16647002974

Committed 31 Jul 2025 10:52AM UTC coverage: 93.906% (+0.005%) from 93.901%

Build # 16647002974

Build Type

push

github

Committed by

mmschlk

Commit Message

fix: Remove defense checks

Run Details

4916 of 5235 relevant lines covered (93.91%)

0.94 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

85.37

/src/shapiq/explainer/tree/explainer.py

"""Implementation of the TreeExplainer class.

The :class:`~shapiq.explainer.tree.explainer.TreeSHAPIQ` uses the
:class:`~shapiq.explainer.tree.treeshapiq.TreeSHAPIQ` algorithm for computing any-order Interactions
for tree ensembles.
"""

from __future__ import annotations

import copy
from typing import TYPE_CHECKING, Any

from shapiq.explainer.base import Explainer

from .treeshapiq import TreeSHAPIQ, TreeSHAPIQIndices
from .validation import validate_tree_model

if TYPE_CHECKING:
    import numpy as np

    from shapiq.interaction_values import InteractionValues
    from shapiq.typing import Model

    from .base import TreeModel


class TreeExplainer(Explainer):
    """The TreeExplainer class for tree-based models.

    The explainer for tree-based models using the
    :class:`~shapiq.explainer.tree.treeshapiq.TreeSHAPIQ` algorithm. For details, refer to
    `Muschalik et al. (2024)` [Mus24]_.

    TreeSHAP-IQ is an algorithm for computing Shapley Interaction values for tree-based models.
    It is based on the Linear TreeSHAP algorithm by `Yu et al. (2022)` [Yu22]_, but extended to
    compute Shapley Interaction values up to a given order. TreeSHAP-IQ needs to visit each node
    only once and makes use of polynomial arithmetic to compute the Shapley Interaction values
    efficiently.

    The TreeExplainer can be used with a variety of tree-based models, including
    ``scikit-learn``, ``XGBoost``, and ``LightGBM``. The explainer can handle both regression and
    classification models.

    References:
        .. [Yu22] Peng Yu, Chao Xu, Albert Bifet, Jesse Read. (2022). Linear Tree Shap. In: Proceedings of 36th Conference on Neural Information Processing Systems. https://openreview.net/forum?id=OzbkiUo24g
        .. [Mus24] Maximilian Muschalik, Fabian Fumagalli, Barbara Hammer, & Eyke Hüllermeier (2024). Beyond TreeSHAP: Efficient Computation of Any-Order Shapley Interactions for Tree Ensembles. In: Proceedings of the AAAI Conference on Artificial Intelligence, 38(13), 14388-14396. https://doi.org/10.1609/aaai.v38i13.29352

    """

    def __init__(
        self,
        model: dict | TreeModel | list[TreeModel] | Model,
        *,
        max_order: int = 2,
        min_order: int = 0,
        index: TreeSHAPIQIndices = "k-SII",
        class_index: int | None = None,
        **kwargs: Any,  # noqa: ARG002
    ) -> None:
        """Initializes the TreeExplainer.

        Args:
            model: A tree-based model to explain.

            max_order: The maximum interaction order to be computed. An interaction order of ``1``
                corresponds to the Shapley value. Any value higher than ``1`` computes the Shapley
                interaction values up to that order. Defaults to ``2``.

            min_order: The minimum interaction order to be computed. Defaults to ``1``.

            index: The type of interaction to be computed. It can be one of
                ``["k-SII", "SII", "STII", "FSII", "BII", "SV"]``. All indices apart from ``"BII"``
                will reduce to the ``"SV"`` (Shapley value) for order 1. Defaults to ``"k-SII"``.

            class_index: The class index of the model to explain. Defaults to ``None``, which will
                set the class index to ``1`` per default for classification models and is ignored
                for regression models.

