Sentimental Classification for Multiple Sentences

[KaiyeYang-git]

Sentimental Classification for Multiple Sentences

[asahi417]

To perform inference in batch on your device, you can pass a list of sentences instead of single string.
For example,

model = tweetnlp.load('sentiment')
model.sentiment(["sentence 1", "sentence 2", etc])

You can also find how the output would look like on the colab here.

[KaiyeYang-git]

To perform inference in batch on your device, you can pass a list of sentences instead of single string. For example,

model = tweetnlp.load('sentiment')
model.sentiment(["sentence 1", "sentence 2", etc])

You can also find how the output would look like on the colab here.

Hi asahi417, it is a great idea to calculate sentences togther but I received a warning for importing tweetnlp, and I hardly can find any information about it when someone met the same issue:
'''cannot import name 'allowed_transitions' from 'allennlp.modules.conditional_random_field'''
(C:\Users\ky002\AppData\Roaming\Python\Python39\site-packages\allennlp\modules\conditional_random_field_init_.py)
Do you have any ideas about it? Or which version of tweetnlp should I choose?

[KaiyeYang-git]

I have found a bug in 'env/lib/python3.8/site-packages/tweetnlp/model_ner/model.py'
the warning results show that 'cannot import name 'allowed_transitions' from 'allennlp.modules.conditional_random_field',
and it works fine it you change the code in 'env/lib/python3.8/site-packages/tweetnlp/model_ner/model.py' lin 12 from 'from allennlp.modules.conditional_random_field import allowed_transitions' to 'from allennlp.modules.conditional_random_field.conditional_random_field import allowed_transitions'.
The reason behind is that there are two 'conditional_random_field' files in the package which have a father-son relationship

[asahi417]

@KaiyeYang-git Could you share your version of allennlp library? That library behaves funky sometime... and it might be probably better to just migrate the CRF part into tweetnlp directly and remove the dependency on allennlp.

[KaiyeYang-git]

@KaiyeYang-git Could you share your version of allennlp library? That library behaves funky sometime... and it might be probably better to just migrate the CRF part into tweetnlp directly and remove the dependency on allennlp.

Here comes all information about library allennlp:

pip show allennlp
Name: allennlp
Version: 2.10.0
Summary: An open-source NLP research library, built on PyTorch.
Home-page: https://github.com/allenai/allennlp
Author: Allen Institute for Artificial Intelligence
Author-email: [email protected]
License: Apache
Location: c:\users\ky002\appdata\roaming\python\python39\site-packages
Requires: base58, cached-path, dill, fairscale, filelock, h5py, huggingface-hub, lmdb, more-itertools, nltk, numpy, protobuf, pytest, requests, rich, sacremoses, scikit-learn, scipy, sentencepiece, spacy, tensorboardX, termcolor, torch, torchvision, tqdm, traitlets, transformers, typer, wandb
Required-by: tweetnlp

[asahi417]

Thanks! Please allow me to take a bit more time to look into the issue. Sorry for the inconvenience 🙏

[asahi417]

It seems that from allennlp version 2.10.0, they use different way of loading a module we used in NER script. I fixed by changing the way of loading it depending on the version.