dc2ba0a4b217d154e9e790be30cb0c41994b44d3,softlearning/algorithms/sac.py,SAC,_init_actor_update,#SAC#,246

Before Change

            name="policy_loss_optimizer",
        ).minimize(
            loss=policy_loss,
            var_list=self._policy.get_params_internal())

        vf_train_op = tf.train.AdamOptimizer(
            self._vf_lr,

After Change

            optimizer=tf.train.AdamOptimizer,
            increment_global_step=False,
            name="policy_optimizer",
            summaries=[
                "loss", "gradients", "gradient_norm", "global_gradient_norm"
            ])

        vf_train_op = tf.contrib.layers.optimize_loss(
            self._vf_loss_t,
            self.global_step,
            learning_rate=self._vf_lr,
            optimizer=tf.train.AdamOptimizer,
            increment_global_step=True,
            name="vf_optimizer",
            summaries=[
                "loss", "gradients", "gradient_norm", "global_gradient_norm"
            ])

        self._training_ops += [policy_train_op, vf_train_op]

    def _init_target_ops(self):
        Create tensorflow operations for updating target value function.

Instances