ProjPursuit

Selecting this button generates a window in which the projection
pursuit index for the current two dimensional view of the data is
plotted.  As the tour progresses the index is plotted continuously.
The value is also displayed directly beneath this button.

Selecting this button also turns on the "PrnCmp Basis" button and
spheres the data by principal components.  Three of the projection
pursuit indices are polynomial-based.  These indices are measures of
the difference between the empirical density and the standard normal
density.  There is no restriction in the software for running the
projection pursuit with "PrnCmp Basis" turned off, but beware that this
may not give sensible results for the polynomial indices.

When "Optimz" is selected, the tour is guided by the gradient of the
projection pursuit index, so that XGobi begins to search for views with
high index values.  Hopefully these views will expose interesting
features of the data.  When a maximum value of the projection pursuit
index is reached the tour may turn itself off (this occurs when the
gradients are all close to zero, meaning we have gone as far as
possible in this direction to capture a maximum, and it is not possible
to find any better direction).  To restart the tour turn "Optimz" off.

There are three types of indices currently implemented.  One type is
based on density estimation using expansions in orthogonal polynomials
(Natural Hermite, Hermite, Legendre), another is based on kernel
density estimations (Friedman-Tukey, Entropy).  In each of these
methods there is a parameter which it is desirable to alter from time
to time.  The scrollbar allows the interactive changing of these
parameters.  For the polynomials it controls the number of terms used
in the expansions.  For the kernel density estimation it controls the
window width of the kernel. A third index type is based on detecting
particular types of structure (Holes, Central Mass, Skewness). There
are no parameter inputs to these indices.