Simple random action agent playing qbert

atari-4-qbert.py

#!/usr/bin/python

# My first Atari simulation on the AIgym
# Not much happening yet

import gym
episodes = 500
max_time = 10000

env = gym.make('Qbert-v0')
print env.action_space
print env.observation_space
env.monitor.start('/tmp/atari-experiment-3', force=True)
env.reset()

for episode in range(episodes):
    obervation = env.reset()
    total_reward = 0
    for time in range(max_time):
        env.render()
        action = env.action_space.sample()
        observation, reward, done, info = env.step(action)
        total_reward = total_reward + reward
        if done:
            print "Episode {}:".format(episode)
            print "  completed in {} steps".format(time+1)
            print "  total_reward was {}".format(total_reward)
            break

env.monitor.close()