446a47f0e996fbae4e6bded7ef746ae329354511,sac/distributions/gmm.py,GMM,_create_graph,#GMM#,79

Before Change

            xz_sig_t = tf.boolean_mask(xz_sigs_t, mask_t)  // N x Dx

            // Sample x.
            x_t = tf.stop_gradient(
                xz_mu_t + xz_sig_t * tf.random_normal((N_t, Dx))
            )  // N x Dx

            // log p(x|z)
            log_p_xz_t = self._create_log_gaussian(

After Change

            // Sample x.
            x_t = xz_mu_t + xz_sig_t * tf.random_normal((N_t, Dx))  // N x Dx
            if not self._reparameterize:
                x_t = tf.stop_gradient(x_t)

            // log p(x|z)
            log_p_xz_t = self._create_log_gaussian(
                xz_mus_t, xz_log_sigs_t, x_t[:, None, :]
            )  // N x K

Instances