            **kwargs: Additional keyword arguments are ignored.

        """
        super().__init__(model, index=index, max_order=max_order)

        # validate and parse model
        validated_model = validate_tree_model(model, class_label=class_index)
        self._trees: list[TreeModel] | TreeModel = copy.deepcopy(validated_model)
        if not isinstance(self._trees, list):
            self._trees = [self._trees]
        self._n_trees = len(self._trees)

        self._min_order: int = min_order
        self._class_label: int | None = class_index

        # setup explainers for all trees
        self._treeshapiq_explainers: list[TreeSHAPIQ] = [
            TreeSHAPIQ(model=_tree, max_order=self._max_order, index=index) for _tree in self._trees
        ]
        self.baseline_value = self._compute_baseline_value()

    def explain_function(
        self,
        x: np.ndarray,
        **kwargs: Any,  # noqa: ARG002
    ) -> InteractionValues:
        """Computes the Shapley Interaction values for a single instance.

        Args:
            x: The instance to explain as a 1-dimensional array.
            **kwargs: Additional keyword arguments are ignored.

        Returns:
            The interaction values for the instance.

        """
        if len(x.shape) != 1:
            msg = "explain expects a single instance, not a batch."
            raise TypeError(msg)
        # run treeshapiq for all trees
        interaction_values: list[InteractionValues] = []
        for explainer in self._treeshapiq_explainers:
            tree_explanation = explainer.explain(x)
            interaction_values.append(tree_explanation)

        # combine the explanations for all trees
        final_explanation = interaction_values[0]
        if len(interaction_values) > 1:
            for i in range(1, len(interaction_values)):
                final_explanation += interaction_values[i]

        if self._min_order == 0 and final_explanation.min_order == 1:
            final_explanation.min_order = 0
            # Add the baseline value to the empty prediction
            # might break for some edge cases
            final_explanation.interactions[()] = float(final_explanation.baseline_value)

        return final_explanation

    def _compute_baseline_value(self) -> float:
        """Computes the baseline value for the explainer.

        The baseline value is the sum of the empty predictions of all trees in the ensemble.

        Returns:
            The baseline value for the explainer.

        """
        return sum(
            [treeshapiq.empty_prediction for treeshapiq in self._treeshapiq_explainers],
        )

