An implementation of various models for sentiment classification (positive vs negative) of IMDB reviews based on the dataset used by the Kaggle Bags of Popcorn Knowledge Competition.

Implemented Approaches include the following, from best to worst, with AUC score shown based on a random 20% Test Dataset. See the code for parameter details as well as info on other variations of these models:

1. Bidirectional Recurrent Neural Net (RNN) using Passage Library: 0.966
2. Convolutional Neural Net (CNN) + RNN with Dropout using Keras: 0.954 Theano Backend, 0.952 TensorFlow Backend
3. Average Word2Vec Features Model based on IMDB Dataset: 0.871
4. Average Word2Vec Features Models based on Google News Dataset (Pre-trained): 0.854
5. RNN using TensorFlow's not fully matured at the time SKFlow: 0.829