Pygpc: A sensitivity and uncertainty analysis toolbox for Python
Konstantin Weise,
Lucas Poßner,
Erik Müller,
Richard Gast,
Thomas R. Knösche
Affiliations
Konstantin Weise
Methods and Development Group Brain Networks, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstr. 1a, 04103 Leipzig, Germany; Technische Universität Ilmenau, Advanced Electromagnetics Group, Helmholtzplatz 2, 98693 Ilmenau, Germany; Corresponding author at: Methods and Development Group Brain Networks, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstr. 1a, 04103 Leipzig, Germany.
Lucas Poßner
Leipzig University of Applied Sciences, Institute for Electronics and Biomedical Information Technology, Wächterstr. 13, 04107 Leipzig, Germany
Erik Müller
Leipzig University of Applied Sciences, Institute for Electronics and Biomedical Information Technology, Wächterstr. 13, 04107 Leipzig, Germany
Richard Gast
Methods and Development Group Brain Networks, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstr. 1a, 04103 Leipzig, Germany
Thomas R. Knösche
Methods and Development Group Brain Networks, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstr. 1a, 04103 Leipzig, Germany; Technische Universität Ilmenau, Institute of Biomedical Engineering and Informatics, Gustav-Kirchhoff-Straße 2, 98693 Ilmenau, Germany
We present a novel Python package for the uncertainty and sensitivity analysis of computational models. The mathematical background is based on the non-intrusive generalized polynomial chaos method allowing one to treat the investigated models as black box systems, without interfering with their legacy code. Pygpc is optimized to analyze models with complex and possibly discontinuous transfer functions that are computationally costly to evaluate. The toolbox determines the uncertainty of multiple quantities of interest in parallel, given the uncertainties of the system parameters and inputs. It also yields gradient-based sensitivity measures and Sobol indices to reveal the relative importance of model parameters.
