FrozenLake4x4-handmade.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

# Just a handmade algorithm. NOT AI.

import gym

### MAP ###
# SFFF
# FHFH
# FFFH
# HFFG
###########

# 0:L, 1:D, 2:R, 3:U
ACTIONS = (
     0,  3,  3,  3,
     0, -1,  0, -1,
     3,  1,  0, -1,
    -1,  2,  1, -1,
)

EPISODE_COUNT = 1000

DEBUG = False
#DEBUG = True


def get_action(ob):
    act = ACTIONS[ob]
    assert 0 <= act <= 3
    return act

def main():
    env = gym.make("FrozenLake-v0")
    env.monitor.start("rec")

    reward_total = 0.0
    for episode in range(EPISODE_COUNT):
        ob = env.reset()
        if DEBUG: env.render()
        while True:
            ob, reward, done, info = env.step(get_action(ob))
            if DEBUG:
                env.render()
                print(ob, reward, done, info)
            if done:
                reward_total += reward
                break

    env.monitor.close()

    print("episodes:       {}".format(EPISODE_COUNT))
    print("total reward:   {}".format(reward_total))
    print("average reward: {:.2f}".format(reward_total / EPISODE_COUNT))

if __name__ == "__main__": main()