wangpeiyi9979 / ie-bert-cnn Goto Github PK

一个关于百度2019语言与智能技术竞赛信息抽取 (http://lic2019.ccf.org.cn/kg) 模型, 模型采用BERT+CNN。DEMO地址 https://github.com/Wangpeiyi9979/InformationExtractionDemo

Python 100.00%

ie-bert-cnn's People

Contributors

Stargazers

Watchers

Forkers

gavingx lxj0276 jxinyee feileyu manduner songjinduo weibobo2015 xumeng123 tangelian nidhoggurz wengbenjue zxlzr mylv1222 nicole1130 qianrenjian liaorongfu bisonx pujing-ai cqray1990 freddiexu gztangde leiyis99 chenjun0210 p3n9w31 rita1223-0727 akailcy joe2hpimn littlestar-angel hominine720202 balance-cqy sxjpage ryanalexton beira-bf jin-guangyin amjoker zhangyuan-bot jx1100370217 gmjustforfun yuconggen jachinlin2022

ie-bert-cnn's Issues

ValueError: Object arrays cannot be loaded when allow_pickle=False

遇到错误烦来请教，请问您知道原因吗，谢谢！

Traceback (most recent call last):
File "helpData.py", line 382, in
dataHelper.process_data()
File "helpData.py", line 377, in process_data
metrics.judge_data_quality(self.opt)
File "/home/github/KNOWLEDGE/IE_BERT_CNN/metrics.py", line 340, in judge_data_quality
train_npy_data = np.load(opt.npy_data_root+'train/relations.npy')
File "/home/anaconda3/lib/python3.7/site-packages/numpy/lib/npyio.py", line 453, in load
pickle_kwargs=pickle_kwargs)
File "/home/anaconda3/lib/python3.7/site-packages/numpy/lib/format.py", line 722, in read_array
raise ValueError("Object arrays cannot be loaded when "
ValueError: Object arrays cannot be loaded when allow_pickle=False

已解决

AttributeError: 'NoneType' object has no attribute 'bert'

您好，作者！
执行python main.py train报错：
File "F:\Skill\Code\RELATION\IE-Bert-CNN-master\models\BERT_MUL_CNN.py", line 112, in forward
sequence_output, _ = self.bertForToken.bert(batch_data, token_type_ids=None, attention_mask=batch_masks, output_all_encoded_layers=False)
AttributeError: 'NoneType' object has no attribute 'bert'
请问有没有解决办法？

关于这个信息抽取比赛的数据

你好：
请问方便提供一下“2019语言与智能技术竞赛信息抽取”比赛的数据么？
非常非常感谢！

你好，很感兴趣您的工作，main.py文件中缺少了data文件，您能提供吗？

能请问下使用cnn的原因吗？

会不会出现比较多的情况，比如，实体不全（北京大学，但抽取出北京大）或者过长（北京，但北京大）呢？

Could you please explain the file “all_50_schemas_new” is for what？Thanks a lot...

As the title, please！

关于任务

这里面BERT是用来训练词向量吗 CNN做特征提取？

pytorch_pretrained_bert 缺失

感觉您的工作非常有意义，想学习一下，看您代码中至少有两个文件都
import pytorch_pretrained_bert
然后报错找不到这个文件或模块，请问能提供下这个文件吗？
谢谢

ValueError: setting an array element with a sequence. 请问怎么解决

Traceback (most recent call last):
File "helpData.py", line 383, in
dataHelper.process_data()
File "helpData.py", line 370, in process_data
self.get_sens_and_tags_and_entsRel(self.origin_train_data, case=0)
File "helpData.py", line 348, in get_sens_and_tags_and_entsRel
ent_rel = np.array(ent_rel)

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (173045,) + inhomogeneous part.

准确率

准确率是用pre除以rel吗

请教

执行python main.py train报错：
File "F:\Skill\Code\RELATION\IE-Bert-CNN-master\models\BERT_MUL_CNN.py", line 112, in forward
sequence_output, _ = self.bertForToken.bert(batch_data, token_type_ids=None, attention_mask=batch_masks, output_all_encoded_layers=False)
AttributeError: 'NoneType' object has no attribute 'bert'
请问有没有解决办法？

代码中一些细节的疑问

s1, e1, s2, e2 = positions
type_obj = torch.zeros(e1-s1+1).long()
type_sbj = torch.ones(e2-s2+1).long()
if e1 < s2:
m_s = e1+1
m_e = s2-1
else:
m_s = e2+1
m_e = s1-1
type_mid = torch.ones(m_e-m_s+1).long() * 2
if self.opt.use_gpu:
type_obj = type_obj.cuda()
type_sbj = type_sbj.cuda()
type_mid = type_mid.cuda()
obj_vecs = sen_matrix[s1:e1+1,:] + self.type_emb(type_obj)
sbj_vecs = sen_matrix[s2:e2+1,:] + self.type_emb(type_sbj)
mid_vecs = sen_matrix[m_s:m_e+1,:] + self.type_emb(type_mid)
请问下这个mid_vecs 是啥？