camir-aes2014: toolboxes/FullBNT-1.0.7/netlab3.3/glminit.m annotate

first hg version after svn

rev	line source
wolffd@0	1 function net = glminit(net, prior)
wolffd@0	2 %GLMINIT Initialise the weights in a generalized linear model.
wolffd@0	3 %
wolffd@0	4 % Description
wolffd@0	5 %
wolffd@0	6 % NET = GLMINIT(NET, PRIOR) takes a generalized linear model NET and
wolffd@0	7 % sets the weights and biases by sampling from a Gaussian distribution.
wolffd@0	8 % If PRIOR is a scalar, then all of the parameters (weights and biases)
wolffd@0	9 % are sampled from a single isotropic Gaussian with inverse variance
wolffd@0	10 % equal to PRIOR. If PRIOR is a data structure similar to that in
wolffd@0	11 % MLPPRIOR but for a single layer of weights, then the parameters are
wolffd@0	12 % sampled from multiple Gaussians according to their groupings (defined
wolffd@0	13 % by the INDEX field) with corresponding variances (defined by the
wolffd@0	14 % ALPHA field).
wolffd@0	15 %
wolffd@0	16 % See also
wolffd@0	17 % GLM, GLMPAK, GLMUNPAK, MLPINIT, MLPPRIOR
wolffd@0	18 %
wolffd@0	19
wolffd@0	20 % Copyright (c) Ian T Nabney (1996-2001)
wolffd@0	21
wolffd@0	22 errstring = consist(net, 'glm');
wolffd@0	23 if ~isempty(errstring);
wolffd@0	24 error(errstring);
wolffd@0	25 end
wolffd@0	26 if isstruct(prior)
wolffd@0	27 sig = 1./sqrt(prior.index*prior.alpha);
wolffd@0	28 w = sig'.*randn(1, net.nwts);
wolffd@0	29 elseif size(prior) == [1 1]
wolffd@0	30 w = randn(1, net.nwts).*sqrt(1/prior);
wolffd@0	31 else
wolffd@0	32 error('prior must be a scalar or a structure');
wolffd@0	33 end
wolffd@0	34
wolffd@0	35 net = glmunpak(net, w);
wolffd@0	36

Mercurial > hg > camir-aes2014