In the previous chapter, we walked through building a simple drawing application that would try to infer what the user was drawing and present them with alternatives based on the most likely predicted categories; the intention of this application was to improve the efficiency of sketching tasks by giving the user completed sketches, obtained through Microsoft's Bing image search, rather than having to spend time fussing over the details.

In this chapter, we'll revisit this application but look at an alternative for inferring what the user is drawing, and, in doing so, we will be exposing ourselves to new types of data and machine learning models. Following the familiar format, we will first revise the task, explore the data and model, and then walk through building up the required functionality in a playground, before migrating it across to our application. Let's get started.