MNT: use builtin nsmallest (#633)

* MNT: use builtin nsmallest

Signed-off-by: DerWeh <andreas.weh@web.de>

* MNT: use stable search order to keep exact behavior

Signed-off-by: DerWeh <andreas.weh@web.de>

---------

Signed-off-by: DerWeh <andreas.weh@web.de>

Author: DerWeh

python/interpret-core/interpret/glassbox/_ebm/_ebm.py

@@ -1343,16 +1343,11 @@ def fit(self, X, y, sample_weight=None, bags=None, init_score=None):
                                     (rank - old_mean) / (n + 1)
                                 )
 
-                        final_ranks = []
-                        total_interactions = 0
-                        for indices in pair_ranks:
-                            heapq.heappush(final_ranks, (pair_ranks[indices], indices))
-                            total_interactions += 1
-
-                        n_interactions = min(interactions, total_interactions)
-                        boost_groups = [
-                            heapq.heappop(final_ranks)[1] for _ in range(n_interactions)
-                        ]
+                        boost_groups = heapq.nsmallest(
+                            interactions,
+                            pair_ranks,
+                            key=lambda indices: (pair_ranks[indices], indices),
+                        )
                     else:
                         # Check and remove duplicate interaction terms
                         uniquifier = set()