random_search.py

import numpy as np
import itertools
import gym
import random
env = gym.make("BowlingNoFrameskip-v4")

def enact(step, partitions, actions):
    step = step % partitions[-1]
    for p, a in zip(partitions, actions):
        if step < p:
            return a

def evaluate(params):
    done = False
    i = 0
    G = 0
    env.reset()
    while not done:
        #env.render()
        i += 1
        action = enact(i, params[0], params[1])
        _, reward, done, _ = env.step(action)
        G += reward
    return G

returns = []
optimal_value = -1
n_episodes = 0
while True:
    n_episodes+= 1
    partitions = [random.choice(range(10, 1000, 10)) for i in range(3)]
    actions = [random.choice(range(3)) for i in range(3)]
    partitions.sort()
    if 1 not in actions:
        continue
    print(n_episodes, "evaluating...", [partitions, actions], '...', end='')
    val = evaluate([partitions, actions])
    if val > optimal_value:
        optimal_value = val
        optimal_policy = [partitions, actions]
    print("has value", val, '; optimal:', optimal_value)