but not in the roc, so I would prefer to hide the roc away from my scripts. But of course, it suffices if one "direct" auc method (apart from auc.roc) survives your cleaning up.