vyraun / cnn_basics

NLP using CNN on Cornell Movie Ratings

CNN basics

Should speed up getting CNN going for NLP.

Built specifically for Cornell movie reviews data, but can be used with any text classification.

Got to around 75% accuracy with a very simple grid search.

It should be easy to run it with your own data though, only a few lines.

Installation

Works at least on Python 3.5

Install using pip (just, spacy, keras; using tensorflow, gensim):

pip3.5 install -r requirements.txt

Post install:

For spacy you will need to install the english corpus: python3.5 -m spacy.en.download

Get word embeddings and place them in ~/embeddings/:

• GoogleNews 300-dim word embeddings