def __init__(self, args):
self.tasks = [self.train_step]
self.args = args
self.device = args.device
self.model = Model().to(self.device)
self.optimizer = optim.Adam(self.model.parameters(),
lr=args.lr,
amsgrad=True)
self.writer = SummaryWriter()
self.scores = deque(maxlen=250)
self.saved_model_score = -1e10
self._scores = deque(maxlen=10000)
self.n_errors = 0
self.env = Environment(args)
self.batch_buffer = list()
self.frames = 0
self.running = True
self.debug_mode = False
backup_dir(src=str(Path(__file__).parent),
dst=str(Path(self.writer.log_dir) / 'backup.zip'))
cprint('READY', 'green')
def __init__(self, args):
self.tasks = [self.train_step]
self.args = args
self.device = args.device
self.model = Model().to(self.device)
# 만약 model.pt가 있다면, model load
# 이전에 학습된 모델이라고 가정한다.
model_path = Path(__file__).parent / 'model.pt'
if os.path.isfile(model_path):
# checkpoint = torch.load(model_path, map_location=torch.device('cuda')) # gpu
checkpoint = torch.load(model_path, map_location=torch.device('cpu')) # cpu
# checkpoint = torch.load(model_path) # gpu environment
self.model.load_state_dict(checkpoint['model_state_dict'])
# gpu
# self.model.to(self.device)
# parameter for model save
# 현재 score가 일정 수치를 넘어 model save가 되면 False로 바뀐다.
# 다시 score가 일정 수치 이하로 내려가면 True로 바뀐다.
self.possible_model_save = True
# 만약 model.pt가 있다면, optimizer load
self.optimizer = optim.Adam(self.model.parameters(), lr=args.lr, amsgrad=True)
if os.path.isfile(model_path):
# checkpoint = torch.load(model_path, map_location=torch.device('cuda')) # gpu
checkpoint = torch.load(model_path, map_location=torch.device('cpu')) # cpu
# checkpoint = torch.load(model_path)
self.optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
self.writer = SummaryWriter()
self.scores = deque(maxlen=250)
self.saved_model_score = -1e10
self._scores = deque(maxlen=10000)
self.n_errors = 0
# for pool
self.mean_score = -1
self.mybot_version = None
self.bot_cache_path = Path(__file__).parent / ('bot_cache.txt')
# version 파일이 있다면 내용을 가져오기.
if os.path.isfile(self.bot_cache_path):
with open(self.bot_cache_path, 'r') as cache_reader:
Info_cache = cache_reader.read()
# 비어 있지 않을 때 내용을 가져와 mybot_version에 넣는다.
if Info_cache != '':
splited_Info = Info_cache.split()
self.mybot_version = int(splited_Info[0])
self.saved_model_score = float(splited_Info[1])
else:
self.mybot_version = 1
# 혹시 모르는 cache 파일은 없고 모델만 생겼을 때를 대비.
for i in range(1, 5):
model_path = Path(__file__).parent / ('model' + str(i) + '.pt')
if os.path.isfile(model_path):
self.mybot_version = (self.mybot_version % 4) + 1
else:
self.mybot_version = 1
# 혹시 모르는 cache 파일은 없고 모델만 생겼을 때를 대비.
for i in range(1, 5):
model_path = Path(__file__).parent / ('model' + str(i) + '.pt')
if os.path.isfile(model_path):
self.mybot_version = (self.mybot_version % 4) + 1
self.env = Environment(args)
self.batch_buffer = list()
self.frames = 0
self.running = True
self.debug_mode = False
backup_dir(
src=str(Path(__file__).parent),
dst=str(Path(self.writer.log_dir) / 'backup.zip')
)
cprint('READY', 'green')