README

Solved version of problem to determine minimum knowledge required by agent to successfully complete task
Observation values may be off as I threw this together pretty quick just as a proof of concept

CartPole-v0.py

import logging
import gym
from Solved import SolvedAgent


def main():
    logger = logging.getLogger()
    logger.setLevel(logging.DEBUG)

    env = gym.make('CartPole-v0')
    agent = SolvedAgent()

    outdir = '/tmp/' + agent.name + '-results'
    env.monitor.start(outdir, force=True)

    episode_count = 200
    max_steps = 200
    reward = 0
    done = False

    for i in xrange(episode_count):
        ob = env.reset()

        for j in xrange(max_steps):
            action = agent.act(ob, reward, done)
            ob, reward, done, _ = env.step(action)
            if done:
                break

    env.monitor.close()

if __name__ == '__main__':
    main()

Solved.py

class SolvedAgent(object):
    def __init__(self):
        self.name = 'solved'

    def act(self, observation, reward, done):

        # Guesses at observations are:
        # observation[0] = pole speed
        # observation[1] = pole top pos
        # observation[2] = pole angle
        # observation[3] = block speed

        if (observation[2] > 0 and observation[3] > -1) or observation[3] > 1:
            return 1

        return 0

Thank you for the feedback. I wasn't intending this to be put on the reviewed list at it's not doing any actual learning, but it could be useful when thinking about how to determine what information is necessary to solve a problem versus what information is available.