Augmenting Reduction Plots with Localized Explanations

Abstract

Multidimensional projection have been a common technique to aid the analysis high dimensional data, techniques exist that attempt to explain the projection using per point or global techniques. Neither can convey to the user why points are placed in a certain neighborhood. We will implement a localized explanation technique which achieves this based on the attributes [1] and dimensionality [2] and extend them to 3D. We aim to make a tool which is both correct and works at interactive rates, by doing so making the techniques more accessible. We will also use thistool to apply the techniques to a series of real world dataset to provide examples of the types of insights possible. We also compare 2D and 3D for 29 projections techniques to investigate the additional value of the 3rddimension.