Metadata-Version: 2.1
Name: swem
Version: 0.0.1
Summary: A portable document embedding using SWEM.
Home-page: https://github.com/yutayamazaki/SWEM-Python
Author: Yuta Yamazaki
Author-email: yu.yamazakii@gmail.com
Maintainer: Yuta Yamazaki
Maintainer-email: yu.yamazakii@gmail.com
License: MIT
Keywords: swem nlp python
Platform: UNKNOWN
Description-Content-Type: text/markdown
Requires-Dist: gensim
Requires-Dist: mecab-python3
Requires-Dist: numpy

# SWEM
Implementation of SWEM(Simple Word-Embedding-based Models)  
[Baseline Needs More Love: On Simple Word-Embedding-Based Models and Associated Pooling Mechanisms (ACL 2018)](https://arxiv.org/abs/1805.09843)

Details are available [here(Japanese)](https://scrapbox.io/whey-memo/SWEM(Simple_Word-Embedding-Based_Models)).

## Example

```python example.py
from gensim.models.word2vec import Word2Vec

from swem import SWEM

if __name__ == '__main__':
    model = Word2Vec.load('wiki_mecab-ipadic-neologd.model')

    swem = SWEM(model)

    doc = '僕の名前はバナナです。'


    for method in ['max', 'average', 'concat']:
        print(swem.infer_vector(doc, method=method).shape)
```

Results  
```shell
(200,)
(200,)
(400,)
```

