wolffd@0: function g = demgpot(x, mix)
wolffd@0: %DEMGPOT Computes the gradient of the negative log likelihood for a mixture model.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	This function computes the gradient of the negative log of the
wolffd@0: %	unconditional data density P(X) with respect to the coefficients of
wolffd@0: %	the data vector X for a Gaussian mixture model.  The data structure
wolffd@0: %	MIX defines the mixture model, while the matrix X contains the data
wolffd@0: %	vector as a row vector. Note the unusual order of the arguments: this
wolffd@0: %	is so that the function can be used in DEMHMC1 directly for sampling
wolffd@0: %	from the distribution P(X).
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	DEMHMC1, DEMMET1, DEMPOT
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: 
wolffd@0: % Computes the potential gradient
wolffd@0: 
wolffd@0: temp = (ones(mix.ncentres,1)*x)-mix.centres;
wolffd@0: temp = temp.*(gmmactiv(mix,x)'*ones(1, mix.nin));
wolffd@0: % Assume spherical covariance structure
wolffd@0: if ~strcmp(mix.covar_type, 'spherical')
wolffd@0:   error('Spherical covariance only.')
wolffd@0: end
wolffd@0: temp = temp./(mix.covars'*ones(1, mix.nin));
wolffd@0: temp = temp.*(mix.priors'*ones(1, mix.nin));
wolffd@0: g = sum(temp, 1)/gmmprob(mix, x);