Merge pull request #59 from mmschlk/tree-explainer-random-forrest

Bugfix and Support for Random Forests in TreeExplainer

Author: mmschlk

shapiq/explainer/tree/base.py

@@ -109,7 +109,7 @@ def reduce_feature_complexity(self) -> None:
         Feature '8' is 'renamed' to '2' such that in the internal representation a one-hot vector
         (and matrices) of length 3 suffices to represent the feature indices.
         """
-        if self.n_features_in_tree < self.max_feature_id:
+        if self.n_features_in_tree < self.max_feature_id + 1:
             new_feature_ids = set(range(self.n_features_in_tree))
             mapping_old_new = {old_id: new_id for new_id, old_id in enumerate(self.feature_ids)}
             mapping_new_old = {new_id: old_id for new_id, old_id in enumerate(self.feature_ids)}

shapiq/explainer/tree/conversion/sklearn.py

@@ -1,24 +1,19 @@
 """This module contains functions for converting scikit-learn decision trees to the format used by
  shapiq."""
 
-from typing import Optional, Union
+from typing import Optional
 
 import numpy as np
 from explainer.tree.base import TreeModel
 
 from shapiq.utils import safe_isinstance
-
-try:
-    from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
-    from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
-except ImportError:
-    pass
+from shapiq.utils.types import Model
 
 
 def convert_sklearn_forest(
-    tree_model: Union["RandomForestRegressor", "RandomForestClassifier"],
-    class_label: int = 0,
-    output_type: Optional[str] = None,
+    tree_model: Model,
+    class_label: Optional[int] = None,
+    output_type: str = "raw",
 ) -> list[TreeModel]:
     """Transforms a scikit-learn random forest to the format used by shapiq.
 
@@ -33,8 +28,6 @@ def convert_sklearn_forest(
         The converted random forest model.
     """
     scaling = 1.0 / len(tree_model.estimators_)
-    if not safe_isinstance(tree_model, "sklearn.ensemble.RandomForestClassifier"):
-        output_type = None
     return [
         convert_sklearn_tree(
             tree, scaling=scaling, class_label=class_label, output_type=output_type
@@ -44,8 +37,8 @@ def convert_sklearn_forest(
 
 
 def convert_sklearn_tree(
-    tree_model: Union["DecisionTreeRegressor", "DecisionTreeClassifier"],
-    class_label: int = 0,
+    tree_model: Model,
+    class_label: Optional[int] = None,
     scaling: float = 1.0,
     output_type: str = "raw",
 ) -> TreeModel:
@@ -63,14 +56,16 @@ def convert_sklearn_tree(
         The converted decision tree model.
     """
     tree_values = tree_model.tree_.value.copy() * scaling
+    # set class label if not given and model is a classifier
+    if safe_isinstance(tree_model, "sklearn.tree.DecisionTreeClassifier") and class_label is None:
+        class_label = 1
+
     if class_label is not None:
         # turn node values into probabilities
         if len(tree_values.shape) == 3:
-            tree_values = tree_values / np.sum(tree_values, axis=2, keepdims=True)
-            tree_values = tree_values[:, 0, class_label]
-        else:
-            tree_values = tree_values / np.sum(tree_values, axis=1, keepdims=True)
-            tree_values = tree_values[:, class_label]
+            tree_values = tree_values[:, 0, :]
+        tree_values = tree_values / np.sum(tree_values, axis=1, keepdims=True)
+        tree_values = tree_values[:, class_label]
     if output_type != "raw":
         # TODO: Add support for logits output type
         raise NotImplementedError("Only raw output types are currently supported.")

shapiq/explainer/tree/conversion/xgboost.py

@@ -1,7 +1,7 @@
 """This module contains the TreeExplainer class making use of the TreeSHAPIQ algorithm for
 computing any-order Shapley Interactions for tree ensembles."""
 import copy
-from typing import Any, Union
+from typing import Any, Optional, Union
 
 import numpy as np
 from explainer._base import Explainer
@@ -17,17 +17,20 @@ def __init__(
         model: Union[dict, TreeModel, Any],
         max_order: int = 2,
         min_order: int = 1,
+        class_label: Optional[int] = None,
+        output_type: str = "raw",
     ) -> None:
         # validate and parse model
-        validated_model = _validate_model(model)  # the parsed and validated model
-
+        validated_model = _validate_model(model, class_label=class_label, output_type=output_type)
         self._trees: Union[TreeModel, list[TreeModel]] = copy.deepcopy(validated_model)
         if not isinstance(self._trees, list):
             self._trees = [self._trees]
         self._n_trees = len(self._trees)
 
-        self._max_order = max_order
-        self._min_order = min_order
+        self._max_order: int = max_order
+        self._min_order: int = min_order
+        self._class_label: Optional[int] = class_label
+        self._output_type: str = output_type
 
         # setup explainers for all trees
         self._treeshapiq_explainers: list[TreeSHAPIQ] = [

shapiq/explainer/tree/explainer.py

@@ -1,6 +1,7 @@
 """This module contains the tree explainer implementation."""
 import copy
 from math import factorial
+from typing import Any, Optional, Union
 
 import numpy as np
 from approximator import transforms_sii_to_ksii
@@ -30,7 +31,7 @@ class TreeSHAPIQ:
 
     def __init__(
         self,
-        model: TreeModel,
+        model: Union[dict, TreeModel, Any],
         max_order: int = 2,
         min_order: int = 1,
         interaction_type: str = "k-SII",
@@ -507,24 +508,22 @@ def _get_N_cii(self, interpolated_poly, order) -> np.ndarray[float]:
             )
         return Ns
 
-    def _get_subset_weight_cii(self, t, order) -> float:
+    def _get_subset_weight_cii(self, t, order) -> Optional[float]:
         # TODO: add docstring
         if self._interaction_type == "STI":
             return self._max_order / (
                 self._n_features_in_tree * binom(self._n_features_in_tree - 1, t)
             )
-        elif self._interaction_type == "FSI":
+        if self._interaction_type == "FSI":
             return (
                 factorial(2 * self._max_order - 1)
                 / factorial(self._max_order - 1) ** 2
                 * factorial(self._max_order + t - 1)
                 * factorial(self._n_features_in_tree - t - 1)
                 / factorial(self._n_features_in_tree + self._max_order - 1)
             )
-        elif self._interaction_type == "BZF":
+        if self._interaction_type == "BZF":
             return 1 / (2 ** (self._n_features_in_tree - order))
-        else:
-            raise ValueError("Interaction type not supported")
 
     @staticmethod
     def _get_N_id(D) -> np.ndarray[float]: