glmgrad.m

来自「Bayes网络工具箱」· M 代码 · 共 58 行

function [g, gdata, gprior] = glmgrad(net, x, t)
%GLMGRAD Evaluate gradient of error function for generalized linear model.
%
%	Description
%	G = GLMGRAD(NET, X, T) takes a generalized linear model data
%	structure NET  together with a matrix X of input vectors and a matrix
%	T of target vectors, and evaluates the gradient G of the error
%	function with respect to the network weights. The error function
%	corresponds to the choice of output unit activation function. Each
%	row of X corresponds to one input vector and each row of T
%	corresponds to one target vector.
%
%	[G, GDATA, GPRIOR] = GLMGRAD(NET, X, T) also returns separately  the
%	data and prior contributions to the gradient.
%
%	See also
%	GLM, GLMPAK, GLMUNPAK, GLMFWD, GLMERR, GLMTRAIN
%

%	Copyright (c) Christopher M Bishop, Ian T Nabney (1996, 1997)

% Check arguments for consistency
errstring = consist(net, 'glm', x, t);
if ~isempty(errstring);
  error(errstring);
end

y = glmfwd(net, x);
delout = y - t;

gw1 = x'*delout;
gb1 = sum(delout, 1);

gdata = [gw1(:)', gb1];

if isfield(net, 'beta')
  g1 = gdata*net.beta;
else
  g1 = gdata;
end

if isfield(net, 'alpha')
  w = mlppak(net);
  if size(net.alpha) == [1 1]
    gprior = w;
    g2 = net.alpha*gprior;
  else
    ngroups = size(net.alpha, 1);
    gprior = net.index'.*(ones(ngroups, 1)*w);
    g2 = net.alpha'*gprior;
  end
else
  gprior = 0;
  g2 = 0;
end

g = g1 + g2;