Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/netlabKPM/glmhess_weighted.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:e9a9cd732c1e |
---|---|
1 function [h, hdata] = glmhess_weighted(net, x, t, eso_w, hdata) | |
2 %GLMHESS Evaluate the Hessian matrix for a generalised linear model. | |
3 % | |
4 % Description | |
5 % H = GLMHESS(NET, X, T) takes a GLM network data structure NET, a | |
6 % matrix X of input values, and a matrix T of target values and returns | |
7 % the full Hessian matrix H corresponding to the second derivatives of | |
8 % the negative log posterior distribution, evaluated for the current | |
9 % weight and bias values as defined by NET. Note that the target data | |
10 % is not required in the calculation, but is included to make the | |
11 % interface uniform with NETHESS. For linear and logistic outputs, the | |
12 % computation is very simple and is done (in effect) in one line in | |
13 % GLMTRAIN. | |
14 % | |
15 % See also | |
16 % GLM, GLMTRAIN, HESSCHEK, NETHESS | |
17 % | |
18 % Copyright (c) Ian T Nabney (1996-9) | |
19 | |
20 % Check arguments for consistency | |
21 errstring = consist(net, 'glm', x, t); | |
22 if ~isempty(errstring); | |
23 error(errstring); | |
24 end | |
25 | |
26 ndata = size(x, 1); | |
27 nparams = net.nwts; | |
28 nout = net.nout; | |
29 p = glmfwd(net, x); | |
30 inputs = [x ones(ndata, 1)]; | |
31 | |
32 if nargin == 4 | |
33 hdata = zeros(nparams); % Full Hessian matrix | |
34 % Calculate data component of Hessian | |
35 switch net.outfn | |
36 | |
37 case 'softmax' | |
38 bb_start = nparams - nout + 1; % Start of bias weights block | |
39 ex_hess = zeros(nparams); % Contribution to Hessian from single example | |
40 for m = 1:ndata | |
41 X = x(m,:)'*x(m,:); | |
42 a = diag(p(m,:))-((p(m,:)')*p(m,:)); | |
43 a=eso_w(m,1)*a; | |
44 ex_hess(1:nparams-nout,1:nparams-nout) = kron(a, X); | |
45 ex_hess(bb_start:nparams, bb_start:nparams) = a.*ones(net.nout, net.nout); | |
46 temp = kron(a, x(m,:)); | |
47 ex_hess(bb_start:nparams, 1:nparams-nout) = temp; | |
48 ex_hess(1:nparams-nout, bb_start:nparams) = temp'; | |
49 hdata = hdata + ex_hess; | |
50 end | |
51 | |
52 otherwise | |
53 error(['Unknown activation function ', net.actfn]); | |
54 end | |
55 end | |
56 | |
57 [h, hdata] = hbayes(net, hdata); |