bb099e4c5b48329b842dbf9884f086f7b514bc0a,gpflow/models/model.py,GPModel,predict_f_samples,#GPModel#Any#Any#Any#Any#,112

Before Change

        num_elems = tf.shape(var)[1]
        var_jitter = ops.add_to_diagonal(var, default_jitter())
        L = tf.linalg.cholesky(var_jitter)  // [P, N, N]
        V = tf.random.normal([num_latent_gps, num_elems, num_samples], dtype=mu.dtype)  // [P, N, S]
        LV = L @ V  // [P, N, S]
        mu_t = tf.linalg.adjoint(mu)  // [P, N]
        return tf.transpose(mu_t[..., np.newaxis] + LV)  // [S, N, P]

    def predict_y(
        self, Xnew: DataPoint, full_cov: bool = False, full_output_cov: bool = False

After Change

        Currently, the method does not support `full_output_cov=True` and `full_cov=True`.
        
        if full_cov and full_output_cov:
            raise NotImplementedError(
                "The combination of both `full_cov` and `full_output_cov` is not supported."
            )

        // check below for shape info
        mean, cov = self.predict_f(Xnew, full_cov=full_cov, full_output_cov=full_output_cov)
        if full_cov:

Instances