Forward propagation through a Sparse OGP.
[y_mean, y_var] = ogpfwd(x, ktest)
y_mean = ogpfwd(x)
using the GLOBAL sparse OGP data structure
net
together with the matrix x
of input vectors, the function
forward propagates the inputs through the model to generate a matrix y_mean
of posterior means. Each row of x
corresponds to one
input vector and each row of y_mean
corresponds to one output
vector, the mean of the approximated posterior GP.
[y_mean, y_var] = ogpfwd(x)
also generates a column
vector y_var
of conditional variances, each row
corresponds to the variance of an input pattern.
If the network output has more than a single dimension, then the outputs
(both y_mean
and y_var
) will be [nout*nX,1]
and
[nout*nX,nout]
respectively.
[y_mean, y_var] = ogpfwd(net, x, ktest)
uses a
pre-computed covariance matrix ktest
between the inputs x
and
the BV set in the forward propagation.
It assumes that the inputs and the BV set elements do not change. This
increases efficiency when several calls to ogpfwd
are made.
Copyright (c) Lehel Csató (2001-2004)