To demonstrate the flexibility of recurrent networks, we are going to do something different in the final section of this chapter. Up until now, we have been working with probably the most-used testing data set, MNIST. This dataset has characteristics that are well known and it is extremely useful for comparing different types of models and testing different architectures and parameter sets. However, there are some tasks, such as natural language processing, that quite obviously require an entirely different type of dataset.
Also, the models we have built so far have been focused on one simple task: classification. This is the most straightforward machine learning task. To give you a flavor ...
