In this chapter, we will be continuing our journey of tackling the billion-dollar worth problem of advertising click-through prediction. We will be focusing on learning a very (probably the most) scalable classification model—logistic regression. We will be exploring what logistic function is, how to train a logistic regression model, adding regularization to the model, and variants of logistic regression that are applicable to very large datasets. Besides the application in classification, we will also be discussing how logistic regression and random forest are used in picking significant features. Again, you won't get bored as there will be lots of implementations from scratch, and with scikit-learn and TensorFlow.

In this chapter, we will cover the following topics:

• Categorical feature encoding
• Logistic function
• What is logistic regression
• Training a logistic regression model via gradient descent
• Training a logistic regression model via stochastic gradient descent
• The implementations of logistic regression from scratch
• The implementations of logistic regression with scikit-learn
• The implementations of logistic regression with TensorFlow
• Click-through prediction with logistic regression
• Logistic regression with L1 and L2 regularization
• Logistic regression for feature selection
• Online learning
• Another way to select features—random forest