Representing Points in Many Dimensions by Trees and Castles

摘要: Abstract A number of points in k dimensions are displayed by associating with each point a symbol: drawing tree or castle. All symbols have the same structure derived from hierarchical clustering algorithm applied to variables (dimensions) over all points, but their parts coded according coordinates individual point. Trees and castles show general size effects, change whole clusters point, trends, outliers. They especially appropriate for evaluating observing points. Their major advantage earlier attempts represent multivariate observations (such as profiles, stars, faces, boxes, Andrews's curves) lies matching relationships between features representing symbol. Several examples given, including one 48 variables.