SHAP-Go¶

A Go library for computing SHAP (SHapley Additive exPlanations) values for ML model explainability.

Overview¶

SHAP-Go provides a Go-native implementation of SHAP value computation for explaining machine learning model predictions. It supports:

• ONNX models via ONNX Runtime bindings
• XGBoost/LightGBM via TreeSHAP (exact, fast)
• Custom models via a simple function interface
• Multiple explainer algorithms with different trade-offs
• JSON-serializable explanations for audit/compliance

Why SHAP?¶

SHAP (SHapley Additive exPlanations) values provide a unified approach to explaining machine learning predictions. They answer: "How much did each feature contribute to this prediction?"

Key properties:

• Local accuracy: SHAP values sum to the difference between prediction and baseline
• Consistency: If a feature's contribution increases, its SHAP value won't decrease
• Missingness: Features with no impact have SHAP value of 0

Why Go?¶

• Performance: Native code, no Python overhead
• Deployment: Single binary, easy containerization
• Concurrency: Built-in parallel processing
• Type safety: Catch errors at compile time

Quick Example¶

package main

import (
    "context"
    "fmt"

    "github.com/plexusone/shap-go/explainer"
    "github.com/plexusone/shap-go/explainer/permutation"
    "github.com/plexusone/shap-go/model"
)

func main() {
    // Define a simple model
    predict := func(ctx context.Context, input []float64) (float64, error) {
        return input[0] + 2*input[1], nil
    }
    m := model.NewFuncModel(predict, 2)

    // Background data for SHAP computation
    background := [][]float64{{0.0, 0.0}}

    // Create explainer
    exp, _ := permutation.New(m, background,
        explainer.WithNumSamples(100),
        explainer.WithFeatureNames([]string{"x1", "x2"}),
    )

    // Explain a prediction
    ctx := context.Background()
    explanation, _ := exp.Explain(ctx, []float64{1.0, 2.0})

    fmt.Printf("Prediction: %.2f\n", explanation.Prediction)
    fmt.Printf("Base Value: %.2f\n", explanation.BaseValue)
    for name, shap := range explanation.Values {
        fmt.Printf("SHAP(%s): %.4f\n", name, shap)
    }
}

Explainer Types¶

Getting Started¶

Ready to start explaining your models?