cur_state = env.reset()
step_counter = 0
while True:
step_counter += 1
if key:
env.display()
action = agent.choose_action(cur_state)
next_state, reward = env.move(action)
agent.learn(
cur_state=cur_state,
action=action,
reward=reward,
next_state=next_state
)
cur_state = next_state
if reward != 0:
break
if reward > 0:
successful_step_counter_arr.append(step_counter)
elif reward < 0:
failed_step_counter_arr.append(step_counter)
if key:
print(
from env import TicTacToeEnv
from agent import QLearningAgent
env = TicTacToeEnv()
agent = QLearningAgent(env)
for game_nr in range(1000000):
if game_nr % 10000 == 0:
print(game_nr)
done = False
s = env.reset().copy()
# print('Init', s)
while not done:
a = agent.take_action(s)
r, s_, done, _ = env.step(a)
agent.learn(s, a, r, s_, done)
# print(s, a, r, s_, done)
s = s_.copy()
V = pd.DataFrame.from_dict(agent._V,
orient='index',
dtype=np.float32,
columns=['V'])
N = pd.DataFrame.from_dict(agent._N,
orient='index',
dtype=np.uint32,
columns=['N'])
df = V.merge(N, how='left', left_index=True, right_index=True)
states = pd.DataFrame(df.index.values.tolist(), index=df.index)
res = states.merge(V, how='left', left_index=True, right_index=True).merge(
N, how='left', left_index=True, right_index=True).reset_index(drop=True)