hetGPy: Heteroskedastic Gaussian Process Modeling in Python

hetGPy is a Python implementation of the hetGP R library. Please see our JOSS paper for an in-depth introduction to the package.

This package has the goals of: * Matching the behavior of the R package * Having minimal Python and external dependencies, which are: * numpy and scipy for computations * matplotlib for visualization * joblib for parallelization * tqdm for progress bars * Eigen (C++) for fast calculations (when vectorization is non-obvious)

The motivation for such a package is due to the rising popularity of implementing simulation models (also known as computer experiments) in Python.

Documentation

The package documentation is available at: https://hetgpy.readthedocs.io/en/latest/

Installing and Environments

pypi

• hetGPy is available on pypi:

pip install hetgpy

Development Version:

python -m pip install git+https://github.com/davidogara/hetGPy.git

• To build from the source files:

1. Clone the repository. Make sure to include --recurve-submodules if you do not already have Eigen installed on your system:

git clone --recurse-submodules https://github.com/davidogara/hetGPy.git

1. With hetGPy as your current working directory: pip install -e .

We recommend installing in a virtual environment. One way to do this with venv is: python3.10 -m venv .venv

After this you should be able to run the examples in the examples folder.

Note on Dependencies

• hetGPy requires scipy>=1.14.0 which fixed a memory leakage issue when using L-BFGS-B in scipy.optimize.minizmize. That version of scipy requires Python 3.10.

• Since hetGPy is designed for large-scale problems, this was chosen as a necessary feature. Experienced users may be able to roll back some of the dependencies, but this is not the recommended use.

• hetGPy also requires a c++17 compiler and Eigen for the underlying covariance functions. Eigen 3.4.0 is included with the source files (and is a submodule of the git repository), but experienced users may wish to link against their own installation.

Contact

For questions regarding this package, please contact:
David O'Gara
Division of Computational and Data Sciences, Washington University in St. Louis
david.ogara@wustl.edu

References

Binois M, Gramacy RB (2021). “hetGP: Heteroskedastic Gaussian Process Modeling and Sequential Design in R.” Journal of Statistical Software, 98(13), 1-44. doi:10.18637/jss.v098.i13 https://doi.org/10.18637/jss.v098.i13