Generalized Linear Models (GLMs)

In the first part of the book, we discussed supervised learning models where the response variable is real-valued. There are many applications where the response variable is of a different type. In this part, we introduce modeling strategies for these different response types, as well as how to use glmnet to obtain sparse versions of these models.

• In 7  Binary classification we describe the logistic regression model for binary responses.

• In 8  Multiclass classification we discuss how the logistic regression model can be extended to handle classification tasks with more than two classes.

• In 9  Poisson regression we introduce the Poisson regression model for modeling count data.

• The three models above are special cases of an important family of models called generalized linear models (GLMs). In 10  Other GLMs we describe what GLMs are, and how you can use glmnet to fit regularized versions of any GLM.

• In 11  Multi-task learning, we introduce the multi-task learning (MTL) paradigm and describe a sparse modeling approach for it.

Much of glmnet’s functionality for real-valued responses extends to these response types as well. As such, these chapters will focus on functionality that is unique to each response type.