marcelomata / embeddings

Fast, DB Backed pretrained word embeddings for natural language processing.

Embeddings

Embeddings is a python package that provides pretrained word embeddings for natural language processing and machine learning.

Instead of loading a large file to query for embeddings, embeddings is backed by a database and fast to load and query:

>>> %timeit GloveEmbedding('common_crawl_840', d_emb=300)
100 loops, best of 3: 12.7 ms per loop

>>> %timeit GloveEmbedding('common_crawl_840', d_emb=300).emb('canada')
100 loops, best of 3: 12.9 ms per loop

>>> g = GloveEmbedding('common_crawl_840', d_emb=300)

>>> %timeit -n1 g.emb('canada')
1 loop, best of 3: 38.2 µs per loop

Installation

pip install embeddings  # from pypi
pip install git+https://github.com/vzhong/embeddings.git  # from github

Usage

Upon first use, the embeddings are first downloaded to disk in the form of a SQLite database. This may take a long time for large embeddings such as GloVe. Further usage of the embeddings are directly queried against the database. Embedding databases are stored in the $EMBEDDINGS_ROOT directory (defaults to ~/.embeddings). Note that this location is probably undesirable if your home directory is on NFS, as it would slow down database queries significantly.

from embeddings import GloveEmbedding, FastTextEmbedding, KazumaCharEmbedding, ConcatEmbedding

g = GloveEmbedding('common_crawl_840', d_emb=300, show_progress=True)
f = FastTextEmbedding()
k = KazumaCharEmbedding()
c = ConcatEmbedding([g, f, k])
for w in ['canada', 'vancouver', 'toronto']:
    print('embedding {}'.format(w))
    print(g.emb(w))
    print(f.emb(w))
    print(k.emb(w))
    print(c.emb(w))

Docker

If you use Docker, an image prepopulated with the Common Crawl 840 GloVe embeddings and Kazuma Hashimoto's character ngram embeddings is available at vzhong/embeddings. To mount volumes from this container, set $EMBEDDINGS_ROOT in your container to /opt/embeddings.

For example:

docker run --volumes-from vzhong/embeddings -e EMBEDDINGS_ROOT='/opt/embeddings' myimage python train.py

Contribution

Pull requests welcome!