Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

关于测试的一个问题 #27

Open
JiyangZhang opened this issue Feb 11, 2019 · 6 comments
Open

关于测试的一个问题 #27

JiyangZhang opened this issue Feb 11, 2019 · 6 comments
Labels
bug Something isn't working

Comments

@JiyangZhang
Copy link

文中提到这个模型是sentence-level的relation extracion模型,把每一个句子看成一个包来训练。您的代码中好像是根据包(entity pair)来测试的。请问您这种测试方法是否可行。谢谢.

@xuyanfu
Copy link
Owner

xuyanfu commented Feb 17, 2019

您好,虽然这篇文章的方法可以做到sentence-level. 但是关于远程监督的关系抽取数据集,之前的评测方式都是bag-level. 第一,主要是方便跟之前的工作进行比较。第二基于远程监督的假设,不能保证句子级别的标注都是正确的。

@JiyangZhang
Copy link
Author

感谢回答。还有一个小问题,据我了解,之前bag-level测评作pr曲线的时候,作者作图的方法是将每个包对每种关系的概率进行排序,然后作pr曲线图。个人觉得这种方法更普遍,参考的是thunlp组的开源代码。

@xuyanfu
Copy link
Owner

xuyanfu commented Feb 17, 2019

我应该也进行排序啦,在test.py的126行。

@JiyangZhang
Copy link
Author

重点是参与排序的概率的个数,您是每个entity_pair的最大的一个概率进行排序。Yankai Lin 如下回复我的邮件:
“ For drawing the aggregated precision-recall curve, if you have n entity pairs (not sentences, an entity pair may have several sentences), and m relations (including NA), the trained model will predict and produce the probability matrix of each sentence ([m-1, n]), which not includes NA, make a list of length of (m-1)*n, ranking it according to probability, and plot the precision-recall curve **accordingly.”
这是”Neural Relation Extraction with Selective Attention over Instances“的画法。

不过好像也没有一个绝对标准,只是thunlp是这么做的。再次感谢您的代码,给我提供了很大帮助!谢谢

@xuyanfu xuyanfu mentioned this issue Sep 9, 2019
@xuyanfu xuyanfu added the bug Something isn't working label Sep 9, 2019
@zxs1995
Copy link

zxs1995 commented Oct 1, 2019

您好,请问如果改进这篇论文,用bag-level的评测标准可以不

@ShellingFord221
Copy link

你好,也就是说,这篇文章的工作是sentence级别的训练(把每个句子当成一个包),然后bag级别的评测,目的是为了将结果与之前的方法比较?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

4 participants