从先前的检查点还原Tensorflow模型

Question

I am tuning my model for 200 episodes by training it 2 times each for 100 episodes. The first run for 100 eps is finished and right now 2nd run is going on. My problem is I want to restore the model after it finished the first run that is from the 100th episode. And now that I have already started the 2nd run, it has overwritten the checkpoint files.

I came across this post which says tensorflow keeps last 5 models. How can I retrieve the model parameters from the previous checkpoint which was created after the first run?

Answer 1

Tensorflow可以在一次运行中保存最好的5个模型，您已经开始了另一次新运行，恐怕您的模型会丢失，您将不得不对其进行重新训练，经验法则始终保持模型的命名动态，因此不会被替换， 例如：

model_name = 'Model_{:d}'.format(episode_number) + '.ckpt'