rq-wu / underwaterranker Goto Github PK

您好，感谢您出色的工作！我有一个疑问，在使用已训练好的URanker网络计算损失函数时，为什么不需要加with torch.no_grad()？这样optimizer.step()岂不是会更新URanker中的参数吗？如果提了愚蠢的问题，还请见谅！

Hello
Thanks for your great paper.
Could you please upload URankerSet to Google Drive?
Thanks.

Hello, URanker is an interesting work! So, I try to train the network on my data, but it seems to have a bug. When I run the ranker_main_train.py, the runner gets the str path of the yaml file. The contents of yaml are not parsed. It seems the ranker_main_train.py need to add a parsing function, which is coded as follows:

from runner import UIE_Runner
import argparse
import utils

parser = argparse.ArgumentParser()
parser.add_argument('--opt_path', type=str)
args = parser.parse_args()
options = utils.get_option(args.opt_path)  # the adding fuction

runner_train = UIE_Runner(options)
runner_train.main_loop()

我用uranker测试某个数据集，结果出现负值是怎么回事