基於tensorflow實現，導入預訓練模型之後，將global_step, epoch重新設置爲0

問題：使用tensoflow導入訓練模型參數後，eoch, global_step仍然是之前訓練所保留的值。因此，在導入預訓練模型參數之後，需要將epoch, global_step的值重新置0.

部分參考代碼：

with tf.Session(config=config) as sess:
            # Initialize all variables
            sess.run(tf.global_variables_initializer())
            print("{}: Start training...".format(datetime.datetime.now()))

            # summary writer for tensorboard
            train_summary_writer = tf.summary.FileWriter(FLAGS.log_dir + '/train', sess.graph)
            test_summary_writer = tf.summary.FileWriter(FLAGS.log_dir + '/test', sess.graph)

            # restore from checkpoint
            if FLAGS.restore_training:
                # check if checkpoint exists
                if os.path.exists(checkpoint_prefix + "-latest"):
                    print("{}: Last checkpoint found at {}, loading...".format(datetime.datetime.now(),
                                                                               FLAGS.checkpoint_dir))
                    latest_checkpoint_path = tf.train.latest_checkpoint(FLAGS.checkpoint_dir,
                                                                        latest_filename="checkpoint-latest")
                    saver.restore(sess, latest_checkpoint_path)
            if FLAGS.pre_training:
                print("{}: pre_train checkpoint found at {}, loading...".format(datetime.datetime.now(),
                                                                                FLAGS.pre_checkpoint_path))
                saver.restore(sess, FLAGS.pre_checkpoint_path)
                # start_epoch.eval()[0]= 0
                op1 =tf.assign(global_step, 0)
                op2 =tf.assign(start_epoch, [0])
                sess.run([op1, op2])
                # sess.run(start_epoch)
            print("{}: Last checkpoint epoch: {}".format(datetime.datetime.now(), start_epoch.eval()[0]))
            print("{}: Last checkpoint global step: {}".format(datetime.datetime.now(),
                                                               tf.train.global_step(sess, global_step)))

其中，將epoch, global_step重新設置爲0的代碼如下：

op1 =tf.assign(global_step, 0)
op2 =tf.assign(start_epoch, [0])
sess.run([op1, op2])