Comments (5)
您好,我来回答一下您的问题
- 是单轮的训练结果。全部实验都是random seed = 0,因为encoding scheme比较重要,所以我在webnlg上也是做过测试的,sequential和parallel大概比joint低个0.5-1.5点左右。SciERC结果不稳定,想能问下浮动能大到多少呢?浮动大的原因可能是超参没有怎么调,可以试试一些稳定训练的trick,比如说warmup等。
- entity extraction确实会出现您说的这种情况,这部分可能是RE部分没有cover实体尾部的结果,可以试试PFN-nested。
- 这个没有做过。因为没有partition,也就得不到三个分区了,这样的话也就不知道是对哪个部分进行filter了。
from pfn.
- 我使用了您的代码,重新运行了在seed = [0,1,2,3,4]的情况。NER的结果在65.67-68.59,RE的结果在36.23到37.86。方差还是比较大的。
- 如果方便的话,请问您是否可以在ablation study添加上entity extraction的结果呢?看到您的代码是同时输出NER和RE两个结果的。
非常感谢您的回答。希望您科研顺利。
from pfn.
RE结果和我这差的有点多。我跑了下1,2,3,4的seed,NER的结果是66.9, 66.4, 68.0, 67.7, RE的结果是36.9, 36.3, 38.9, 37.7。我在readme上更新了cuda版本要求,貌似这个也是会有影响的。
ablation我这里只存了sequential和parallel,sequential是68.7, parallel是67.0,scierc上确实波动很大,这部分我觉得可能是因为训练数据集本身的原因导致的。webnlg上NER sequential和parallel比joint差个0.3-0.5左右。
由于论文是最终稿,emnlp不让改了,所以可能没法添加结果。希望上边几个数值对您有用。
from pfn.
好的,谢谢您的回复。论文中的内容虽然无法额外变动,但是希望能够在github项目的readme上更新实验结果。同时建议您的ablation study能够在5-10个random seed上取NER和RE的平均结果和方差。单次的RE结果恐怕不能够使得您论文中ablation study的结论令人信服。
from pfn.
嗯,谢谢您的建议,ablation的多次结果之后会考虑加上的。
from pfn.
Related Issues (20)
- Small inconsistency, or not? HOT 1
- question about eval_metric HOT 1
- When training model, is it necessary to set args.do_eval=True?
- FileNotFoundError: [Errno 2] No such file or directory: 'data/data/NYT/ner2idx.json' HOT 1
- OOM for my own bigger datasets HOT 2
- Question about entity extraction of out-of-triples HOT 1
- chinese custom dataset HOT 2
- evaluate the model with customized input HOT 2
- 请问消融实验的编码模式encoding scheme应该怎么设置呀 HOT 2
- Nest下的处理 HOT 2
- SEMEVAL dataset HOT 1
- 为什么PFN-nested model在关系训练时候可以利用实体tail信息呢? HOT 2
- 你好,能否提供适用于nested-pfn的inference.py呢?感谢! HOT 1
- 关于头尾实体特征和关系特征的问题 HOT 5
- 关于实验结果复现的问题 HOT 2
- 关于实体抽取和关系抽取F1分数问题 HOT 2
- 关于OOT(Out-of-triple)数据集问题 HOT 2
- why use albert-xxlarge instead of bert-base when training on some datasets? HOT 2
- Question about the tail re unit HOT 6
- 有一个报错问题,不知道是环境不对还是, HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pfn.