function likelihood = marginalProbability(x,parameters,method,post)

[obs,dim] = size(x);


if(method == 1)
n = dim/2;
normal = (2*pi)^(n);
for i = 1:parameters.nc
  diffs = x - (ones(obs, 1) * parameters.mus(i,:));
  % Use Cholesky decomposition of covariance matrix to speed computation
  c = chol(parameters.covariances(:,:,i));
  temp = diffs/c;
  sample(:,i) = exp(-0.5*sum(temp.*temp, 2))./(normal*prod(diag(c)));
end
likelihood = sum((sample .* (ones(obs,1)*parameters.alphas)),2);
end

if(method == 2)
n = dim/2;
for class = 1:parameters.nc
    errors = x - (ones(obs, 1) * parameters.mus(class,:));
    K = parameters.covariances(:,:,class);
    Kinv = inv(K);
    G = det(K);
    for man = 1:obs
      sample(man,class)=(1/((2*pi)^n)*sqrt(G))*exp(-.5*errors(man,:)*Kinv*errors(man,:)');
    end
end
likelihood = sum((sample .* (ones(obs,1)*parameters.alphas)),2);
end