a55f9be217b2320cec7c7dc44a8245496f851af5,softlearning/algorithms/sac.py,SAC,_init_placeholders,#SAC#,120

Before Change

        )

        if self._store_extra_policy_info:
            self._log_pis_ph = tf.placeholder(
                tf.float32,
                shape=(None, 1),
                name="log_pis",
            )
            self._raw_actions_ph = tf.placeholder(
                tf.float32,
                shape=(None, *self._action_shape),
                name="raw_actions",
            )

    def _get_Q_target(self):
        next_actions = self._policy.actions([self._next_observations_ph])
        next_log_pis = self._policy.log_pis(

After Change

            - reward
            - terminals
        
        self._placeholders = {
            "iteration": tf.placeholder(
                tf.int64, shape=None, name="iteration",
            ),
            "observations": tf.placeholder(
                tf.float32,
                shape=(None, *self._observation_shape),
                name="observation",
            ),
            "next_observations": tf.placeholder(
                tf.float32,
                shape=(None, *self._observation_shape),
                name="next_observation",
            ),
            "actions": tf.placeholder(
                tf.float32,
                shape=(None, *self._action_shape),
                name="actions",
            ),
            "rewards": tf.placeholder(
                tf.float32,
                shape=(None, 1),
                name="rewards",
            ),
            "terminals": tf.placeholder(
                tf.bool,
                shape=(None, 1),
                name="terminals",
            ),
        }

    def _get_Q_target(self):
        next_observations = self._placeholders["next_observations"]
        next_actions = self._policy.actions([next_observations])

Instances