Daniel@0: function net = glm(nin, nout, outfunc, prior, beta)
Daniel@0: %GLM	Create a generalized linear model.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %
Daniel@0: %	NET = GLM(NIN, NOUT, FUNC) takes the number of inputs and outputs for
Daniel@0: %	a generalized linear model, together with a string FUNC which
Daniel@0: %	specifies the output unit activation function, and returns a data
Daniel@0: %	structure NET. The weights are drawn from a zero mean, isotropic
Daniel@0: %	Gaussian, with variance scaled by the fan-in of the output units.
Daniel@0: %	This makes use of the Matlab function RANDN and so the seed for the
Daniel@0: %	random weight initialization can be  set using RANDN('STATE', S)
Daniel@0: %	where S is the seed value. The optional argument ALPHA sets the
Daniel@0: %	inverse variance for the weight initialization.
Daniel@0: %
Daniel@0: %	The fields in NET are
Daniel@0: %	  type = 'glm'
Daniel@0: %	  nin = number of inputs
Daniel@0: %	  nout = number of outputs
Daniel@0: %	  nwts = total number of weights and biases
Daniel@0: %	  actfn = string describing the output unit activation function:
Daniel@0: %	      'linear'
Daniel@0: %	      'logistic'
Daniel@0: %	      'softmax'
Daniel@0: %	  w1 = first-layer weight matrix
Daniel@0: %	  b1 = first-layer bias vector
Daniel@0: %
Daniel@0: %	NET = GLM(NIN, NOUT, FUNC, PRIOR), in which PRIOR is a scalar, allows
Daniel@0: %	the field  NET.ALPHA in the data structure NET to be set,
Daniel@0: %	corresponding  to a zero-mean isotropic Gaussian prior with inverse
Daniel@0: %	variance with value PRIOR. Alternatively, PRIOR can consist of a data
Daniel@0: %	structure with fields ALPHA and INDEX, allowing individual Gaussian
Daniel@0: %	priors to be set over groups of weights in the network. Here ALPHA is
Daniel@0: %	a column vector in which each element corresponds to a  separate
Daniel@0: %	group of weights, which need not be mutually exclusive.  The
Daniel@0: %	membership of the groups is defined by the matrix INDEX in which the
Daniel@0: %	columns correspond to the elements of ALPHA. Each column has one
Daniel@0: %	element for each weight in the matrix, in the order defined by the
Daniel@0: %	function GLMPAK, and each element is 1 or 0 according to whether the
Daniel@0: %	weight is a member of the corresponding group or not.
Daniel@0: %
Daniel@0: %	NET = GLM(NIN, NOUT, FUNC, PRIOR, BETA) also sets the  additional
Daniel@0: %	field NET.BETA in the data structure NET, where beta corresponds to
Daniel@0: %	the inverse noise variance.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	GLMPAK, GLMUNPAK, GLMFWD, GLMERR, GLMGRAD, GLMTRAIN
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: net.type = 'glm';
Daniel@0: net.nin = nin;
Daniel@0: net.nout = nout;
Daniel@0: net.nwts = (nin + 1)*nout;
Daniel@0: 
Daniel@0: outtfns = {'linear', 'logistic', 'softmax'};
Daniel@0: 
Daniel@0: if sum(strcmp(outfunc, outtfns)) == 0
Daniel@0:   error('Undefined activation function. Exiting.');
Daniel@0: else
Daniel@0:   net.outfn = outfunc;
Daniel@0: end
Daniel@0: 
Daniel@0: if nargin > 3
Daniel@0:   if isstruct(prior)
Daniel@0:     net.alpha = prior.alpha;
Daniel@0:     net.index = prior.index;
Daniel@0:   elseif size(prior) == [1 1]
Daniel@0:     net.alpha = prior;
Daniel@0:   else
Daniel@0:     error('prior must be a scalar or structure');
Daniel@0:   end
Daniel@0: end
Daniel@0:   
Daniel@0: net.w1 = randn(nin, nout)/sqrt(nin + 1);
Daniel@0: net.b1 = randn(1, nout)/sqrt(nin + 1);
Daniel@0: 
Daniel@0: if nargin == 5
Daniel@0:   net.beta = beta;
Daniel@0: end
Daniel@0: