MIGSA

Moment Independent Global Sensitivity Analysis — a set of tools in Rust and Python for computing global sensitivity measures using kernel density estimation (KDE) and hierarchical Dirichlet process (HDP) methods for Borgonovo δ (Delta) sensitivity indices.

Overview

MIGSA provides efficient implementations of moment-independent sensitivity analysis methods. These tools are useful for understanding the influence of input variables on model outputs without relying on variance-based approaches — instead using density-based measures to detect changes in output distributions.

The libraries include:

• Rust core library (migsa) for performant sensitivity computation
• Python bindings (pymigsa) for easy integration into data science workflows

Features

• Compute δ (Delta) sensitivity measures using:

  • Kernel Density Estimation (KDE)
  • Hierarchical Dirichlet Process (HDP) density estimation

• Designed for global sensitivity analysis

• Works with multivariate models and complex output distributions

• High performance from Rust core and seamless Python interface

Installation

Rust (core library)

Add to your Cargo.toml:

[dependencies]
migsa = "0.x"

Then in your Rust code:

use migsa::sensitivity;

Python (bindings)

Install from PyPI (if published):

pip install migsa

Or install directly from the GitHub repo:

git clone https://github.com/schmidmt/migsm.git
cd migsm
pip install ./pymigsa

Then import in Python:

import pymigsa

These examples assume tabular data with input variables and corresponding model outputs.

Tests

To run the Rust tests:

cargo test

To test the Python bindings:

pytest tests/

License

MIGSA is released under the MIT License. See LICENSE for details.

Citation

If you use MIGSA in academic work, consider citing it (provide citation here once available).