xiaoleimiao / udtl_lable_inconsistent Goto Github PK

您好 在UAN代码的测试阶段的一行代码可能存在问题 for (each_predict, each_label,each_target_share_weight) in zip(logits, labels.cpu(), target_share_weight): 这个target_share_weight应该利用测试数据重新计算一下

datasets文件夹下
from datasets.CWRU_label_inconsistent import CWRU_inconsistent
from datasets.CWRUFFT_label_inconsistent import CWRUFFT_inconsistent
总是会报如下错误：TypeError: 'module' object is not callable（“模块”对象不可调用）
但是我认真比对了代码 并没有出现包的引入错误 这是为什么呢？

• train_utils_open_univ.py line 273
  I checked the official UAN code and they get the domain_prob by adding or not adding .detach(), while your code seems to use only self.AdversarialNet_auxiliary(), which is non-adversarial because no grl layer is added. So the adv_loss here might be wrong? self.AdversarialNet() should be used here right?

• train_utils_open_univ.py line 330
  > should be changed to <? Because the weight < threshold means that it belongs to the unknown class.

• train_utils_open_univ.py line 368-415
  Code indentation may be wrong？

Thanks for your effort! This code helped me a lot : )