What is GPyS?

This is a prototypical implementation of Gaussian Process Subspace (GPS) Prediction in the Python programming language. For the original research article documenting the method, see the Citation section.

Table of Contents

• Citation
• Installation
• Example Use
  • For GPS Preprocessor
  • For GPS Hyperparameter Training
  • For GPS Prediction

Citation

• Ruda Zhang, Simon Mak, and David Dunson. Gaussian Process Subspace Prediction for Model Reduction. SIAM Journal on Scientific Computing, 2022. https://epubs.siam.org/doi/10.1137/21M1432739

Installation

Install the package¹ via pip using the following command:

• pip install GPyS==0.1.2

Example Use

After installing the package you can load all modules as shown below:

from GPyS import GPyS_preprocessor, GPyS_prediction, GPyS_LOOCV_error

For GPS Preprocessor:

• Note that only GPyS_preprocessor.Preprocessor.setup(X) takes in argument X and this must be called first before any other functions
• The remaining functions merely return preprocessing quantities of interests

For GPS Hyperparameter Training:

• Utilize GPyS_LOOCV_error.LOOCV.hSSDist(length) method for the objective function computation at a given (default) length scale
• Please take a look at the LOOCV_script.py to see an example computation of optimal lengthscale for GPS.
• Also, all the functions can be independently called here.

For GPS Prediction:

• Call GPyS_prediction.Prediction.GPS_Prediction() to immediately obtain prediction results
• Also, all the functions can be independently called here.

Footnotes

1. this package is created and maintained by Ruda Zhang and Taiwo Adebiyi of the UQ-UH Lab. ↩