In this demo, we will get hands-on experiences on Sequential Models

We will be building and training a basic character-level RNN to classify words. The model receives a series of characters as input and returns a prediction of the next character together with the hidden state at each step. The hidden step will be fed onto the next step. At the end the model returns the category (language) a given word belongs to.

The dataset consists of surnames from 18 different languages and returns the language of a given surname. While pytorch has a set of NLP pre-processing functionalities lying on the library torchtext, this notebook builds from scratch all the desired functionallity.