Those of us who do high-dimensional approximation know the value and frustration inherent in using Gaussian basis functions.
To try and combat ill-conditioning for increasingly flat Gaussians we have developed an RBFQR approach for use in general dimensions.
This builds directly on the work of Fornberg, Larsson, Flyer, Piret and many others.
This material is currently being revised for preparation in my thesis, and once a final version is completed it will be made available here.
For the time being, the original paper is available here.
The purpose of this website is to give you access to code which will allow you to reproduce our results and experiment yourself.
At the moment the code isn't terribly well commented, but it's being addressed as time permits.
For those of you who want to see what the code looks like, you can download the latest stable version below:
If you have any issues, feel free to email firstname.lastname@example.org or email@example.com with questions.
This code is infrequently, if ever, run on a Mac, so be warned. There shouldn't be any issues on OS X (because it is Unix based)
but I'm not a Mac expert.
Code Development Repository
For those of you interested in running the latest version of the code, which is quite volatile, you may feel free to access
the Mercurial repository we have set up. The website is this address. For those of you who are command line fans, the line is
Those of you using a Windows machine can either access HG through Cygwin or download TortoiseHG.
Changes are being made to this code often, so it might break something you have written. On the other hand if you have code requests,
we would send you a response through the repository.