Sequence classifier with a scikit-learn interface

Convolutional neural network sequence classifier in the spirit of [1]. Wraps a Keras implementation as a scikit-learn classifier.

Software requirements

• Python (2.7 or >= 3.5)
• scikit-learn (tested on 0.19)
• Keras (tested on 2.0.5, with the Theano 0.9.0 backend)

Installation

Install Keras. This has been tested on the Theano backend, should work on other backends, too.

pip3 install --user sklearn-sequence-classifiers

Installing from the source code:

git clone git@github.com:aajanki/sequence-classification.git
cd sequence-classification
python3 setup.py install --user

Running tests

python3 setup.py test

Usage example

Predicting IMDB review sentiments.

from keras.datasets import imdb
from keras.preprocessing import sequence
from sequence_classifiers import CNNSequenceClassifier

maxlen = 400
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=5000)
x_train = sequence.pad_sequences(x_train, maxlen=maxlen)
x_test = sequence.pad_sequences(x_test, maxlen=maxlen)

clf = CNNSequenceClassifier(epochs=2)
clf.fit(x_train, y_train)
print(clf.score(x_test, y_test))

License

BSD

References

[1] Yoon Kim: Convolutional Neural Networks for Sentence Classification