1

We are making a reinforcement learning algorithm for our class. And in this progress we want to make a savestate for a nes-game which is made using the gym-retro package. We have tried using pickle to save the game environment but it does not pickle gamedate objects.

TypeError: can't pickle GameData objects

Is there any way to save the gamestate that we don't know about? This is part of the code we use:

def SaveEnv(self, env, level):
    self.env[level] = open('store.pckl', 'wb')
    envDict = {}
    envDict[level] = env
    pk.dump(envDict, self.env[level])
    self.env[level].close()
    print("Save Successful")    

env = retro.make(game='TinyToonAdventures-Nes')
env.reset()
SaveEnv(env,level)
Tijmbee
  • 11
  • 1

1 Answers1

0

I think that I'm a bit late and don't know if it will be useful for you, but, for anyone that runs into the same problem, here is my solution.

For saving your state, you would need to import the gzip library and, in the simulation (when you're in your desired state), do this:

content = env.em.get_state()
with gzip.open("the_saved_state.state", 'wb') as f:
    f.write(content)

And now, for loading:

f = gzip.open('the_saved_state.state','rb')
file_content = f.read()
f.close()
env.em.set_state(file_content)

Additionally, I want to give credit to this fella: https://github.com/openai/retro/issues/235 (from whom I based on)

Jorvan
  • 109
  • 2
  • 13