JKCooper2

import logging
import os

import gym

# The world's simplest agent!
class RandomAgent(object):
    def __init__(self, action_space):
        self.action_space = action_space

JKCooper2

import logging
import gym
from Solved import SolvedAgent


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

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

JKCooper2

import logging
import gym
from SimulatedAnnealing import SimulatedAnnealingAgent


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

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

JKCooper2

import logging
import gym
from SimulatedAnnealing import SimulatedAnnealingAgent


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

    env = gym.make('Acrobot-v0')

JKCooper2

Unachievable score resulting from action returning 20 rather than 2
Environment currently sets: velocity += (action-1)*0.001 + math.cos(3*position)*(-0.0025) without a bound checks on the input
Will look at adding range check to the environment

JKCooper2

# CARTPOLE MULTI AGENT
# Set up to allow for using a pool of agents

import logging
import gym
from CrossEntropyMethod import CrossEntropyMethodPool

import gym.scoreboard.scoring
import gym.monitoring.monitor

JKCooper2

For Section 1: https://openai.com/requests-for-research/#cartpole

Requirement of environment for algorithm to work:
 - Action space has two discrete actions
 - Ratio of observations can decide to the best action to take

JKCooper2

For part 1 of https://openai.com/requests-for-research/#cartpole

Quite often it doesn't solve (because of local minimum)

JKCooper2

Alteration to [standard hill climbing model](https://gym.openai.com/algorithms/alg_WKinUO3TNabzwPeaD7A)

Uses biased update that allows for worse performance to becomes new standard with reduced probability

For CartPole environment should result in larger percentage of tests solving the problem

JKCooper2

import numpy as np


class Discrete:
    def __init__(self, values):
        self.values = values
        self.max = np.prod(self.values)

    def __validate(self, observation):
        for i in range(len(self.values)):