zhpmatrix / bertem

论文实现(ACL2019)：《Matching the Blanks: Distributional Similarity for Relation Learning》

Introduction

• SHOW ME YOUR CODE!
• email:[email protected]
• 微信公众号《KBQA沉思录》主理人
• 个人博客:zhpmatrix.github.io
• 🎓 Master of Computer Science
• 养了好几盆绿植🌳，还有一只巴西龟和台湾花龟，还有一只精力无限的猫🐱，想养一只柴犬
• 写码，看书，写文章，打篮球，吹牛
• 年少时想成为一个作家，很酷的工作

Visitor count

bertem's Issues

"Entity start state"相关代码

entity_emb_output = batch_entity_emb.sum(dim=1)

这个与paper中所采用的concat方式不一样呢，有什么特殊的考虑么？

why: num_embeddings = 2 * (seq_length - 1) + 1

I can't understand this code.

你好 请问如何获取TACRED数据集呢

求助 谢谢哥

在fewrel数据集上实验过吗

paper的源代码库里没有给出few shot的实验，请问你这儿有在fewrel数据集上实验过吗

SemEval dataset question

How is the SemEval data set handled as input? Is it convenient to provide the input sample data？

求助楼主

tacred数据集在官方下不到了 楼主可以私发一下吗 万分感谢啊
[email protected]

token error

I use SemEval Task8 data.
Arguments: (['[CLS]', 'the', 'system', 'as', 'described', 'above', 'has', 'its', 'greatest', 'application', 'in', 'an', 'array', '##ed', '<s1>', 'configuration', '<e1>', 'of', 'antenna', '<s2>', 'elements', '<e2>', '.', '[SEP]'],)
but there is no '<s1>' '<e1>' in vocab!

Question regarding reproducing reported scores on TACRED with BERT-EM model

Hi,
Have you been able to reproduce 70.1 F1 on TACRED dataset with BERT-EM model as reported in the paper?

Thanks and regards,

Could you please give a sample of TACRED dataset format for your code?

Thank you for your kind open sourcing. I have tried to run your code about tacred_run_classifier.py on TACRED dataset. It seems that your input dataset is different from the original dataset.
According to the original dataset format, I changed the input entity_pos as "entity_pos = [[line['subj_start'],line['subj_end']+1],[line['obj_start'],line['obj_end']+1]]". Is it right?

Even so, when running the code, some errors about GPU or Cuda happened. Is this the problem of input data format? Or are there any other parameters should I change?

The errors are like:“RuntimeError: cublas runtime error : resource allocation****aten/src/THC/THCGeneral.cpp:250" etc.

Where is the model/ folder

On your train script you specified a model folder that is not in the repository, from where should I download?