def learn(self, env, episode_count=1000, gamma=0.9,
learning_rate=0.1, render=False, report_interval=50):
"""QLearningAgent.learn
env:環境のデータが格納された変数です。
episode_count:エピソード回数を指定します。default:1000
gamma:割引率を指定します。default:0.9
render:画面に様子を表示するかどうか設定します。default:False
report_interval:ログを保存する間隔を設定します。default:50"""
self.init_log()
actions = list(range(env.action_space.n))
self.Q = defaultdict(lambda: [0] * len(actions))
for e in tqdm(range(episode_count)):
s = env.reset()
done = False
while not done:
if render:
cuitools.reset()
env.render()
time.sleep(0.01)
a = self.policy(s, actions)
n_state, reward, done, info = env.step(a)
gain = reward + gamma * max(self.Q[n_state])
estimated = self.Q[s][a]
self.Q[s][a] += learning_rate * (gain - estimated)
s = n_state
else:
self.log(reward)
if e != 0 and e % report_interval == 0:
pass
def learn(self,
env,
episode_count=1000,
gamma=0.9,
render=False,
report_interval=50):
"""MonteCarloAgent.learn
env:環境のデータが格納された変数です。
episode_count:エピソード回数を指定します。default:1000
gamma:割引率を指定します。default:0.9
render:画面に様子を表示するかどうか設定します。default:False
report_interval:ログを保存する間隔を設定します。default:50"""
self.init_log()
actions = list(range(env.action_space.n))
self.Q = defaultdict(lambda: [0] * len(actions))
N = defaultdict(lambda: [0] * len(actions))
for e in tqdm(range(episode_count)):
s = env.reset()
done = False
# Play until the end of episode.
experience = []
while not done:
if render:
cuitools.reset()
env.render()
time.sleep(0.01)
a = self.policy(s, actions)
n_state, reward, done, info = env.step(a)
experience.append({"state": s, "action": a, "reward": reward})
s = n_state
else:
self.log(reward)
# Evaluate each state, action.
for i, x in enumerate(experience):
s, a = x["state"], x["action"]
# Calculate discounted future reward of s.
G, t = 0, 0
for j in range(i, len(experience)):
G += math.pow(gamma, t) * experience[j]["reward"]
t += 1
N[s][a] += 1 # count of s, a pair
alpha = 1 / N[s][a]
self.Q[s][a] += alpha * (G - self.Q[s][a])
if e != 0 and e % report_interval == 0:
pass
def train(self,
env,
episode_count=1000,
gamma=0.9,
learning_rate=0.1,
render=False,
report_interval=50):
"""actor_critic.train
env:環境のデータが格納された変数です。
episode_count:エピソード回数を指定します。default:1000
gamma:割引率を指定します。default:0.9
render:画面に様子を表示するかどうか設定します。default:False
report_interval:ログを保存する間隔を設定します。default:50"""
actor = self.actor_class(env)
critic = self.critic_class(env)
actor.init_log()
for e in tqdm(range(episode_count)):
s = env.reset()
done = False
while not done:
if render:
cuitools.reset()
env.render()
time.sleep(0.01)
a = actor.policy(s)
n_state, reward, done, info = env.step(a)
gain = reward + gamma * critic.V[n_state]
estimated = critic.V[s]
td = gain - estimated
actor.Q[s][a] += learning_rate * td
critic.V[s] += learning_rate * td
s = n_state
else:
actor.log(reward)
if e != 0 and e % report_interval == 0:
pass
# actor.show_reward_log(episode=e)
return actor, critic
_ = plus_proche(T, ref, diss)
fin = time()
duree = fin - debut
print("Duree (moyen): ", duree)
n = 100000 # grand
T = []
for _ in range(n):
T.append(randint(0, n))
ref = randint(0, n)
debut = time()
_ = plus_proche(T, ref, diss)
fin = time()
duree = fin - debut
print("Duree (grand): ", duree)
print(f"Duree (grand): {tps_plus_proche(n)}")
N = []
Tps = []
for n in tqdm(range(10, 10**5, 200)):
N.append(n)
# calcul
Tps.append(tps_plus_proche(n))
#affichage de la courbe N vs Tps
plt.figure()
plt.plot(N, Tps)
plt.show()
# tqdm a installer
#from tqdm import tqdm # programme python classique
#from tqdm.notebook import tqdm # dans un notebook
from tqdm.gui import tqdm # pour thonny
from time import sleep
from random import randint
for i in tqdm(range(5), desc='Boucle sur i'):
for j in tqdm(range(3), desc=f'Boucle sur j', leave=False):
duree_sommeil = randint(1, 2)
sleep(duree_sommeil)