- All Implemented Interfaces:
- org.apache.hadoop.io.Writable, OnlineAuc
public class GroupedOnlineAuc
extends Object
implements OnlineAuc
Implements a variant on AUC where the result returned is an average of several AUC measurements
made on sub-groups of the overall data. Controlling for the grouping factor allows the effects
of the grouping factor on the model to be ignored. This is useful, for instance, when using a
classifier as a click prediction engine. In that case you want AUC to refer only to the ranking
of items for a particular user, not to the discrimination of users from each other. Grouping by
user (or user cluster) helps avoid optimizing for the wrong quality.