Chapter 5. Introducing Models

“Where does he get those wonderful toys?”

—Jack Nicholson (Batman)

Now you’re in the big leagues. Way back in Chapter 2 you accessed a fully trained model, but you didn’t need to understand tensors at all. Here in Chapter 5, you will get to utilize your tensor skills to work directly with your models, with no training wheels.

Finally, you’re going to dive into utilizing the brain of most machine learning. Models can seem like a black box. Generally, they expect a specific tensor shape in, and a specific tensor shape comes out. For instance, let’s say you’ve trained a dog or cat classifier. The input might be a 32 x 32 3D RGB tensor, and the output might be a single tensor value of zero to one to indicate the prediction. Even if you don’t know the inner workings of such a device, at the least, consuming and utilizing models with a defined structure should be simple.

We will:

  • Utilize trained models to predict a variety of answers

  • Identify the benefits of our existing tensor manipulation skills

  • Learn about Google’s hosting

  • Learn about object localization

  • Learn how to overlay a bounding box to identify some aspect of an image

This chapter will teach you direct access to models. You won’t be dependent on cute wrapper libraries for coddling. If you want, you’ll even be able to write your own wrapper library around existing TensorFlow.js models. Armed with the skills in this chapter, you can start applying breakthrough machine learning ...

Get Learning TensorFlow.js now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.