"""
# MACROS
LEFT = 0
DOWN = 1
RIGHT = 2
UP = 3
# Key mapping
arrow_keys = {'\x1b[A': UP, '\x1b[B': DOWN, '\x1b[C': RIGHT, '\x1b[D': LEFT}
# is_slippery True
env = gym.make('FrozenLake-v0')
env.reset()
print_utils.clear_screen()
env.render() # Show the initial board
while True:
# Choose an action from keyboard
key = readchar.readkey()
if key not in arrow_keys.keys():
print("Game aborted!")
break
action = arrow_keys[key]
state, reward, done, info = env.step(action)
# Show the board after action
print_utils.clear_screen()
def main():
"""Main"""
frozone_lake_env = gym.make("FrozenLake-v0")
# Initialize table with all zeros
Q = np.zeros([N_STATES, N_ACTIONS])
# Set learning parameters
# create lists to contain total rewards and steps per episode
rewards = []
for i in range(N_EPISODES):
# Reset environment and get first new observation
state = frozone_lake_env.reset()
episode_reward = 0
done = False
# The Q-Table learning algorithm
while not done:
# Choose an action by greedily (with noise) picking from Q table
action = np.argmax(Q[state, :] + np.random.randn(1, N_ACTIONS) /
(i + 1))
# Get new state and reward from environment
new_state, reward, done, _ = frozone_lake_env.step(action)
# Update Q-Table with new knowledge using learning rate
Q[state,
action] = (1 -
LEARNING_RATE) * Q[state, action] + LEARNING_RATE * (
reward + DISCOUNT_RATE * np.max(Q[new_state, :]))
episode_reward += reward
state = new_state
rewards.append(episode_reward)
print("Score over time: " + str(sum(rewards) / N_EPISODES))
print("Final Q-Table Values")
for i in range(10):
# Reset environment and get first new observation
state = frozone_lake_env.reset()
episode_reward = 0
done = False
# The Q-Table learning algorithm
while not done:
# Choose an action by greedily (with noise) picking from Q table
action = np.argmax(Q[state, :])
# Get new state and reward from environment
new_state, reward, done, _ = frozone_lake_env.step(action)
print_utils.clear_screen()
frozone_lake_env.render()
time.sleep(.1)
# Update Q-Table with new knowledge using learning rate
Q[state, action] = (1 - LEARNING_RATE) * Q[state, action] \
+ LEARNING_RATE * (reward + DISCOUNT_RATE * np.max(Q[new_state, :]))
episode_reward += reward
state = new_state
if done:
print("Episode Reward: {}".format(episode_reward))
print_utils.print_result(episode_reward)
rewards.append(episode_reward)
frozone_lake_env.close()