Ziqiang Liu

Results 7 issues of Ziqiang Liu

在metanet的代码中这里传入了None ``` fewshot_re_kit.framework.FewShotREModel.__init__(self, None) ``` 但是在FewShotREModel中Dataparallel需要传入参数,否则报错 ``` class FewShotREModel(nn.Module): def __init__(self, my_sentence_encoder): ''' sentence_encoder: Sentence encoder You need to set self.cost as your own loss function. ''' nn.Module.__init__(self) self.sentence_encoder =...

您好,我尝试了很多次都无法复现出论文的结果,请问可不可以提供一下能够复现出论文结果的参数?感激不尽!

请问为什么代码中直接拿测试数据集来做验证?验证集似乎没有使用到。请问可以解释一下一下处理分割好的数据集分别都是用在哪个阶段嘛?谢谢!

请问如何得到论文中Proto、HATT-Proto、BERT-PAIR等这些Few-shot模型的结果,是在训练阶段保留了base relation的原型吗? 另外,我在复现BERT 5-shot时发现,base acc越来越高,但是novel acc却越来越低了,虽然both acc达到了论文中的结果,但是novel acc 却远低于论文结果,请问如何改善呢?参数基本都是代码中的默认参数。

Now the uploaded code is just the experiment about "nyt-10" dataset. Could please upload another part of the experiment "noisy-TACRED" in the paper? Thank you!

#### Reference Issues/PRs #### What does this implement/fix? Explain your changes. In the current evaluation logic, whenever either `pred_try` or `pred_minimal_try` fails, it will be judged as `no apply`, which...

请问可以share一下训练数据嘛?或者数据要处理成什么样的格式才能训练呢?