|
wolffd@0
|
1 function [y, sigsq] = gpfwd(net, x, cninv)
|
|
wolffd@0
|
2 %GPFWD Forward propagation through Gaussian Process.
|
|
wolffd@0
|
3 %
|
|
wolffd@0
|
4 % Description
|
|
wolffd@0
|
5 % Y = GPFWD(NET, X) takes a Gaussian Process data structure NET
|
|
wolffd@0
|
6 % together with a matrix X of input vectors, and forward propagates
|
|
wolffd@0
|
7 % the inputs through the model to generate a matrix Y of output
|
|
wolffd@0
|
8 % vectors. Each row of X corresponds to one input vector and each row
|
|
wolffd@0
|
9 % of Y corresponds to one output vector. This assumes that the
|
|
wolffd@0
|
10 % training data (both inputs and targets) has been stored in NET by a
|
|
wolffd@0
|
11 % call to GPINIT; these are needed to compute the training data
|
|
wolffd@0
|
12 % covariance matrix.
|
|
wolffd@0
|
13 %
|
|
wolffd@0
|
14 % [Y, SIGSQ] = GPFWD(NET, X) also generates a column vector SIGSQ of
|
|
wolffd@0
|
15 % conditional variances (or squared error bars) where each value
|
|
wolffd@0
|
16 % corresponds to a pattern.
|
|
wolffd@0
|
17 %
|
|
wolffd@0
|
18 % [Y, SIGSQ] = GPFWD(NET, X, CNINV) uses the pre-computed inverse
|
|
wolffd@0
|
19 % covariance matrix CNINV in the forward propagation. This increases
|
|
wolffd@0
|
20 % efficiency if several calls to GPFWD are made.
|
|
wolffd@0
|
21 %
|
|
wolffd@0
|
22 % See also
|
|
wolffd@0
|
23 % GP, DEMGP, GPINIT
|
|
wolffd@0
|
24 %
|
|
wolffd@0
|
25
|
|
wolffd@0
|
26 % Copyright (c) Ian T Nabney (1996-2001)
|
|
wolffd@0
|
27
|
|
wolffd@0
|
28 errstring = consist(net, 'gp', x);
|
|
wolffd@0
|
29 if ~isempty(errstring);
|
|
wolffd@0
|
30 error(errstring);
|
|
wolffd@0
|
31 end
|
|
wolffd@0
|
32
|
|
wolffd@0
|
33 if ~(isfield(net, 'tr_in') & isfield(net, 'tr_targets'))
|
|
wolffd@0
|
34 error('Require training inputs and targets');
|
|
wolffd@0
|
35 end
|
|
wolffd@0
|
36
|
|
wolffd@0
|
37 if nargin == 2
|
|
wolffd@0
|
38 % Inverse covariance matrix not supplied.
|
|
wolffd@0
|
39 cninv = inv(gpcovar(net, net.tr_in));
|
|
wolffd@0
|
40 end
|
|
wolffd@0
|
41 ktest = gpcovarp(net, x, net.tr_in);
|
|
wolffd@0
|
42
|
|
wolffd@0
|
43 % Predict mean
|
|
wolffd@0
|
44 y = ktest*cninv*net.tr_targets;
|
|
wolffd@0
|
45
|
|
wolffd@0
|
46 if nargout >= 2
|
|
wolffd@0
|
47 % Predict error bar
|
|
wolffd@0
|
48 ndata = size(x, 1);
|
|
wolffd@0
|
49 sigsq = (ones(ndata, 1) * gpcovarp(net, x(1,:), x(1,:))) ...
|
|
wolffd@0
|
50 - sum((ktest*cninv).*ktest, 2);
|
|
wolffd@0
|
51 end
|
