diff options
Diffstat (limited to 'sourcecodes/bnt-master/netlab3.3/mlppak.m')
| -rw-r--r-- | sourcecodes/bnt-master/netlab3.3/mlppak.m | 31 |
1 files changed, 31 insertions, 0 deletions
diff --git a/sourcecodes/bnt-master/netlab3.3/mlppak.m b/sourcecodes/bnt-master/netlab3.3/mlppak.m new file mode 100644 index 00000000..30ec375d --- /dev/null +++ b/sourcecodes/bnt-master/netlab3.3/mlppak.m @@ -0,0 +1,31 @@ +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]; + |