wolffd@0: function [prob,a] = mdnprob(mixparams, t)
wolffd@0: %MDNPROB Computes the data probability likelihood for an MDN mixture structure.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	PROB = MDNPROB(MIXPARAMS, T) computes the probability P(T) of each
wolffd@0: %	data vector in T under the Gaussian mixture model represented by the
wolffd@0: %	corresponding entries in MIXPARAMS. Each row of T represents a single
wolffd@0: %	vector.
wolffd@0: %
wolffd@0: %	[PROB, A] = MDNPROB(MIXPARAMS, T) also computes the activations A
wolffd@0: %	(i.e. the probability P(T|J) of the data conditioned on each
wolffd@0: %	component density) for a Gaussian mixture model.
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	MDNERR, MDNPOST
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: %	David J Evans (1998)
wolffd@0: 
wolffd@0: % Check arguments for consistency
wolffd@0: errstring = consist(mixparams, 'mdnmixes');
wolffd@0: if ~isempty(errstring)
wolffd@0:   error(errstring);
wolffd@0: end
wolffd@0: 
wolffd@0: ntarget    = size(t, 1);
wolffd@0: if ntarget ~= size(mixparams.centres, 1)
wolffd@0:   error('Number of targets does not match number of mixtures')
wolffd@0: end
wolffd@0: if size(t, 2) ~= mixparams.dim_target
wolffd@0:   error('Target dimension does not match mixture dimension')
wolffd@0: end
wolffd@0: 
wolffd@0: dim_target = mixparams.dim_target;
wolffd@0: ntarget    = size(t, 1);
wolffd@0: 
wolffd@0: % Calculate squared norm matrix, of dimension (ndata, ncentres)
wolffd@0: % vector (ntarget * ncentres)
wolffd@0: dist2 = mdndist2(mixparams, t);
wolffd@0: 
wolffd@0: % Calculate variance factors
wolffd@0: variance = 2.*mixparams.covars;
wolffd@0: 
wolffd@0: % Compute the normalisation term
wolffd@0: normal  = ((2.*pi).*mixparams.covars).^(dim_target./2);
wolffd@0: 
wolffd@0: % Now compute the activations
wolffd@0: a = exp(-(dist2./variance))./normal;
wolffd@0: 
wolffd@0: % Accumulate negative log likelihood of targets
wolffd@0: prob = mixparams.mixcoeffs.*a;