TreeSHAP Explanations¶

Shinrin provides TreeSHAP-based explanations for Mondrian Trees and Forests through the TreeExplainer class.

TreeExplainer¶

The TreeExplainer computes SHAP (SHapley Additive exPlanations) values for tree models.

from shinrin import TreeExplainer

# Create an explainer
explainer = TreeExplainer(model)

# Compute SHAP values
shap_values = explainer.shap_values(X)

# Get the expected value (base score)
expected_value = explainer.expected_value

explanation() Helper¶

For quick visualization, use the convenience explanation() function:

from shinrin import explanation

# Opens a matplotlib visualization
explanation(model, X)

SHAP Value Interpretation¶

SHAP values represent the contribution of each feature to a specific prediction:

Positive values push the prediction higher than the expected value
Negative values push the prediction lower than the expected value
The sum of SHAP values plus the expected value equals the model output

Example¶

import numpy as np
from shinrin import MondrianForestRegressor, TreeExplainer

# Train a model
X = np.random.rand(100, 4)
y = np.random.rand(100)
model = MondrianForestRegressor(n_estimators=10, max_depth=8, random_state=0)
model.fit(X, y)

# Create explainer and compute values
explainer = TreeExplainer(model)
shap_values = explainer.shap_values(X)

# Analyze feature importance
import pandas as pd
feature_importance = np.abs(shap_values).mean(axis=0)
print(pd.Series(feature_importance, index=[f"Feature {i}" for i in range(4)]))