tensorflow 训练完后如何测试？尝试读取文件，报错了。

Question

我写了一个类，想在fit里训练，在predict里测试训练结果，我的predict函数应该怎么写？我尝试把计算图保存到文件里，然后在predict函数里读取出来，但是总是报错。我用的是tensorflow1.13.1。
下面是相关的代码

    def fit(self, x, y, learning_rate=0.1, iter_num=10):
        (m, n_h, n_w, n_c) = x.shape
        (m, n_y) = y.shape
        self.learning_rate = learning_rate
        self.x_ = tf.placeholder(tf.float32, [None, n_h, n_w, n_c])
        self.y_ = tf.placeholder(tf.float32, [None, n_y])
        self.init_paras()
        self.f_p(self.x_, self.y_)
        self.b_p()
        init = tf.global_variables_initializer()
        with tf.Session() as session:
            session.run(init)
            for cnt in range(iter_num):
                _, cost = session.run([self.cache["optimizer"], self.cache["cost"]], feed_dict={self.x_: x, self.y_: y})
                if cnt % 5 == 0:
                    print("第{}次代价为：{}".format(cnt, cost))
            saver = tf.train.Saver()
            saver.save(session, "./checkpoint_dir/model")

    def predict(self, x, y):
        with tf.Session() as sess:
            saver = tf.train.Saver()
            saver.restore(sess, "./checkpoint_dir/model.meta")
            a = self.cache["output"]
            accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(a, axis=1), tf.argmax(y, axis=1)), tf.float32))
            sess.run(accuracy, feed_dict={self.x_: x, self.y_: y})
            print(accuracy)****

下面是报错的结果

tensorflow.python.framework.errors_impl.DataLossError: Unable to open table file ./checkpoint_dir/model.meta: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?
     [[{{node save_1/RestoreV2}}]]

DataLossError (see above for traceback): Unable to open table file ./checkpoint_dir/model.meta: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?
     [[node save_1/RestoreV2 (defined at ~/PycharmProjects/pythonProject1/test.py:67) ]]

Answer 1

Set up your data format vector and pass it into the Model for inference