treeshap 0.1.1

Exact TreeSHAP in Rust — fast Shapley values for XGBoost, LightGBM, and ONNX tree ensembles

treeshap

Exact TreeSHAP in Rust with Python bindings. Computes Shapley values for XGBoost, LightGBM, and ONNX tree ensemble models.

Installation

pip install treeshap

Pre-built wheels are available for Linux (x86_64, aarch64), macOS (x86_64, Apple Silicon), and Windows (x86_64). Python 3.8+ required.

Usage

import numpy as np
from treeshap import TreeEnsemble, ShapExplainer

# Load a model
model = TreeEnsemble.from_file("model.json", "xgboost")

# Explain predictions
explainer = ShapExplainer(model)
explanation = explainer.explain(X)  # X is a numpy array

# Access results
print(explanation.shap_values)   # numpy array (n_samples, n_features)
print(explanation.base_value)    # float or list[float]
print(explanation.predictions)   # numpy array (n_samples,)

# Verify local accuracy
report = explanation.verify()
assert report.is_pass

# Generate plots (returns SVG bytes)
svg = explanation.plot_waterfall(sample_index=0)
svg = explanation.plot_beeswarm()
svg = explanation.plot_importance()

Supported Formats

Format	Loader	Example
XGBoost JSON	TreeEnsemble.from_file(path, "xgboost")	booster.save_model("model.json")
LightGBM text	TreeEnsemble.from_file(path, "lightgbm")	booster.save_model("model.txt")
ONNX	TreeEnsemble.from_file(path, "onnx")	TreeEnsembleRegressor/Classifier

Performance

Sub-millisecond single-sample latency. 10,000 samples with 100 trees at depth 6 in 2.8 seconds on Apple M3. GIL is released during computation for full multi-core utilization.

License

MIT OR Apache-2.0