mAP

Anyone know of good and FAST ways to perform mAP calculations? The naive approach of calculating PR for each possible score value is going to be very inexcusably slow. There should be an online formulation? When I get time I will think about this - it may involve serious refactoring.

Start by locating the existing mAP calculation and its precision-recall loop; the issue does not name a file or test. Determine whether an online formulation can replace the naive per-score calculation without changing the metric, then benchmark the result against the current approach. Done means substantially faster mAP calculation with equivalent results.

Written by the indexing model from the issue text.