Pythagorean trees for visualizing trees in Orange

Pavlin Gregor Poličar (2016) Pythagorean trees for visualizing trees in Orange. EngD thesis.

Preview

PDF Download (3432Kb)

Abstract

Orange is a data mining toolkit that has, up to this point, only supported the classic graph method using nodes and edges for visualizing classification and regression trees. Due to space requirements this method is only useful when dealing with smaller trees. Alternative methods for visualizing trees are therefore needed. We have implemented an interactive widget that uses Pythagorean trees which conveys the tree structure clearly with small as well as with large trees. We also implemented a widget to visualize random forest using Pythagorean trees, something that Orange and similar programs did not yet support. The visualization enables users to inspect sections of trees in detail along with the selection of corresponding data in interesting branches. This could lead to better understanding of the data and underlying model. The random forest visualization is visually appealing and nicely shows each of the tree models that this popular method method infers from the data.

Actions (login required)