Class 16 Stat701 Fall 1997

Graphical Analysis of High Dimensional Data

Todays class.

• The need for exploring high dimensional data.
• Interaction.
• Conditioning plots and Trellis.
• A technical paper out of Bell Labs on Trellis.

This is going to be an unusual class as it will be based entirely on issues concerning the graphical representation of high dimensional data. There is not alot of theory but here are some examples of items we will cover:

Image 1 shows a Trellised scatterplot conditioned over the weight variable, with a smooth superimposed.

Image 2 shows a 3-D rotatable scatterplot, that is infact dynamically linked to a scatterplot matrix.

Image 3 shows a 3-D surface, formed by using a smoothing spline. The color coded contours produce a reasonable representation of 3-D surface .

Image 4 shows the same 3-D surface, but now viewed from 4 different angles.

Finally Image 5 displays a two dimensional scatterplot, Trellised or conditioned over the levels of two other variables. Hence it is looking at a representation of a four dimensional relationship. Furthermore, with the addition of color an extra dimension is added, and together with varying symbol iszes 5 different variables are represented in this plot.