ENH Add interaction constraint shortcuts to HistGradientBoosting* (#2…

…4849)

Co-authored-by: jeremie du boisberranger <jeremiedbb@yahoo.fr>

doc/whats_new/v1.2.rst

@@ -328,6 +328,12 @@ Changelog
   as value to specify monotonicity constraints for each feature.
   :pr:`24855` by :user:`Olivier Grisel <ogrisel>`.
 
+- |Enhancement| Interaction constraints for
+  :class:`ensemble.HistGradientBoostingClassifier`
+  and :class:`ensemble.HistGradientBoostingRegressor` can now be specified
+  as strings for two common cases: "no_interactions" and "pairwise" interactions.
+  :pr:`24849` by :user:`Tim Head <betatim>`.
+
 - |Fix| Fixed the issue where :class:`ensemble.AdaBoostClassifier` outputs
   NaN in feature importance when fitted with very small sample weight.
   :pr:`20415` by :user:`Zhehao Liu <MaxwellLZH>`.

examples/inspection/plot_partial_dependence.py

@@ -270,9 +270,7 @@
 
 print("Training interaction constraint HistGradientBoostingRegressor...")
 tic = time()
-est_no_interactions = HistGradientBoostingRegressor(
-    interaction_cst=[[i] for i in range(X_train.shape[1])]
-)
+est_no_interactions = HistGradientBoostingRegressor(interaction_cst="no_interactions")
 est_no_interactions.fit(X_train, y_train)
 print(f"done in {time() - tic:.3f}s")
 

sklearn/ensemble/_hist_gradient_boosting/gradient_boosting.py

@@ -3,6 +3,7 @@
 
 from abc import ABC, abstractmethod
 from functools import partial
+import itertools
 from numbers import Real, Integral
 import warnings
 
@@ -92,7 +93,12 @@ class BaseHistGradientBoosting(BaseEstimator, ABC):
         "min_samples_leaf": [Interval(Integral, 1, None, closed="left")],
         "l2_regularization": [Interval(Real, 0, None, closed="left")],
         "monotonic_cst": ["array-like", dict, None],
-        "interaction_cst": [list, tuple, None],
+        "interaction_cst": [
+            list,
+            tuple,
+            StrOptions({"pairwise", "no_interactions"}),
+            None,
+        ],
         "n_iter_no_change": [Interval(Integral, 1, None, closed="left")],
         "validation_fraction": [
             Interval(Real, 0, 1, closed="neither"),
@@ -288,8 +294,15 @@ def _check_interaction_cst(self, n_features):
         if self.interaction_cst is None:
             return None
 
+        if self.interaction_cst == "no_interactions":
+            interaction_cst = [[i] for i in range(n_features)]
+        elif self.interaction_cst == "pairwise":
+            interaction_cst = itertools.combinations(range(n_features), 2)
+        else:
+            interaction_cst = self.interaction_cst
+
         try:
-            constraints = [set(group) for group in self.interaction_cst]
+            constraints = [set(group) for group in interaction_cst]
         except TypeError:
             raise ValueError(
                 "Interaction constraints must be a sequence of tuples or lists, got:"
@@ -1275,7 +1288,8 @@ class HistGradientBoostingRegressor(RegressorMixin, BaseHistGradientBoosting):
         .. versionchanged:: 1.2
            Accept dict of constraints with feature names as keys.
 
-    interaction_cst : sequence of lists/tuples/sets of int, default=None
+    interaction_cst : {"pairwise", "no_interaction"} or sequence of lists/tuples/sets \
+            of int, default=None
         Specify interaction constraints, the sets of features which can
         interact with each other in child node splits.
 
@@ -1284,6 +1298,9 @@ class HistGradientBoostingRegressor(RegressorMixin, BaseHistGradientBoosting):
         specified in these constraints, they are treated as if they were
         specified as an additional set.
 
+        The strings "pairwise" and "no_interactions" are shorthands for
+        allowing only pairwise or no interactions, respectively.
+
         For instance, with 5 features in total, `interaction_cst=[{0, 1}]`
         is equivalent to `interaction_cst=[{0, 1}, {2, 3, 4}]`,
         and specifies that each branch of a tree will either only split
@@ -1623,7 +1640,8 @@ class HistGradientBoostingClassifier(ClassifierMixin, BaseHistGradientBoosting):
         .. versionchanged:: 1.2
            Accept dict of constraints with feature names as keys.
 
-    interaction_cst : sequence of lists/tuples/sets of int, default=None
+    interaction_cst : {"pairwise", "no_interaction"} or sequence of lists/tuples/sets \
+            of int, default=None
         Specify interaction constraints, the sets of features which can
         interact with each other in child node splits.
 
@@ -1632,6 +1650,9 @@ class HistGradientBoostingClassifier(ClassifierMixin, BaseHistGradientBoosting):
         specified in these constraints, they are treated as if they were
         specified as an additional set.
 
+        The strings "pairwise" and "no_interactions" are shorthands for
+        allowing only pairwise or no interactions, respectively.
+
         For instance, with 5 features in total, `interaction_cst=[{0, 1}]`
         is equivalent to `interaction_cst=[{0, 1}, {2, 3, 4}]`,
         and specifies that each branch of a tree will either only split

sklearn/ensemble/_hist_gradient_boosting/tests/test_gradient_boosting.py

@@ -1187,6 +1187,10 @@ def test_uint8_predict(Est):
     [
         (None, 931, None),
         ([{0, 1}], 2, [{0, 1}]),
+        ("pairwise", 2, [{0, 1}]),
+        ("pairwise", 4, [{0, 1}, {0, 2}, {0, 3}, {1, 2}, {1, 3}, {2, 3}]),
+        ("no_interactions", 2, [{0}, {1}]),
+        ("no_interactions", 4, [{0}, {1}, {2}, {3}]),
         ([(1, 0), [5, 1]], 6, [{0, 1}, {1, 5}, {2, 3, 4}]),
     ],
 )