from scipy.stats import spearmanr
from tensorboardX import SummaryWriter
from sklearn.metrics import accuracy_score
# from models.u_model import NIMA
# from models.e_model import NIMA
# from models.relic_model import NIMA
from models.relic1_model import NIMA
# from models.relic2_model import NIMA
from dataset import AVADataset
from util import EDMLoss,AverageMeter
import option
f = open('/data/mayme/git/AVA/checkpoint/log_test.txt', 'w')
opt = option.init()
opt.device = torch.device("cuda:{}".format(opt.gpu_id))
def adjust_learning_rate(params, optimizer, epoch):
"""Sets the learning rate to the initial LR
decayed by 10 every 30 epochs"""
lr = params.init_lr * (0.1 ** (epoch // 10))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def get_score(opt,y_pred):
w = torch.from_numpy(np.linspace(1,10, 10))
w = w.type(torch.FloatTensor)
w = w.to(opt.device)
w_batch = w.repeat(y_pred.size(0), 1)
next(generator)
# send state
self._notify(state='running')
# do
while True:
next(generator)
except StopIteration:
self._notify(state='end')
def _load_worker(self, server, args):
assert self._args.worker_path is not None
spec = importlib.util.spec_from_file_location("__importlib_module__",
self._args.worker_path)
m = importlib.util.module_from_spec(spec)
spec.loader.exec_module(m)
g = m.main(server, args)
return g
def _notify(self, **kwargs):
if self._manager is None: return
self._manager.send(name=self._args.name, **kwargs)
if __name__ == '__main__':
import option
option.init()
Process(option.args, option.cluster)()