1	"""Implementation of the TreeExplainer class.
2
3	The :class:`~shapiq.explainer.tree.explainer.TreeSHAPIQ` uses the
4	:class:`~shapiq.explainer.tree.treeshapiq.TreeSHAPIQ` algorithm for computing any-order Interactions
5	for tree ensembles.
6	"""
7
8	from __future__ import annotations	1✔
9
10	import copy	1✔
11	from typing import TYPE_CHECKING, Any	1✔
12
13	from shapiq.explainer.base import Explainer	1✔
14
15	from .treeshapiq import TreeSHAPIQ, TreeSHAPIQIndices	1✔
16	from .validation import validate_tree_model	1✔
17
18	if TYPE_CHECKING:	1✔
19	import numpy as np	×
20
21	from shapiq.interaction_values import InteractionValues	×
22	from shapiq.typing import Model	×
23
24	from .base import TreeModel	×
25
26
27	class TreeExplainer(Explainer):	1✔
28	"""The TreeExplainer class for tree-based models.
29
30	The explainer for tree-based models using the
31	:class:`~shapiq.explainer.tree.treeshapiq.TreeSHAPIQ` algorithm. For details, refer to
32	`Muschalik et al. (2024)` [Mus24]_.
33
34	TreeSHAP-IQ is an algorithm for computing Shapley Interaction values for tree-based models.
35	It is based on the Linear TreeSHAP algorithm by `Yu et al. (2022)` [Yu22]_, but extended to
36	compute Shapley Interaction values up to a given order. TreeSHAP-IQ needs to visit each node
37	only once and makes use of polynomial arithmetic to compute the Shapley Interaction values
38	efficiently.
39
40	The TreeExplainer can be used with a variety of tree-based models, including
41	``scikit-learn``, ``XGBoost``, and ``LightGBM``. The explainer can handle both regression and
42	classification models.
43
44	References:
45	.. [Yu22] Peng Yu, Chao Xu, Albert Bifet, Jesse Read. (2022). Linear Tree Shap. In: Proceedings of 36th Conference on Neural Information Processing Systems. https://openreview.net/forum?id=OzbkiUo24g
46	.. [Mus24] Maximilian Muschalik, Fabian Fumagalli, Barbara Hammer, & Eyke Hüllermeier (2024). Beyond TreeSHAP: Efficient Computation of Any-Order Shapley Interactions for Tree Ensembles. In: Proceedings of the AAAI Conference on Artificial Intelligence, 38(13), 14388-14396. https://doi.org/10.1609/aaai.v38i13.29352
47
48	"""
49
50	def __init__(	1✔
51	self,
52	model: dict \| TreeModel \| list[TreeModel] \| Model,
53	*,
54	max_order: int = 2,
55	min_order: int = 0,
56	index: TreeSHAPIQIndices = "k-SII",
57	class_index: int \| None = None,
58	**kwargs: Any, # noqa: ARG002
59	) -> None:
60	"""Initializes the TreeExplainer.
61
62	Args:
63	model: A tree-based model to explain.
64
65	max_order: The maximum interaction order to be computed. An interaction order of ``1``
66	corresponds to the Shapley value. Any value higher than ``1`` computes the Shapley
67	interaction values up to that order. Defaults to ``2``.
68
69	min_order: The minimum interaction order to be computed. Defaults to ``1``.
70
71	index: The type of interaction to be computed. It can be one of
72	``["k-SII", "SII", "STII", "FSII", "BII", "SV"]``. All indices apart from ``"BII"``
73	will reduce to the ``"SV"`` (Shapley value) for order 1. Defaults to ``"k-SII"``.
74
75	class_index: The class index of the model to explain. Defaults to ``None``, which will
76	set the class index to ``1`` per default for classification models and is ignored
77	for regression models.
78
79	**kwargs: Additional keyword arguments are ignored.
80
81	"""
82	super().__init__(model, index=index, max_order=max_order)	1✔
83
84	# validate and parse model
85	validated_model = validate_tree_model(model, class_label=class_index)	1✔
86	self._trees: list[TreeModel] \| TreeModel = copy.deepcopy(validated_model)	1✔
87	if not isinstance(self._trees, list):	1✔
88	self._trees = [self._trees]	1✔
89	self._n_trees = len(self._trees)	1✔
90
91	self._min_order: int = min_order	1✔
92	self._class_label: int \| None = class_index	1✔
93
94	# setup explainers for all trees
95	self._treeshapiq_explainers: list[TreeSHAPIQ] = [	1✔
96	TreeSHAPIQ(model=_tree, max_order=self._max_order, index=index) for _tree in self._trees
97	]
98	self.baseline_value = self._compute_baseline_value()	1✔
99
100	def explain_function(	1✔
101	self,
102	x: np.ndarray,
103	**kwargs: Any, # noqa: ARG002
104	) -> InteractionValues:
105	"""Computes the Shapley Interaction values for a single instance.
106
107	Args:
108	x: The instance to explain as a 1-dimensional array.
109	**kwargs: Additional keyword arguments are ignored.
110
111	Returns:
112	The interaction values for the instance.
113
114	"""
115	if len(x.shape) != 1:	1✔
116	msg = "explain expects a single instance, not a batch."	×
117	raise TypeError(msg)	×
118	# run treeshapiq for all trees
119	interaction_values: list[InteractionValues] = []	1✔
120	for explainer in self._treeshapiq_explainers:	1✔
121	tree_explanation = explainer.explain(x)	1✔
122	interaction_values.append(tree_explanation)	1✔
123
124	# combine the explanations for all trees
125	final_explanation = interaction_values[0]	1✔
126	if len(interaction_values) > 1:	1✔
127	for i in range(1, len(interaction_values)):	1✔
128	final_explanation += interaction_values[i]	1✔
129
130	if self._min_order == 0 and final_explanation.min_order == 1:	1✔
131	final_explanation.min_order = 0	1✔
132	# Add the baseline value to the empty prediction
133	# might break for some edge cases
134	final_explanation.interactions[()] = float(final_explanation.baseline_value)	1✔
135
136	return final_explanation	1✔
137
138	def _compute_baseline_value(self) -> float:	1✔
139	"""Computes the baseline value for the explainer.
140
141	The baseline value is the sum of the empty predictions of all trees in the ensemble.
142
143	Returns:
144	The baseline value for the explainer.
145
146	"""
147	return sum(	1✔
148	[treeshapiq.empty_prediction for treeshapiq in self._treeshapiq_explainers],
149	)

mmschlk / shapiq / 16647002974

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous