User Guide

This guide covers how to use shap-monitor effectively in your ML pipelines.

Contents

• Monitoring - Using SHAPMonitor to log explanations
• Analysis - Analyzing logged data with SHAPAnalyzer
• Storage Backends - Understanding storage options
• Examples - Complete working examples

Overview

shap-monitor provides two main components:

SHAPMonitor

Logs SHAP explanations in production with:

• Configurable sampling rate
• Automatic SHAP value computation
• Efficient Parquet storage
• Model version tracking

SHAPAnalyzer

Analyzes logged explanations with:

• Summary statistics over time periods
• Time period comparisons
• Feature importance rankings
• Change detection