Cannot load bots trained with self_play_train.py

Thank you for your efforts in releasing this exciting library. I am trying to train an agent, and then load that agent to view its behaviour rendered in a pygame window. However, the agents that I train are cannot be loaded into a PermissiveModel because they do not have ‘function_tables’ (or signatures for that matter).

I add the following to examples/rllib/self_play_train.py

   # 7. Train once
   trainer.train()

   # Additions
   trainer.get_policy("av").export_model("./some_path")
   from meltingpot.python import bot
   agent = bot.SavedModelPolicy("./some_path")

Running self_play_train.py results in the following error:

Traceback (most recent call last):
  File "<repo_path>/meltingpot/examples/rllib/self_play_train.py", line 117, in <module>
    main()
  File "<repo_path>/meltingpot/examples/rllib/self_play_train.py", line 114, in main
    agent = bot.SavedModelPolicy("./some_path")
  File "<repo_path>/meltingpot/meltingpot/python/bot.py", line 103, in __init__
    self._model = permissive_model.PermissiveModel(model)
  File "<repo_path>/meltingpot/meltingpot/python/utils/bots/permissive_model.py", line 51, in __init__
    self._tables = self.model.function_tables()
AttributeError: 'AutoTrackable' object has no attribute 'function_tables'

Would it please be possible for you to extend this example to include the saving and loading of bots?

[Muff2n]

@duenez that PR #30 fixed issue #26, not this one. The confusion has come because I removed my first PR and split it into two. PR #30 was the first part, I will submit the second part that addresses this issue next week.

[duenez]

This should be fixed now.