def main():
new_map = ["SFFF", "FHFH", "FFFH", "HFFG"]
env = FrozenLakeEnv(desc=new_map, is_slippery=IS_SLIPPERY)
env = env.unwrapped
succeed_episode = 0
for i_episode in range(1000000):
if use_random_map and i_episode % 10 == 0:
env.close()
new_map = random_map(HOLE_NUM)
env = FrozenLakeEnv(desc=new_map, is_slippery=IS_SLIPPERY)
env = env.unwrapped
pos = env.reset()
state = encode_state(new_map, pos)
ep_r = 0
while True:
a = select_action(state)
pos_next, r, done, info = env.step(a)
ep_r += r
#state_next = encode_state(new_map, pos_next)
if args.render:
env.render()
model.rewards.append(r)
if done:
break
finish_episode()
episode_durations.append(ep_r)
if ep_r > 0:
# EPSILON = 1 - 1. / ((i_episode / 500) + 10)
succeed_episode += 1
if i_episode % 1000 == 1:
print('EP: {:d} succeed rate {:4f}'.format(i_episode,
succeed_episode / 1000))
succeed_episode = 0
if i_episode % 5000 == 1:
plot_durations()
# print(episode, qtable)
# print(total_rewards)
print("Score over time: " + str(sum(rewards) / total_episodes))
print(qtable)
env.reset()
for episode in range(1):
state = env.reset()
print("state",state)
step = 0
done = False
print("****************************************************")
print("EPISODE ", episode)
for step in range(max_steps):
env.render()
# Take the action (index) that have the maximum expected future reward given that state
action = np.argmax(qtable[state, :])
new_state, reward, done, info = env.step(action)
if done:
print("done..........")
# env.render()
break
state = new_state
env.close()
