CARPoolGP is an sampling and regression technique developed in https://arxiv.org/abs/2403.10609 . The basic idea, is that when we can force correlations between samples in parameter space, we can reduce variance on emulated quantities. CARPoolGP leverages the CARPool method of https://arxiv.org/abs/2009.08970 and Gaussian process regression.
CARPoolGP can be used:
- To emulate a quantity throughout some parameter space given preexisting samples
- Learn the best place in parameter space to generate new samples at (Active Learning)
We provide here a tutorial with a one dimensional toy example, an application using simulations from GZ here, and a an application to emulate profiles again using the simulations of:
If using in your own work, please cite our work!
Note
This project is under active development.
See the installation section for details on getting started with CARPoolGP. To find a brief description of the theoretical framework for CARPoolGP see theory. We include tutorials in the tutorial section.
installation theory tutorial contact