wolffd@0: function g = mlpderiv(net, x)
wolffd@0: %MLPDERIV Evaluate derivatives of network outputs with respect to weights.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	G = MLPDERIV(NET, X) takes a network data structure NET and a matrix
wolffd@0: %	of input vectors X and returns a three-index matrix G whose I, J, K
wolffd@0: %	element contains the derivative of network output K with respect to
wolffd@0: %	weight or bias parameter J for input pattern I. The ordering of the
wolffd@0: %	weight and bias parameters is defined by MLPUNPAK.
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	MLP, MLPPAK, MLPGRAD, MLPBKP
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: 
wolffd@0: % Check arguments for consistency
wolffd@0: errstring = consist(net, 'mlp', x);
wolffd@0: if ~isempty(errstring);
wolffd@0:   error(errstring);
wolffd@0: end
wolffd@0: 
wolffd@0: [y, z] = mlpfwd(net, x);
wolffd@0: 
wolffd@0: ndata = size(x, 1);
wolffd@0: 
wolffd@0: if isfield(net, 'mask')
wolffd@0:   nwts = size(find(net.mask), 1);
wolffd@0:   temp = zeros(1, net.nwts);
wolffd@0: else
wolffd@0:   nwts = net.nwts;
wolffd@0: end
wolffd@0: 
wolffd@0: g = zeros(ndata, nwts, net.nout);
wolffd@0: for k = 1 : net.nout
wolffd@0:   delta = zeros(1, net.nout);
wolffd@0:   delta(1, k) = 1;
wolffd@0:   for n = 1 : ndata
wolffd@0:     if isfield(net, 'mask')
wolffd@0:       temp = mlpbkp(net, x(n, :), z(n, :), delta);
wolffd@0:       g(n, :, k) = temp(logical(net.mask));
wolffd@0:     else
wolffd@0:       g(n, :, k) = mlpbkp(net, x(n, :), z(n, :),...
wolffd@0: 	delta);
wolffd@0:     end
wolffd@0:   end
wolffd@0: end