return 0
elif action == 1:
return 2
elif action == 2:
return 3
elif action == 3:
return 3
if __name__ == "__main__":
env = gym.make('Breakout-ramDeterministic-v4')
env = wrappers.Monitor(env, "./results", force=True)
state_size = 128
action_size = 3
agent = DoubleDQNAgent(state_size, action_size)
agent.load_model()
agent.epsilon = -1 # Q value에 의해서만 움직이게끔.
agent.render = True
scores, episodes = [], []
random_success_cnt = 0
model_success_cnt = 0
# 랜덤액션 진행시
for e in range(EPISODES):
done = False
score = 0
life = 5
env.reset()
for i in range(5):
env.step(1) # 시작 action.
def change_action(action):
if action == 0:
return 0
elif action == 1:
return 2
elif action == 2:
return 3
if __name__ == "__main__":
env = gym.make(
'Breakout-ramDeterministic-v4') #자동으로 frameskip해주는 더 간단한 env.
state_size = 128
action_size = 3 # start action 제외.
agent = DoubleDQNAgent(state_size, action_size)
agent.load_model() #@@@@@@@@@@@@@@@@@@@@모델 로드하기.
#agent.render = True
scores, episodes, avg_q_max_record = [], [], []
for e in range(EPISODES):
done = False
score = 0
epi_step = 0
life = 5
agent.avg_q_max = 0
env.reset()
for i in range(5): # 시작 action 1을 한번만 실행해주면 자주 무시되는듯함. 정확한 이유 모름.
state, _, _, _ = env.step(1) # 시작 action.
state = np.reshape(state,
[1, 128]) # (1,128) => model의 메소드에는 이 형식으로 넣어야함.
# 0,1,2 =>0,2,3 action으로 치환해주기.
def change_action(action):
if action == 0:
return 0
elif action == 1:
return 2
elif action == 2:
return 3
if __name__ == "__main__":
env = gym.make('Breakout-ram-v0')
state_size = 128
action_size = 3 # start action 제외.
agent = DoubleDQNAgent(state_size, action_size)
#agent.load_model() #@@@@@@@@@@@@@@@@@@@@모델 로드하기.
scores, episodes, avg_q_max_record = [], [], []
for e in range(EPISODES):
done = False
score = 0
epi_step = 0
life = 5
agent.avg_q_max = 0
env.reset()
for i in range(5): # 시작 action 1을 한번만 실행해주면 자주 무시되는듯함. 정확한 이유 모름.
state, _, _, _ = env.step(1) # 시작 action.
state = np.reshape(state,
[1, 128]) # (1,128) => model의 메소드에는 이 형식으로 넣어야함.