OpenAI Gym FrozenLake-v0 random search https://gym.openai.com/evaluations/eval_iinFj8fUSvOOYHWXwjjAw

frozenlake_v0_random.py

import gym
import numpy as np
from gym import wrappers

env = gym.make("FrozenLake-v0")
env = wrappers.Monitor(env, '/tmp/cartpole-experiment-2')

env.reset();
n_states = env.observation_space.n
n_actions = env.action_space.n

def get_random_policy():
    return np.random.choice(n_actions, size=n_states)

def sample_reward(env, policy, t_max=100):
    s = env.reset()
    total_reward = 0
    
    for _ in range(t_max):
        s, reward, is_done, _ = env.step(policy[s])
        total_reward += reward
        if is_done:
            break
    return total_reward

def evaluate(policy, n_times=100):
    rewards = [sample_reward(env, policy) for _ in range(n_times)]
    return float(np.mean(rewards))

def print_policy(policy):
    lake = "SFFFFHFHFFFHHFFG"

    arrows = ['>^v<'[a] for a in policy]
    
    signs = [arrow if tile in "SF" else tile for arrow, tile in zip(arrows, lake)]
    
    for i in range(0, 16, 4):
        print(' '.join(signs[i:i+4]))
        
best_policy = None
best_score = -float('inf')

from tqdm import tqdm
for i in tqdm(range(10000)):
    policy = get_random_policy()
    score = evaluate(policy)
    if score > best_score:
        best_score = score
        best_policy = policy
        print("\n")
        print("New best score:", score)
        print("Best policy:")
        print_policy(best_policy)
        
env.close()