SubglobalSensitivityAnalysis

Estimating Sobol sensitivity indices for a subspace of the global space around a parameter vector, p0.

Problem

Results of global sensitivity analysis (SA) are sometimes strongly influenced by outliers resulting from unreasonable parameter combinations.

The idea is to still apply global SA, but only to a subset of the entire possible parameter space, specifically to a region around a reasonable parameter set.

The user specifies a probability distribution function of each parameter, and the subglobal method ensures that a parameter range is sampled, so that a given proportion (default %20) of the area under its prior pdf is covered.

This ensures that for a parameter with wide distribution also a wide range is sampled, and that more samples are drawn where the prior probability of the parameter is higher.

Getting started

Setup arguments and call the main function estimate_subglobal_sobol_indices, as described in Getting started.

Handle foreign dependencies

This Julia package depends on RCall.jl and the sensitivity R package. If the R package is missing, this Julia package will try to automatically install it into a R session specific library and has to do it on each new R session.

In order to permanently install the sensitivity package into one's R user library execute:

using SubglobalSensitivityAnalysis
install_R_dependencies(["sensitivity"])

Caution, this may interfere with other R projects (see docu).

Note, this installation to R user library needs to be run in a Julia session before running other commands from the package. This is because otherwise the R package is maybe already installed at the R session specific library and the installation for already available packages is skipped.