48270681afc13081094f7f398a1e194c6b07ba9b,gpflow/models/gplvm.py,BayesianGPLVM,_build_likelihood,#BayesianGPLVM#,105

Before Change

        Construct a tensorflow function to compute the bound on the marginal
        likelihood.
        
        num_inducing = tf.shape(self.Z)[0]
        psi0 = tf.reduce_sum(self.kern.eKdiag(self.X_mean, self.X_var), 0)
        psi1 = self.kern.eKxz(self.Z, self.X_mean, self.X_var)
        psi2 = tf.reduce_sum(self.kern.eKzxKxz(self.Z, self.X_mean, self.X_var), 0)
        Kuu = self.kern.K(self.Z) + tf.eye(num_inducing, dtype=settings.float_type) * settings.numerics.jitter_level
        L = tf.cholesky(Kuu)
        sigma2 = self.likelihood.variance
        sigma = tf.sqrt(sigma2)

After Change

        assert self.X_prior_var.shape[1] == self.num_latent

    @params_as_tensors
    def _build_likelihood(self):
        
        Construct a tensorflow function to compute the bound on the marginal
        likelihood.
        
        pX = DiagonalGaussian(self.X_mean, self.X_var)

        num_inducing = len(self.feature)
        psi0 = tf.reduce_sum(expectation(pX, self.kern))
        psi1 = expectation(pX, (self.feature, self.kern))
        psi2 = tf.reduce_sum(expectation(pX, (self.feature, self.kern), (self.feature, self.kern)), axis=0)
        Kuu = self.feature.Kuu(self.kern, jitter=settings.numerics.jitter_level)
        L = tf.cholesky(Kuu)
        sigma2 = self.likelihood.variance
        sigma = tf.sqrt(sigma2)

Instances