We now compute all @k metrics for all possible values of k. At each call to each metric, we are also sorting the data, which means we have a O(k^2 log(k)) algorithm. It takes more than 5 minutes to run it once on one call to rescore. This is too slow. It needs to be optimized to be practical.
bjschoenfeld
commented
5 years ago
We now compute all @k metrics for all possible values of k. At each call to each metric, we are also sorting the data, which means we have a O(k^2 log(k)) algorithm. It takes more than 5 minutes to run it once on one call to rescore. This is too slow. It needs to be optimized to be practical.