Contrast Trees and Boosting

Contrast trees represent a new approach for assessing the accuracy of many types of machine learning estimates that are not amenable to standard (cross) validation methods; see "Contrast trees and distribution boosting", Jerome H. Friedman (2020) . In situations where inaccuracies are detected, boosted contrast trees can often improve performance. Functions are provided to to build such trees in addition to a special case, distribution boosting, an assumption free method for estimating the full probability distribution of an outcome variable given any set of joint input predictor variable values.

Contrast Trees and Distribution Boosting

Contrast trees are used to assess the accuracy of many types of machine learning estimates that are not amenable to standard validation techniques. These include properties of the conditional distribution $p_{y}(y,|,\mathbf{x})$ (means, quantiles, complete distribution) as functions of $\mathbf{x}$. Given a set of predictor variables $\mathbf{x}=(x_{1},x_{2},$$,x_{p})$ and two outcome variables $y$ and $z$ associated with each $\mathbf{x}$, a contrast tree attempts to partition the space of $\mathbf{x}$ values into local regions within which the respective distributions of $y,|,\mathbf{x}$ and $z,|,\mathbf{x}$, or selected properties of those distributions such as means or quantiles, are most different.

For more details, please see the tutorial.