/keras-text

Text Classification Library in Keras

Primary LanguagePythonMIT LicenseMIT

Keras Text Classification Library

Build Status license Slack

keras-text is a one-stop text classification library implementing various state of the art models with a clean and extendable interface to implement custom architectures.

Quick start

Create a tokenizer to build your vocabulary

  • To represent you dataset as (docs, words) use WordTokenizer
  • To represent you dataset as (docs, sentences, words) use SentenceWordTokenizer
  • To create arbitrary hierarchies, extend Tokenizer and implement the token_generator method.
from keras_text.processing import WordTokenizer


tokenizer = WordTokenizer()
tokenizer.build_vocab(texts)

Want to tokenize with character tokens to leverage character models? Use CharTokenizer.

Build a dataset

A dataset encapsulates tokenizer, X, y and the test set. This allows you to focus your efforts on trying various architectures/hyperparameters without having to worry about inconsistent evaluation. A dataset can be saved and loaded from the disk.

from keras_text.data import Dataset


ds = Dataset(X, y, tokenizer=tokenizer)
ds.update_test_indices(test_size=0.1)
ds.save('dataset')

The update_test_indices method automatically stratifies multi-class or multi-label data correctly.

Build text classification models

See tests/ folder for usage.

Word based models

When dataset represented as (docs, words) word based models can be created using TokenModelFactory.

from keras_text.models import TokenModelFactory
from keras_text.models import YoonKimCNN, AttentionRNN, StackedRNN


# RNN models can use `max_tokens=None` to indicate variable length words per mini-batch.
factory = TokenModelFactory(1, tokenizer.token_index, max_tokens=100, embedding_type='glove.6B.100d')
word_encoder_model = YoonKimCNN()
model = factory.build_model(token_encoder_model=word_encoder_model)
model.compile(optimizer='adam', loss='categorical_crossentropy')
model.summary()

Currently supported models include:

  • Yoon Kim CNN
  • Stacked RNNs
  • Attention (with/without context) based RNN encoders.

TokenModelFactory.build_model uses the provided word encoder which is then classified via Dense block.

Sentence based models

When dataset represented as (docs, sentences, words) sentence based models can be created using SentenceModelFactory.

from keras_text.models import SentenceModelFactory
from keras_text.models import YoonKimCNN, AttentionRNN, StackedRNN, AveragingEncoder


# Pad max sentences per doc to 500 and max words per sentence to 200.
# Can also use `max_sents=None` to allow variable sized max_sents per mini-batch.
factory = SentenceModelFactory(10, tokenizer.token_index, max_sents=500, max_tokens=200, embedding_type='glove.6B.100d')
word_encoder_model = AttentionRNN()
sentence_encoder_model = AttentionRNN()

# Allows you to compose arbitrary word encoders followed by sentence encoder.
model = factory.build_model(word_encoder_model, sentence_encoder_model)
model.compile(optimizer='adam', loss='categorical_crossentropy')
model.summary()

Currently supported models include:

  • Yoon Kim CNN
  • Stacked RNNs
  • Attention (with/without context) based RNN encoders.

SentenceModelFactory.build_model created a tiered model where words within a sentence is first encoded using
word_encoder_model. All such encodings per sentence is then encoded using sentence_encoder_model.

  • Hierarchical attention networks (HANs) can be build by composing two attention based RNN models. This is useful when a document is very large.
  • For smaller document a reasonable way to encode sentences is to average words within it. This can be done by using token_encoder_model=AveragingEncoder()
  • Mix and match encoders as you see fit for your problem.

Resources

TODO: Update documentation and add notebook examples.

Stay tuned for better documentation and examples. Until then, the best resource is to refer to the API docs

Installation

  1. Install keras with theano or tensorflow backend. Note that this library requires Keras > 2.0

  2. Install keras-text

From sources

sudo python setup.py install

PyPI package

sudo pip install keras-text
  1. Download target spacy model

keras-text uses the excellent spacy library for tokenization. See instructions on how to download model for target language.

Citation

Please cite keras-text in your publications if it helped your research. Here is an example BibTeX entry:

@misc{raghakotkerastext
  title={keras-text},
  author={Kotikalapudi, Raghavendra and contributors},
  year={2017},
  publisher={GitHub},
  howpublished={\url{https://github.com/raghakot/keras-text}},
}