naf_cartpole.py 文件源码

def train(self, batch):
    _, _, l = tf.get_default_session().run([self.check_numerics, self.train_op, self.loss],
                                 feed_dict={self.input_state: batch.state_1,
                                            self.input_action: batch.action,
                                            self.reward: batch.reward,
                                            self.terminal_mask: batch.terminal_mask,
                                            self.input_state_2: batch.state_2,
                                            base_network.IS_TRAINING: True})
    return l