The agent hit the global step limit, how do I restore from checkpoints and restore trainning

[syhdog]

I am working on a ideal that requires long time training, about 10 days. I forgot to modify the global step limit. So the agent stopped at 100M step. I want to restore the model and go on my training. I have been look through the code and wandered what should I do.

Sincerely thank you to open this project, it has been a very very respectable work and help a lot with my research. Truly, we find it really enjoyable to develop agents. Thank you a lot.

[KaixiangLin]

In worker.py, add those two lines to specify the variables you want to restore:

    variables_to_restore = [v for v in tf.all_variables() if v.name.startswith("global")]
    pre_train_saver = FastSaver(variables_to_restore)

Add one more line in init_fn

    def init_fn(ses):
        logger.info("Initializing all parameters.")
        ses.run(init_all_op)
        pre_train_saver.restore(ses,
                                "THE_PATH_TO_YOUR_MODEL/model.ckpt-4986751")

[syhdog]

Than you very much, it's be a great help.