This is a repository for the paper Uncertainty-aware Sensitivity Analysis Using Rényi Divergences by Topi Paananen, Michael Riis Andersen, and Aki Vehtari.
The folder demo-1D contains a 1D demo for plotting the R-sens uncertainty-aware sensitivity for user-defined predictive distributions. The folder contains demos for a Gaussian predictive distribution and Bernoulli predictive distribution.
The folder gpytorch contains the R-sens and R-sens2 codes for GPyTorch. There are also several scripts that show examples of how to use the code. Currently supported likelihoods are Gaussian and Bernoulli. More to come soon.
Paananen, T., Andersen, M. R., and Vehtari, A. (2021). Uncertainty-aware Sensitivity Analysis Using Rényi Divergences. UAI 2021, accepted for publication. (arXiv Preprint)