1918f160796525d2b5f241f81d3224a941418935,agents/ppo_agent.py,PPOAgent,update_kl_coefficient,#PPOAgent#,205

Before Change

            self.tp.sess.run(self.increase_kl_coefficient, feed_dict={self.kl_coefficient: kl_coefficient})
        elif self.total_kl_divergence_during_training_process < 0.7 * kl_target:
            // kl too low => decrease regularization
            self.tp.sess.run(self.decrease_kl_coefficient, feed_dict={self.kl_coefficient: kl_coefficient})
        screen.log_title("KL penalty coefficient change = {} -> {}".format(
            kl_coefficient, self.tp.sess.run(self.policy_network.online_network.output_heads[0].kl_coefficient)))

    def post_training_commands(self):

After Change

                new_kl_coefficient,
                self.policy_network.online_network.output_heads[0].kl_coefficient_ph)

        screen.log_title("KL penalty coefficient change = {} -> {}".format(kl_coefficient, new_kl_coefficient))

    def post_training_commands(self):
        if self.tp.agent.use_kl_regularization:

Instances