67472e8698ea7764012bf2d97519167878383b05,sac/policies/real_nvp.py,RealNVPPolicy,build,#RealNVPPolicy#,62

Before Change

            tf.stop_gradient(self.distribution.bijector.forward(self.x)),
            (None, 2),
            name="y")
        self.inverse_x = self.distribution.bijector.inverse(self.y)


        if self.config["squash"]:
            self._action = tf.tanh(self.y)

After Change

        self.Q = self._qf(self._observations_ph, self._action)
        self.log_pi = self.distribution.log_prob(
            self.y, bijector_kwargs={"observations": observations})
        self.pi = tf.exp(self.log_pi)

        log_Z = 0.0

Instances