|
wolffd@0
|
1 function g = glmderiv(net, x)
|
|
wolffd@0
|
2 %GLMDERIV Evaluate derivatives of GLM outputs with respect to weights.
|
|
wolffd@0
|
3 %
|
|
wolffd@0
|
4 % Description
|
|
wolffd@0
|
5 % G = GLMDERIV(NET, X) takes a network data structure NET and a matrix
|
|
wolffd@0
|
6 % of input vectors X and returns a three-index matrix mat{g} whose I,
|
|
wolffd@0
|
7 % J, K element contains the derivative of network output K with respect
|
|
wolffd@0
|
8 % to weight or bias parameter J for input pattern I. The ordering of
|
|
wolffd@0
|
9 % the weight and bias parameters is defined by GLMUNPAK.
|
|
wolffd@0
|
10 %
|
|
wolffd@0
|
11
|
|
wolffd@0
|
12 % Copyright (c) Ian T Nabney (1996-2001)
|
|
wolffd@0
|
13
|
|
wolffd@0
|
14 % Check arguments for consistency
|
|
wolffd@0
|
15 errstring = consist(net, 'glm', x);
|
|
wolffd@0
|
16 if ~isempty(errstring)
|
|
wolffd@0
|
17 error(errstring);
|
|
wolffd@0
|
18 end
|
|
wolffd@0
|
19
|
|
wolffd@0
|
20 ndata = size(x, 1);
|
|
wolffd@0
|
21 if isfield(net, 'mask')
|
|
wolffd@0
|
22 nwts = size(find(net.mask), 1);
|
|
wolffd@0
|
23 mask_array = logical(net.mask)*ones(1, net.nout);
|
|
wolffd@0
|
24 else
|
|
wolffd@0
|
25 nwts = net.nwts;
|
|
wolffd@0
|
26 end
|
|
wolffd@0
|
27 g = zeros(ndata, nwts, net.nout);
|
|
wolffd@0
|
28
|
|
wolffd@0
|
29 temp = zeros(net.nwts, net.nout);
|
|
wolffd@0
|
30 for n = 1:ndata
|
|
wolffd@0
|
31 % Weight matrix w1
|
|
wolffd@0
|
32 temp(1:(net.nin*net.nout), :) = kron(eye(net.nout), (x(n, :))');
|
|
wolffd@0
|
33 % Bias term b1
|
|
wolffd@0
|
34 temp(net.nin*net.nout+1:end, :) = eye(net.nout);
|
|
wolffd@0
|
35 if isfield(net, 'mask')
|
|
wolffd@0
|
36 g(n, :, :) = reshape(temp(find(mask_array)), nwts, net.nout);
|
|
wolffd@0
|
37 else
|
|
wolffd@0
|
38 g(n, :, :) = temp;
|
|
wolffd@0
|
39 end
|
|
wolffd@0
|
40 end
|
