CHAPTER 4Deep Learning Using Keras

In Chapter 2, you learned about Machine Learning algorithms and techniques. You saw code samples of how to build ML models and evaluate your models using metrics of precision and recall. These models were pretty straightforward to understand with some clever ways of capturing patterns in the data. This chapter gets to the much more complex types of learning models. These models have many learning units organized in layers and many such layers—making the architecture “deep.” Though they are complex to build and train, you will see how effective they are at handling big and complex unstructured data like images. Finally, you will use one of the most popular Deep Learning libraries today—called Keras—to build models that can classify images of handwritten digits and learn to label these digits. I am hoping that these simple examples trigger some big ideas in your mind. You can reuse this code to apply learning to your images to build deep models in your domain area.

Handling Unstructured Data

We saw data used in earlier problems like wine quality analysis. Here each column had a particular significance and meaning. We used the term feature to describe each column and this was an important part of our learning method to understand how these features are correlated. We used techniques like normalization to scale the features so they were on the same value scale. Also, we saw that we could use fewer features to make our models learn faster. In short, ...

Get Keras to Kubernetes now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.