device=device,
n_actions=1,
max_action=env.action_space.high,
n_features=env.observation_space.shape[0],
learning_rate=0.001,
gamma=0.99,
tau=0.01,
noiseStart=np.max(env.action_space.high),
noiseEnd=0.2,
noiseDecayFreq=10000,
updateTargetFreq=600,
mSize=10000,
batchSize=100,
startTrainSize=100,
)
if __name__ == "__main__":
RENDER = False # 顯示模擬會拖慢運行速度, 等學得差不多了再顯示
env.seed(1) # 固定隨機種子 for 再現性
# env = env.unwrapped # 不限定 episode
torch.manual_seed(500) # 固定隨機種子 for 再現性
env_run(
env=env,
agent=agent,
callerPath=__file__,
stopRewardFunc=lambda x: x > -100,
RENDER=RENDER,
test=False,
)
from Gym.tools.utils import env_run
from .train import env, agent
if __name__ == "__main__":
RENDER = True # 顯示模擬會拖慢運行速度, 等學得差不多了再顯示
env_run(env=env,
agent=agent,
callerPath=__file__,
RENDER=RENDER,
test=True)
