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