-
Book Overview & Buying
-
Table Of Contents
-
Feedback & Rating

The Kaggle Book
By :

Not all objective functions are suitable for all problems. From a general point of view, you’ll find two kinds of problems in Kaggle competitions: regression tasks and classification tasks. Recently, there have also been reinforcement learning (RL) tasks, but RL doesn’t use metrics for evaluation; instead, it relies on a ranking derived from direct match-ups against other competitors whose solutions are assumed to be as well-performing as yours (performing better in this match-up than your peers will raise your ranking, performing worse will lower it). Since RL doesn’t use metrics, we will keep on referring to the regression-classification dichotomy, though ordinal tasks, where you predict ordered labels represented by integer numbers, may elude such categorization and can be dealt with successfully either using a regression or classification approach.
Regression requires you to build a model that can predict a real number...