Gaussian Process regression using a fixed set of basis vectors.
demogp_fixed - script that implements Online Gaussian Process
regression with a fixed set of BV set.
The script generates data from the noisy sinc function where the
inputs are 2-dimensional and the radius of the input is the argument
to sinc.
In this script the BV set is taken from the test set, thus independent from the training set and no addition or removal to/from the BV set is made during learning - using the sparse online algorithm.
The script performs hyper-parameter optimisation and the performance of the fixed-size OGP is compared with the "normal" algorithm using different BV set sizes.
Similar to one-dimensional regression, the script computes test error and
displays the averages over different runs. Additionally, it displays the
position of the BV elements and the posterior mean. If visual inspection of
the results is wanted, then one must set the variable isBreak to a
nonzero value.
Copyright (c) Lehel Csató (2001-2004)