For rank:pairwise objective, training data and validation data include group information.
For testing, is the group required for ranking?
Will there be a difference if test instances in one group be evaluated separately or in batch?
Rank:pairwise - predict of multiple instance in a group
1 Like