neukg/grte

grte's Issues

crossattention的创建

crossattention的创建直接Bertconfig文件那么他和创建的self-attention应该没有区别把

The return value is not match

The code of data_generator.iter() in main.py is
token_ids, _ ,mask = self.tokenizer.encode(d['text'], max_length=self.max_len)
while the tokenizer in bert4keras only return 2 value, first_token_ids and first_segment_ids, how do you make it?

The problem of extracting entity pairs

Hello, thanks a lot for you releasing the source code!
As the paper says, there are three routes to extract entity pair. However, I did only see two routes in the source code. How do you operate the second search route ( from tail tokens to head tokens )? I think you just put all the entity pairs into a big list and it's difficult to sovle the overlapping problem especially in "MMH→MMT”.
Thank you for your kind help. I'm looking forward to your reply.

请问一下，训练中的rounds是怎么选择的？

图片中的训练参数rounds是怎么选择的？

How to deal with subjects or objects that appear multiple times in the text？

Thanks for your wonderful work!
I have a question. Since there is only text of subject and object in the dataset and there is no location information, is the first found location used as the subject or object when reading data?
We look forward to your reply!

encode() got an unexpected keyword argument 'max_length'

请问一下，token_ids,mask = self.tokenizer.encode(
d['text'], max_length=self.max_len）这段代码报错，怎么修改

File "/ddnstor/imu_tzt1/sulisha/GRTE-main/main.py", line 48, in iter
d['text'], max_length=self.max_len
TypeError: encode() got an unexpected keyword argument 'max_length'

bert4keras

Hi, may I ask what the version of bert4keras is

hi，which version of bert4kreas is it?

实验结果

您好，能问一下关于不同重叠模式和不同三元组数量的F1得分的计算源码吗