wolffd@0: function  [likXandY, likYgivenX, post] = cwr_prob(cwr, X, Y);
wolffd@0: % CWR_EVAL_PDF cluster weighted regression: evaluate likelihood of Y given X 
wolffd@0: % function  [likXandY, likYgivenX, post] = cwr_prob(cwr, X, Y);
wolffd@0: % 
wolffd@0: % likXandY(t) = p(x(:,t), y(:,t))
wolffd@0: % likXgivenY(t) = p(x(:,t)| y(:,t))
wolffd@0: % post(c,t) = p(c | x(:,t), y(:,t))
wolffd@0: 
wolffd@0: [nx N] = size(X);
wolffd@0: nc = length(cwr.priorC);
wolffd@0: 
wolffd@0: if nc == 1
wolffd@0:   [mu, Sigma] = cwr_predict(cwr, X);
wolffd@0:   likY = gaussian_prob(Y, mu, Sigma);
wolffd@0:   likXandY = likY;
wolffd@0:   likYgivenX = likY;
wolffd@0:   post = ones(1,N);
wolffd@0:   return;
wolffd@0: end
wolffd@0: 
wolffd@0: 
wolffd@0: % likY(c,t) = p(y(:,t) | c)
wolffd@0: likY = clg_prob(X, Y, cwr.muY, cwr.SigmaY, cwr.weightsY);
wolffd@0: 
wolffd@0: % likX(c,t) = p(x(:,t) | c)
wolffd@0: [junk, likX] = mixgauss_prob(X, cwr.muX, cwr.SigmaX);
wolffd@0: likX = squeeze(likX);
wolffd@0: 
wolffd@0: % prior(c,t) = p(c)
wolffd@0: prior = repmat(cwr.priorC(:), 1, N);
wolffd@0: 
wolffd@0: post = likX .* likY .* prior;
wolffd@0: likXandY = sum(post, 1);
wolffd@0: post = post ./ repmat(likXandY, nc, 1);
wolffd@0: %loglik = sum(log(lik));
wolffd@0: %loglik = log(lik);
wolffd@0: 
wolffd@0: likX = sum(likX .* prior, 1);
wolffd@0: likYgivenX = likXandY ./ likX;