Tony O'Hagan - Resources

The GEM Software

The Gaussian Emulation Machine (GEM) software is designed to provide user-friendly tools to implement statistical analyses of uncertainty in the outputs of computer models, using the techniques of Gaussian process emulation. To download an introduction to these tools and the concepts that underlie them, click here.

The GEM software was developed by Marc Kennedy with funding from the Natural Environment Research Council (NERC).

Work on GEM ended around 2009, but the software is still in use and is still available here for download.

GEM-SA

The Gaussian Emulation Machine for Sensitivity Analysis (GEM-SA) is the first product of the GEM project. It will allow you to build an emulator of a computer code from a set of input and output points. It will also perform prediction, uncertainty analysis and sensitivity analysis of the code using far fewer code runs than Monte-Carlo based methods.

GEM-SA is offered free of charge, provided it is not used for profit. It may be freely copied and distributed, provided that files have not been edited or altered in any way, and provided that no charge is made for distributing GEM-SA. Published work that uses GEM-SA results should include a citation for this web page. Requests to use GEM-SA for any commercial purpose must be directed to Marc Kennedy.

If you use GEM-SA please let Marc Kennedy know (marc.kennedy@fera.co.uk).

GEM-SA runs on Windows, but Linux users can also run it. The required package (called "Wine") can be freely downloaded in both binary and source form for various Linux distributions from http://www.winehq.com/site/download.

Tutorials

A course was held on March 21^st and March 22^nd 2005 for NERC related researchers, which included instruction in the use of GEM-SA. This was revised and given again for Engineers at the University of Sheffield on January 30^th and 31^st 2008. It has been revised again and updated for a course aimed at researchers in the biological sciences given on July 3^rd 2008. The practical sessions (PowerPoint) have been made available for download here since the 2005 course, and the revised 2008 versions below fully reflect the changes in GEM-SA that were implemented in version 1.1:

• Getting started with GEM-SA
• Uncertainty analysis with GEM-SA
• Sensitivity analysis with GEM-SA (example uses additional data files SAex1_inputs.txt, SAex1_output.txt)

• Gaussian process modelling

Download instructions

Download V1.1 zip file (1.3MB) MS Windows version.

Extract/unzip the zip file to produce a directory GEM-SA containing the following:

• GEM-SA.exe, the executable
• baccodoc.chm, the help file
• GEM-SA.exe.manifest, used within Windows XP to give the appropriate style windows
• emulator.doc, a MS-Word document describing how to use the emulator independently of GEM-SA
• modeldescription.pdf, some brief details of the statistical model used to build the emulator
• GEM-SA DEMO DATA, a directory containing a number of example code input and output files

GEM-Cal

The second product of the GEM project was intended to be GEM-Cal, implementing the calibration theory of Kennedy and O'Hagan. A beta test version of GEM-Cal may be downloaded here. Installation is similar to the procedure above for GEM-SA. Please contact Marc Kennedy for further information.