06 Nov 2002
02 Dec 2002)
Pattern analysis toolbox.
function w = mlppak(net)
%MLPPAK Combines weights and biases into one weights vector.
% W = MLPPAK(NET) takes a network data structure NET and combines the
% component weight matrices bias vectors into a single row vector W.
% The facility to switch between these two representations for the
% network parameters is useful, for example, in training a network by
% error function minimization, since a single vector of parameters can
% be handled by general-purpose optimization routines.
% The ordering of the paramters in W is defined by
% w = [net.w1(:)', net.b1, net.w2(:)', net.b2];
% where W1 is the first-layer weight matrix, B1 is the first-layer
% bias vector, W2 is the second-layer weight matrix, and B2 is the
% second-layer bias vector.
% See also
% MLP, MLPUNPAK, MLPFWD, MLPERR, MLPBKP, MLPGRAD
% Copyright (c) Ian T Nabney (1996-2001)
% Check arguments for consistency
errstring = consist(net, 'mlp');
w = [net.w1(:)', net.b1, net.w2(:)', net.b2];