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

first hg version after svn

rev	line source
wolffd@0	1 function net = mlpinit(net, prior)
wolffd@0	2 %MLPINIT Initialise the weights in a 2-layer feedforward network.
wolffd@0	3 %
wolffd@0	4 % Description
wolffd@0	5 %
wolffd@0	6 % NET = MLPINIT(NET, PRIOR) takes a 2-layer feedforward network 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 of the kind generated by
wolffd@0	11 % MLPPRIOR, then the parameters are sampled from multiple Gaussians
wolffd@0	12 % according to their groupings (defined by the INDEX field) with
wolffd@0	13 % corresponding variances (defined by the ALPHA field).
wolffd@0	14 %
wolffd@0	15 % See also
wolffd@0	16 % MLP, MLPPRIOR, MLPPAK, MLPUNPAK
wolffd@0	17 %
wolffd@0	18
wolffd@0	19 % Copyright (c) Ian T Nabney (1996-2001)
wolffd@0	20
wolffd@0	21 if isstruct(prior)
wolffd@0	22 sig = 1./sqrt(prior.index*prior.alpha);
wolffd@0	23 w = sig'.*randn(1, net.nwts);
wolffd@0	24 elseif size(prior) == [1 1]
wolffd@0	25 w = randn(1, net.nwts).*sqrt(1/prior);
wolffd@0	26 else
wolffd@0	27 error('prior must be a scalar or a structure');
wolffd@0	28 end
wolffd@0	29
wolffd@0	30 net = mlpunpak(net, w);
wolffd@0	31

Mercurial > hg > camir-aes2014