hackerxiaobai / opentag_2019 Goto Github PK

Hello and thank you for sharing your code.
I am trying to understand better the OpenTag implementation and am getting the below error;
Traceback (most recent call last):
File "C:\Project\OpenTag_2019\env\lib\site-packages\fire\core.py", line 471, in _Fire
target=component.name)
File "C:\Project\OpenTag_2019\env\lib\site-packages\fire\core.py", line 681, in _CallAndUpdateTrace
component = fn(*varargs, **kwargs)
File "C:/Project/OpenTag_2019/main.py", line 76, in train
loss = model.log_likelihood(inputs)
File "C:\Project\OpenTag_2019\models\OpenTag_2019.py", line 102, in log_likelihood
return - self.crf(outputs, target)
File "C:\Project\OpenTag_2019\env\lib\site-packages\torch\nn\modules\module.py", line 1110, in call_impl
return forward_call(*input, **kwargs)
File "C:\Project\OpenTag_2019\env\lib\site-packages\torchcrf_init.py", line 102, in forward
numerator = self.compute_score(emissions, tags, mask)
File "C:\Project\OpenTag_2019\env\lib\site-packages\torchcrf_init.py", line 186, in _compute_score
score = self.start_transitions[tags[0]]
IndexError: tensors used as indices must be long, byte, or bool tensors

I am not sure if this is a result of a version issue.
Can you advise the versions for the requirements you make reference to?

The above is produced when running the code using main.py train --batch_size=128

Thank you in advance.

it seems that we still need data file process.pkl to run the code.
could you provide the file or tell me how to generate it from ./data/raw.txt maybe?

Thanks

如题，谢谢。

hello，
看论文中，你们重新复现了 open tag for Keras，这个代码可以发出来学习一下吗？因为想用自己的数据，对这两种方式都进行实验学习一下，万分感谢啦

Hi xiaobai,

Thank you for sharing your great repository.
May I know whether I can use the raw.txt data for my own project?
Thank you very much.

Is the dataset used in this project the same as the dataset used in the paper(《Scaling Up Open Tagging from Tens to Thousands: ComprehensionEmpowered Attribute Value Extraction from Product Title》)? If not, what is the source of this dataset?

您好，我看了论文仅仅采用 bio的标注方式，这样模型的输出怎么让 属性值（value） 和 属性（attribute）一一正确对应呢？举个例子， title:“女童网面透气运动2019夏季新款韩版" attibute: "性别，季节，年份"，我们可以看出 {性别：女童，季节：”夏季“,年份：”2019“，}, 然而模型（假设可以做到）只是依次在 ”女童“，”2019“，”夏季“ 中打了B I label,他并不可以自动让attibute 和 value 正确的对齐呀？ 如果是这样的话，让attributes特别多的话，那完全不能对齐啊。以上是我浅薄的理解（应该是理解错了），希望您在任何有空的时间给我回复，谢谢！

I have found three packages under that name., and with all three of them I have run into problems. Would it be possible for to tell me which exact package I need to run the code, including the version?

Hello， 非常感谢作者的分享。我阅读您的代码，发现OpenTag模型里面并没有用到attention layer吧？我只看到了calculate_cosin这个函数，计算context和att的的距离。故而想向您确认下，模型是否用到了attention layer呢？