Potential new feature: GSA on arbitrary domain (i.e. non hypercube)

[MaxBalmus]

Summary

Global sensitivity analysis, as performed with Sobol indices, assumes that samples are not correlated. However, this may not always be the case:

• we encounter failed simulations
• we want to perform a sensitivity analysis strictly on plausible samples, for example at the end of History Matching

I have attached the following paper which offers a solution for this problem
Sobol’ indices for problems defined in non-rectangular domains.pdf

Basic Example

A basic example would be the following setup: We are trying to perform a GSA on a model of cardiovascular physiology, but we are trying to limit this analysis to a physiological regime.

1. we perform a history matching step to ensure we have identified the plausible parameter subdomain
2. we use the new GSA to perform the analysis on the subdomain only.

Drawbacks

Not applicable.

Unresolved questions

No response

Implementation PR

No response

Reference Issues

No response

[radka-j]

Thanks for the suggestion @MaxBalmus!

[radka-j]

@all-contributors please add @MaxBalmus for ideas

[allcontributors]

@radka-j

I've put up a pull request to add @MaxBalmus! 🎉