Chapter5 - Pong

[ma-ra-bu]

The German translation of the book (p. 134) promises:
"However, if you want, you can easily adapt the algorithm to a more difficult game like Pong in OpenAI Gym; you can find such an implementation on the GitHub page for this chapter: http://mng.bz/JzKp."
Unfortunately I couldn't find anything about it! Anyone know where the code is?

Auf Seite 134 im Buch (Hanser) steht:
"Wenn Sie möchten, können Sie den Algorithmus jedoch leicht an ein schwierigeres Spiel wie Pong in OpenAI Gym anpassen; eine solche Implementierung finden Sie auf der GitHub-Seite zu diesem Kapitel: http://mng.bz/JzKp."
Leider habe ich dazu nichts gefunden! Wer weiss, wo der Code ist?

[walebey]

Look at the comments in all Ch5 notebooks. But that didn´t help me, because the code is not working.
Running the code with pong gives following error message:
....
File "", line 41, in run_episode
raw_state = np.array(worker_env.env.state)
AttributeError: 'AtariEnv' object has no attribute 'state'

[outlace]

We will try to update the code. I think the OpenAI Gym library has changed since publication and is breaking some of the code unfortunately.

[walebey]

As of 2020: for Atari environments, you need to use env.env.ale.getScreenRGB() instead to access the state.
But it doesn´t seem to work for both games.

import gym
env = gym.make("CartPole-v1")
env = gym.make("Pong-v0")
observation = env.reset()
print(observation)
print("----------------")
if(type(env.env) is gym.envs.classic_control.cartpole.CartPoleEnv):
print(env.env.state)
else:
print(env.env.ale.getScreenRGB())