A major bottleneck with MAE fitting is the branch routing ran post discretizer fit. Right now it is using coordinate descent with no summary stats on leaf partitions. This results in O(n log n) any time we try to add or remove a leaf to a partition which occurs many times on a single execution of coordinate descent.
A major bottleneck with MAE fitting is the branch routing ran post discretizer fit. Right now it is using coordinate descent with no summary stats on leaf partitions. This results in O(n log n) any time we try to add or remove a leaf to a partition which occurs many times on a single execution of coordinate descent.