The Tiny Tour
It starts with nine views
of random or user-specified projections. Figures in
sun.cwru.edu/~jiayang/nsf/ipp.html provide a birds-eye
view of some features of our ipp. It then continues, by a
user's click on the ``optimize'' button, with nine (not one) most
informative solutions (maxima), based on Friedman's (1987) indices.
Once an interesting projection is uncovered, the user may
investigate it further, entering a new display that features only a
single projection and/or its neighborhood and provides the P-value
(calculated following Sun, 1991 and Fleischer and Sun, 1998) of the
projection and summary statistics of the first four moments of that
projection. Other options, such as changing the order, J, of the
Legendre polynomials, assessing the importance of a particular
variable on the PP index (for variable selection), and removing the
current interesting projection (called structural removal by
Friedman, 1987), are available for going into the next level of
projections. Other multivariate analysis tools such as dendrograms,
perspective plots and contour plots are also linked to the 2-d PP.
ipp is very easy to use. It was successfully applied to a
large messy data set collected by Bill Black on college football,
which has many missing values (see sun.cwru.edu/~jiayang/ft.html).
For a quick but not so small a tour, go
here for a tour
of version 2.