lyutyuh / gazetteer-ner-acl19 Goto Github PK

您好，请问一下我用不同的语料做训练，但是tokens.txt和预训练的embedding都一样，为什么vocab.get_vocab_size('tokens')返回的size会不一样呢
谢谢

https://github.com/lyutyuh/acl19_subtagger/blob/227d6b0389a6d01de7c43c229e293ffe09915b3d/modules/hscrf_layer_SoftDict.py#L229
这一行是不是应该注释掉啊，上面已经得到了BILOU_features，为什么又重新置0了呢

hello，我尝试运行 allennlp train configs/softDictionary_HSCRF.config -s dump_directory/ --include-package models 来预训练subtagger，但是报了错误，如下：

Looks like the server that hosts pre trained models is down. I'm getting Bad Gateway error at https://www.jeffliu.page

I am keep getting error - "CUDA out of memory", even though I have reduced batch size (to 2) and model size. Tried to use multiple GPUs, doesn't work. Any ideas?

Hi I am trying to train this on CPU and I am getting the below error.

result = self.forward(*input, **kwargs)
File "./models/HSCRF_SoftDict.py", line 228, in forward
] += self.end_token_embedding.cuda(util.get_device_of(spans))
RuntimeError: Device index must not be negative

I am not very much aware of torch. Let me know how to resolve it.

请问您有没有遇到过子分类器单独预测时候用model.tar.gz加载模型得到的参数和在代码中用.th文件加载的参数不一致的情况？

Hi, I am getting the following error while running the training script.
FileNotFoundError: [Errno 2] No such file or directory: 'https://www.jeffliu.page/files/state.th'

It seems that jeffliu.page website is down. Any idea how to resolve this issue

Hi,
could you link back from the main README to the https://github.com/microsoft/vert-papers repo from KC?

请问如何预训练一个subtagger呢，因为想把这个框架用在别的gazetteer上

is there any difference between this code and https://github.com/microsoft/vert-papers/tree/master/papers/SubTagger , which should I use for the ACL 19 paper? Thanks .

hi，我想请问一下，论文里另外两个模型HSCRF+concat和HSCRF+gazemb代码啥时候能上传啊？
HSCRF+concat主要用的哪些特征啊？

Hi，我想请教一个问题，论文Table 1中HSCRF是指用Hybrid semi-Markov CRF模型在CONLL数据集上的结果吗？如果是的话，为什么会相差1个百分点？谢谢

Unable to access the data on the page : "https://www.jeffliu.page/files/softdict.zip" . Probably the page is down.

Throws this error :

2020-10-16 01:15:42,574 - WARNING - urllib3.connectionpool - Retrying (Retry(total=3, connect=None, read=None, redirect=None, status=None)) after connection broken by 'NewConnectionError('<urllib3.connection.VerifiedHTTPSConnection object at 0x7f60d4afb240>: Failed to establish a new connection: [Errno 111] Connection refused',)': /files/DATA/conll2003/train.txt

能问一下这个算法Viterbi的动态规划是怎么写的吗，多加了一层循环不太会了，代码看的有点儿不懂
谢谢

按照soft_dictionary_span_classifier_HSCRF.py里的写法，是对gazetteers的条目进行了序列标注而不是像论文里写的是对条目进行分类呀