shapleyx 0.5.0

Global sensitivity analysis with RS-HDMR — Sobol, Shapley, PAWN, and more

ShapleyX — Global Sensitivity Analysis with RS-HDMR

ShapleyX is a Python package for global sensitivity analysis using Random Sampling High-Dimensional Model Representation (RS-HDMR). It builds sparse polynomial surrogate models via Automatic Relevance Determination (ARD) regression and extracts Sobol indices, Shapley effects, and moment-free sensitivity measures — with full support for correlated inputs through Monte Carlo Shapley estimation.

Features

• Variance-based indices:

  • Sobol sensitivity indices to arbitrary order
  • Shapley effects (independent inputs, via coefficient decomposition)
  • Monte Carlo Shapley effects for correlated inputs (Owen & Prieur 2017)
  • Owen-Shapley interaction indices
  • Total sensitivity indices

• Moment-free measures:

  • PAWN (density-based)
  • Delta index (moment-independent)
  • H-index (distribution-based)

• Distribution classes for correlated inputs:

  • GaussianCopulaUniform — uniform marginals with latent normal dependence
  • MultivariateNormal — jointly normal with analytical conditional sampling
  • TruncatedMultivariateNormal — per-dimension truncation bounds with Gibbs sampling

• Computation methods:

  • Exhaustive subset enumeration (exact, for $d \le 8$)
  • Random permutation method (scalable, for larger $d$)

• Infrastructure:

  • Legendre polynomial expansion on $[0,1]^d$
  • Automatic Relevance Determination (ARD) with Bayesian cross-validation
  • Bootstrap resampling for confidence intervals
  • Progress bars (via tqdm, optional) and Numba-accelerated bootstrap (optional)

Installation

From GitHub (latest version)

pip install https://github.com/frbennett/shapleyx/archive/main.zip

To upgrade:

pip uninstall -y shapleyx
pip install https://github.com/frbennett/shapleyx/archive/main.zip

Development Installation

git clone https://github.com/frbennett/shapleyx.git
cd shapleyx
python setup.py develop

Dependencies

Required	Optional
numpy	numba (compiled bootstrap & Legendre evals)
scipy	tqdm (progress bars)
pandas
matplotlib
scikit-learn

Documentation

Full documentation is available at:

• Document Home
• Quick Start Guide
• MC Shapley How-to
• Theory Background
• API Reference

Examples

Jupyter notebooks demonstrating usage:

Notebook	Description
Ishigami Function	Basic RS-HDMR workflow with Sobol & Shapley
MC Shapley	Monte Carlo Shapley for correlated inputs
Truncated Normal	TruncatedMultivariateNormal with the Ishigami function
Owen Product Function	Higher-dimensional example workflow

License

This project is licensed under the MIT License — see the LICENSE file for details.