Limit tree growth to the system recursion limit

docs/releases/unreleased.md

@@ -1,3 +1,7 @@
 # Unreleased
 
 - The units used in River have been corrected to be based on powers of 2 (KiB, MiB). This only changes the display, the behaviour is unchanged.
+
+## tree
+
+- Instead of letting trees grow indefinitely, setting the `max_depth` parameter to `None` will stop the trees from growing when they reach the system recursion limit.

river/forest/adaptive_random_forest.py

@@ -490,7 +490,7 @@ class ARFClassifier(BaseForest, base.Classifier):
         split attempts.
     max_depth
         [*Tree parameter*] The maximum depth a tree can reach. If `None`, the
-        tree will grow indefinitely.
+        tree will grow until the system recursion limit.
     split_criterion
         [*Tree parameter*] Split criterion to use.<br/>
         - 'gini' - Gini<br/>
@@ -767,7 +767,7 @@ class ARFRegressor(BaseForest, base.Regressor):
         split attempts.
     max_depth
         [*Tree parameter*] The maximum depth a tree can reach. If `None`, the
-        tree will grow indefinitely.
+        tree will grow until the system recursion limit.
     delta
         [*Tree parameter*] Allowed error in split decision, a value closer to 0
         takes longer to decide.

river/tree/extremely_fast_decision_tree.py

@@ -38,7 +38,8 @@ class ExtremelyFastDecisionTreeClassifier(HoeffdingTreeClassifier):
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     min_samples_reevaluate
         Number of instances a node should observe before reevaluating the best split.
     split_criterion

river/tree/hoeffding_adaptive_tree_classifier.py

@@ -27,7 +27,8 @@ class HoeffdingAdaptiveTreeClassifier(HoeffdingTreeClassifier):
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     split_criterion
         Split criterion to use.</br>
         - 'gini' - Gini</br>

river/tree/hoeffding_adaptive_tree_regressor.py

@@ -36,7 +36,8 @@ class HoeffdingAdaptiveTreeRegressor(HoeffdingTreeRegressor):
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     delta
         Significance level to calculate the Hoeffding bound. The significance level is given by
         `1 - delta`. Values closer to zero imply longer split decision delays.

river/tree/hoeffding_tree.py

@@ -4,6 +4,7 @@
 import functools
 import io
 import math
+import sys
 from abc import ABC, abstractmethod
 
 from river import base
@@ -30,7 +31,8 @@ class HoeffdingTree(ABC):
     Parameters
     ----------
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     binary_split
         If True, only allow binary splits.
     max_size
@@ -60,7 +62,7 @@ def __init__(
         self._split_criterion: str = ""
         self._leaf_prediction: str = ""
 
-        self.max_depth: float = max_depth if max_depth is not None else math.inf
+        self.max_depth: int = max_depth if max_depth is not None else (sys.getrecursionlimit() - 20)
         self.binary_split: bool = binary_split
         self._max_size: float = max_size
         self._max_byte_size: float = self._max_size * (2**20)  # convert to byte

river/tree/hoeffding_tree_classifier.py

@@ -18,7 +18,8 @@ class HoeffdingTreeClassifier(HoeffdingTree, base.Classifier):
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     split_criterion
         Split criterion to use.</br>
         - 'gini' - Gini</br>

river/tree/hoeffding_tree_regressor.py

@@ -20,7 +20,8 @@ class HoeffdingTreeRegressor(HoeffdingTree, base.Regressor):
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     delta
         Significance level to calculate the Hoeffding bound. The significance level is given by
         `1 - delta`. Values closer to zero imply longer split decision delays.

river/tree/isoup_tree_regressor.py

@@ -21,7 +21,8 @@ class iSOUPTreeRegressor(tree.HoeffdingTreeRegressor, base.MultiTargetRegressor)
     grace_period
         Number of instances a leaf should observe between split attempts.
     max_depth
-        The maximum depth a tree can reach. If `None`, the tree will grow indefinitely.
+        The maximum depth a tree can reach. If `None`, the tree will grow until
+          the system recursion limit.
     delta
         Allowed error in split decision, a value closer to 0 takes longer to
         decide.

river/tree/stochastic_gradient_tree.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 import abc
-import math
+import sys
 
 from scipy.stats import f as f_dist
 
@@ -38,7 +38,7 @@ def __init__(
         self.delta = delta
         self.grace_period = grace_period
         self.init_pred = init_pred
-        self.max_depth = max_depth if max_depth else math.inf
+        self.max_depth = max_depth if max_depth else (sys.getrecursionlimit() - 20)
 
         if lambda_value < 0.0:
             raise ValueError('Invalid value: "lambda_value" must be positive.')