def __init__(self, env):
self.sess = tf.Session()
K.set_session(self.sess)
# hyperparameters
self.GAMMA = 0.95
self.BATCH_SIZE = 32
self.ACTOR_LEARNING_RATE = 0.0001
self.CRITIC_LEARNING_RATE = 0.001
self.env = env
# get state dimension
self.state_dim = env.observation_space.shape[0]
# get action dimension
self.action_dim = env.action_space.shape[0]
# get action bound
self.action_bound = env.action_space.high[0]
# create actor and critic networks
self.actor = Actor(self.sess, self.state_dim, self.action_dim,
self.action_bound, self.ACTOR_LEARNING_RATE)
self.critic = Critic(self.state_dim, self.action_dim,
self.CRITIC_LEARNING_RATE)
# initialize for later gradient calculation
self.sess.run(
tf.global_variables_initializer()) #<-- no problem without it
# save the results
self.save_epi_reward = []
def __init__(self, env):
# 텐서플로우 세션 설정
self.sess = tf.Session()
K.set_session(self.sess)
# 하이퍼파라미터
self.GAMMMA = .95
self.BATCH_SIZE = 32
self.ACTOR_LEARNING_RATE = .0001
self.CRITIC_LEARNING_RATE = .001
#환경
self.env = env
#상태변수 차원
self.state_dim = env.observation_space.shape[0]
#행동 차원(dimension)
self.action_dim = env.action_space.shape[0]
#행동의 최대 크기
self.action_bound = env.action_space.high[0]
#액터 신경망 및 크리틱 신경망 생성
self.actor = Actor(self.sess, self.state_dim, self.action_dim,
self.action_bound, self.ACTOR_LEARNING_RATE)
self.critic = Critic(self.state_dim, self.action_dim,
self.CRITIC_LEARNING_RATE)
# 그래디언트 계산을 위한 초기화
self.sess.run(tf.global_variables_initializer())
#에피소드에서 얻은 총 보상값 저장
self.save_epi_reward = []
def __init__(self, env):
# hyperparameters
self.GAMMA = 0.95
self.BATCH_SIZE = 32
self.ACTOR_LEARNING_RATE = 0.0001
self.CRITIC_LEARNING_RATE = 0.001
self.env = env
# get state dimension
self.state_dim = env.observation_space.shape[0]
# get action dimension
self.action_dim = env.action_space.shape[0]
# get action bound
self.action_bound = env.action_space.high[0]
# create actor and critic networks
self.actor = Actor(self.state_dim, self.action_dim, self.action_bound,
self.ACTOR_LEARNING_RATE)
self.critic = Critic(self.state_dim, self.action_dim,
self.CRITIC_LEARNING_RATE)
# save the results
self.save_epi_reward = []
def __init__(self, env): # 클래스 초기화 메서드
# hyper parameters
self.GAMMA = 0.95
self.BATCH_SIZE = 32
self.ACTOR_LEARNING_RATE = 0.0001
self.CRITIC_LEARNING_RATE = 0.001
# 환경
self.env = env
# get state dimension
self.state_dim = env.observation_space.shape[0]
# get action dimension
self.action_dim = env.action_space.shape[0]
# get action bound (행동의 최대 크기)
self.action_bound = env.action_space.high[0]
## A2C 알고리즘 1. critic 과 actor 신경망의 파라메터 phi (critic 신경망) 와 theta (actor 신경망)를 초기화한다.
# create actor and critic networks
self.actor = Actor(self.state_dim, self.action_dim, self.action_bound, self.ACTOR_LEARNING_RATE)
self.critic = Critic(self.state_dim, self.action_dim, self.CRITIC_LEARNING_RATE);
# save the results (에피소드에서 얻은 총 보상값을 저장하기 위한 변수)
self.save_epi_reward = [];