TreeSHAP-IQ for Custom Tree Models

This example demonstrates how to use TreeExplainer to explain a custom tree model built from scratch using the Play Tennis dataset.

from __future__ import annotations

import numpy as np

import shapiq

The Play Tennis Dataset

A classic dataset with 4 features (Outlook, Temperature, Humidity, Wind) and a binary target (Play Tennis: Yes/No). Features are numerically encoded.

X = np.array(
    [
        [1, 1, 1, 1],
        [1, 1, 1, 2],
        [2, 1, 1, 1],
        [3, 2, 1, 1],
        [3, 3, 2, 1],
        [3, 3, 2, 2],
        [2, 3, 2, 2],
        [1, 2, 1, 1],
        [1, 3, 2, 1],
        [3, 2, 2, 1],
        [1, 2, 2, 2],
        [2, 2, 1, 2],
        [2, 1, 2, 1],
        [3, 2, 1, 2],
    ]
)
y = np.array([0, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 0])

Define a Custom Tree Model

We define a 3-node decision tree:

 #0: Outlook
 /         \
#1: 0      #2: Humidity
           /           \
          #3: 0        #4: Wind
                      /        \
                     #5: 0     #6: 1

tree = shapiq.tree.TreeModel(
    children_left=np.array([1, -1, 3, -1, 5, -1, -1]),
    children_right=np.array([2, -1, 4, -1, 6, -1, -1]),
    children_missing=np.array([-1, -1, -1, -1, -1, -1, -1]),
    features=np.array([0, -2, 2, -2, 3, -2, -2]),
    thresholds=np.array([2.5, -2, 1.5, -2, 1.5, -2, -2]),
    values=np.array([0.5, 0.0, 0.5, 0.0, 0.0, 0.0, 1.0]),
    node_sample_weight=np.array([14, 5, 9, 5, 4, 1, 3]),
)

Explain with TreeSHAP-IQ

Compute exact Shapley values for a single instance.

explainer = shapiq.TreeExplainer(model=tree, index="SV", max_order=1)
sv = explainer.explain(X[5])
print(sv)
sv.plot_force(feature_names=["Outlook", "Temperature", "Humidity", "Wind"])

InteractionValues(
    index=SV, max_order=1, min_order=0, estimated=True, estimation_budget=None,
    n_players=4, baseline_value=0.21428571428571427,
    Top 10 interactions:
        (2,): 0.40343915343915326
        (0,): 0.22982804232804221
        (): 0.21428571428571427
        (3,): 0.1524470899470899
        (1,): 0.0
)