Finds the BV that contributes the least to the GP
[imin, scores] = ogpbvmin(gInd);
imin = ogpbvmin
- returns the index if the BV that has the smallest
score (i.e. contribution) to the Gaussian Process net
. The
loss measure is the KL-distance between the GP with the full GP set and the
GP with the BV with index imin
optimally removed.
If there is a second output variable scores
, then the loss vector for each
BV element is returned.
If an additional indicator input variable gInd
is present, then the
KL-distance is replaced with the ``Euclidean'' distance in the feature
space, i.e. the values of net.C
and net.w
are ignored.
Copyright (c) Lehel Csató (2001-2004)