Summary
This chapter acted as a review of the basic concepts introduced in the previous chapters, while introducing a new application, sentiment analysis, and a high-level library, Keras, to simplify the development of models with the Theano engine.
Among these basic concepts were recurrent networks, word embeddings, batch sequence padding, and class one-hot encoding. Bidirectional recurrency was presented to improve the results.
In the next chapter, we'll see how to apply recurrency to images, with another library, Lasagne, which is more lightweight than Keras, and will let you mix the library modules with your own code for Theano more smoothly.