1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
|
function [y, a] = glmfwd(net, x)
%GLMFWD Forward propagation through generalized linear model.
%
% Description
% Y = GLMFWD(NET, X) takes a generalized linear model data structure
% NET together with a matrix X of input vectors, and forward propagates
% the inputs through the network to generate a matrix Y of output
% vectors. Each row of X corresponds to one input vector and each row
% of Y corresponds to one output vector.
%
% [Y, A] = GLMFWD(NET, X) also returns a matrix A giving the summed
% inputs to each output unit, where each row corresponds to one
% pattern.
%
% See also
% GLM, GLMPAK, GLMUNPAK, GLMERR, GLMGRAD
%
% Copyright (c) Ian T Nabney (1996-2001)
% Check arguments for consistency
errstring = consist(net, 'glm', x);
if ~isempty(errstring);
error(errstring);
end
ndata = size(x, 1);
a = x*net.w1 + ones(ndata, 1)*net.b1;
switch net.outfn
case 'linear' % Linear outputs
y = a;
case 'logistic' % Logistic outputs
% Prevent overflow and underflow: use same bounds as glmerr
% Ensure that log(1-y) is computable: need exp(a) > eps
maxcut = -log(eps);
% Ensure that log(y) is computable
mincut = -log(1/realmin - 1);
a = min(a, maxcut);
a = max(a, mincut);
y = 1./(1 + exp(-a));
case 'softmax' % Softmax outputs
nout = size(a,2);
% Prevent overflow and underflow: use same bounds as glmerr
% Ensure that sum(exp(a), 2) does not overflow
maxcut = log(realmax) - log(nout);
% Ensure that exp(a) > 0
mincut = log(realmin);
a = min(a, maxcut);
a = max(a, mincut);
temp = exp(a);
y = temp./(sum(temp, 2)*ones(1,nout));
% Ensure that log(y) is computable
y(y<realmin) = realmin;
otherwise
error(['Unknown activation function ', net.outfn]);
end
|
