Learning TensorFlow.js

Chapter 9. Classification Models and Data Analysis

“Forethought spares afterthought.”

—Amelia Barr

There’s a reason you don’t just dump data into a model. Neural networks operate at intense speeds and perform complex calculations the same way humans can have an instantaneous reaction. However, for both humans and machine learning models, a reaction rarely contains a reasoned context. Dealing with dirty and confusing data creates subpar models, if anything at all. In this chapter, you’ll explore the process of identifying, loading, cleaning, and refining data to improve the training accuracy of a model in TensorFlow.js.

We will:

• Identify how to make a classification model

• Learn how to handle CSV data

• Learn about Danfo.js and DataFrames

• Identify how to get messy data into training (wrangle your data)

• Practice graphing and analyzing data

• Learn about machine learning notebooks

• Expose core concepts of feature engineering

When you finish this chapter, you’ll feel confident in gathering large amounts of data, analyzing it, and testing your intuitions by using context to create features that help models train.

In this chapter, you’ll build a Titanic life-or-death classifier. Will Miss Kate Connolly, a 30-year-old woman with a third-class ticket, survive? Let’s train a model to take that information and give us a likelihood of survival.

Classification Models

So far, you’ve trained a model that outputs numbers. Most of the models you’ve consumed behave a bit differently ...