Training a logistic regression model

Logistic regression is really close to linear regression conceptually. Once linear regression is fully understood, logistic regression is just a couple of tricks away. But unlike linear regression, logistic regression is most commonly used for classification tasks.

Let’s first explain what logistic regression is, and then train a model on the breast cancer dataset using scikit-learn.

Getting ready

Unlike linear regression, logistic regression’s output is limited to a range of 0 to 1. The first idea is exactly the same as linear regression, having for each feature a parameter :

There is one more step to limit the range to 0 to 1, which is to apply the logistic function to this output z. As a reminder, the logistic function (also called the sigmoid function, although it’s a more generic function) is the following:

The logistic function has an S-shape, with values...