Daniel@0: function g = glmderiv(net, x)
Daniel@0: %GLMDERIV Evaluate derivatives of GLM outputs with respect to weights.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %	G = GLMDERIV(NET, X) takes a network data structure NET and a matrix
Daniel@0: %	of input vectors X and returns a three-index matrix mat{g} whose  I,
Daniel@0: %	J, K element contains the derivative of network output K with respect
Daniel@0: %	to weight or bias parameter J for input pattern I. The ordering of
Daniel@0: %	the weight and bias parameters is defined by GLMUNPAK.
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: % Check arguments for consistency
Daniel@0: errstring = consist(net, 'glm', x);
Daniel@0: if ~isempty(errstring)
Daniel@0:     error(errstring);
Daniel@0: end
Daniel@0: 
Daniel@0: ndata = size(x, 1);
Daniel@0: if isfield(net, 'mask')
Daniel@0:   nwts = size(find(net.mask), 1);
Daniel@0:   mask_array = logical(net.mask)*ones(1, net.nout);
Daniel@0: else
Daniel@0:   nwts = net.nwts;
Daniel@0: end
Daniel@0: g = zeros(ndata, nwts, net.nout);
Daniel@0: 
Daniel@0: temp = zeros(net.nwts, net.nout);
Daniel@0: for n = 1:ndata
Daniel@0:     % Weight matrix w1
Daniel@0:     temp(1:(net.nin*net.nout), :) = kron(eye(net.nout), (x(n, :))');
Daniel@0:     % Bias term b1
Daniel@0:     temp(net.nin*net.nout+1:end, :) = eye(net.nout);
Daniel@0:     if isfield(net, 'mask')
Daniel@0: 	g(n, :, :) = reshape(temp(find(mask_array)), nwts, net.nout);
Daniel@0:     else
Daniel@0: 	g(n, :, :) = temp;
Daniel@0:     end
Daniel@0: end