Abstract
The output of a machine learning algorithm can usually be represented by oneor more multivariate functions of its input variables. Knowing the globalproperties of such functions can help in understanding the system that producedthe data as well as interpreting and explaining corresponding modelpredictions. A method is presented for representing a general multivariatefunction as a tree of simpler functions. This tree exposes the global internalstructure of the function by uncovering and describing the combined jointinfluences of subsets of its input variables. Given the inputs andcorresponding function values, a function tree is constructed that can be usedto rapidly identify and compute all of the function's main and interactioneffects up to high order. Interaction effects involving up to four variablesare graphically visualized.