comparison toolboxes/FullBNT-1.0.7/netlab3.3/mlppak.m @ 0:e9a9cd732c1e tip

first hg version after svn

+function w = mlppak(net)
+%MLPPAK	Combines weights and biases into one weights vector.
+%
+%	Description
+%	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');
+if ~isempty(errstring);
+  error(errstring);
+end
+
+w = [net.w1(:)', net.b1, net.w2(:)', net.b2];
+