global_step = 0
scores, episodes = [], []

for e in range(EPISODES):
    done = False
    score = 0
    # env 초기화
    state = env.reset()
    state = np.reshape(state, [1, 15])

    while not done:
        global_step += 1
        # 현재 상태에 대한 행동 선택
        action = agent.get_action(state)

def get_action(self, state):
    policy = self.model.predict(state)[0]
    return np.random.choice(self.action_size, 1, p=policy)[0]

def __init__(self):
    self.load_model = False
    # 가능한 모든 행동 정의
    self.action_space = [0, 1, 2, 3, 4]
    # 상태와 행동의 크기 정의
    self.action_size = len(self.action_space)
    self.state_size = 15
    self.discount_factor = 0.99
    self.learning_rate = 0.001

    self.model = self.build_model()
    self.optimizer = self.optimizer()
    self.states, self.actions, self.rewards = [], [], []

    if self.load_model:
        self.model.load_weights('./save_model/reinforce_trained.h5')