b3dab0288ddcd165e2ba6f95061b5f3d7bf82a1a,GPflow/vgp.py,VGP,build_predict,#VGP#Any#Any#,122

Before Change

        f_mean = tf.matmul(Kx, self.q_alpha) + self.mean_function(Xnew)

        // predictive var
        f_var = []
        for d in range(self.num_latent):
            b = self.q_lambda[:, d]
            A = K + tf.diag(1./tf.square(b))
            L = tf.cholesky(A)

After Change

        f_mean = tf.matmul(tf.transpose(Kx), self.q_alpha) + self.mean_function(Xnew)

        // predictive var
        A = K + tf.batch_matrix_diag(tf.transpose(1./tf.square(self.q_lambda)))
        L = tf.batch_cholesky(A)
        Kx_tiled = tf.tile(tf.expand_dims(Kx, 0), [self.num_latent, 1, 1])
        LiKx = tf.batch_matrix_triangular_solve(L, Kx_tiled)
        if full_cov:
            f_var = self.kern.K(Xnew) - tf.batch_matmul(LiKx, LiKx, adj_x=True)
        else:
            f_var = self.kern.Kdiag(Xnew) - tf.reduce_sum(tf.square(LiKx), 1)
        return f_mean, tf.transpose(f_var)

Instances