Inspiration for implementing the Gaussian process methods came from
the work of Carl Rasmussen and Chris Williams.  I also benefited from
looking at Carl Rasmussen's implementation.

The sampling method for top-level hyperparameters using adaptive
rejection sampling, and the features allowing cpu time to be stored,
displayed, and used to control the number of iterations, are also
based on work by Carl Edward Rasmussen.

Thanks also to David MacKay, Chris Williams, Tim Howells, Tony Plate,
Dirk Husmeier, and Marcus Lee for comments on the documentation, for
testing the programs, and for finding bugs.