/gluon-nlp

NLP made easy

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

GluonNLP: Your Choice of Deep Learning for NLP

GluonNLP is a toolkit that enables easy text preprocessing, datasets loading and neural models building to help you speed up your Natural Language Processing (NLP) research.

Installation

Make sure you have Python 2.7 or Python 3.6 and recent version of MXNet. You can install MXNet and GluonNLP using pip:

pip install --pre --upgrade mxnet
pip install gluonnlp

Docs 📖

GluonNLP documentation is available at our website.

Community

For questions and comments, please visit our forum (and Chinese version). For bug reports, please submit Github issues.

How to Contribute

GluonNLP has been developed by community members. Everyone is more than welcome to contribute. We together can make the GluonNLP better and more user-friendly to more users.

Read our contributing guide to get to know about our development procedure, how to propose bug fixes and improvements, as well as how to build and test your changes to GluonNLP.

Join our contributors.

Resources

Check out how to use GluonNLP for your own research or projects.

If you are new to Gluon, please check out our 60-minute crash course.

For getting started quickly, refer to notebook runnable examples at Examples.

For advanced examples, check out our Scripts.

For experienced users, check out our API Notes.

Quick Start Guide

Load the Wikitext-2 dataset, for example:

>>> import gluonnlp as nlp
>>> train = nlp.data.WikiText2(segment='train')
>>> train[0][0:5]
['=', 'Valkyria', 'Chronicles', 'III', '=']

Build vocabulary based on the above dataset, for example:

>>> vocab = nlp.Vocab(counter=nlp.data.Counter(train[0]))
>>> vocab
Vocab(size=33280, unk="<unk>", reserved="['<pad>', '<bos>', '<eos>']")

From the models package, apply an Standard RNN langauge model to the above dataset:

>>> model = nlp.model.language_model.StandardRNN('lstm', len(vocab),
...                                              200, 200, 2, 0.5, True)
>>> model
StandardRNN(
  (embedding): HybridSequential(
    (0): Embedding(33280 -> 200, float32)
    (1): Dropout(p = 0.5, axes=())
  )
  (encoder): LSTM(200 -> 200.0, TNC, num_layers=2, dropout=0.5)
  (decoder): HybridSequential(
    (0): Dense(200 -> 33280, linear)
  )
)

For example, load a GloVe word embedding, one of the state-of-the-art English word embeddings:

>>> glove = nlp.embedding.create('glove', source='glove.6B.50d')
# Obtain vectors for 'baby' in the GloVe word embedding
>>> type(glove['baby'])
<class 'mxnet.ndarray.ndarray.NDArray'>
>>> glove['baby'].shape
(50,)