DM: Re: CSI DVW Preservation of "Nearness"

[Monte Hancock]

Dr. Sarle writes:


>
>In any linear 2D projection, points which are proximate in Rn remain
>proximate on the display. What's special about what CSI Visualizer
>does?
>

   The CSI Visualizer is a tool intended to support manual data 
mining.  It
integrates
descriptive statistics with an interactive perspective view of
high-dimensional
point-sets:  the user can "grab" visual structures (e.g., clusters) 
directly
from
the display, assign them names, color code them, apply mensuration
techniques,
fly to another perspective, perform autoclustering, etc..  Users have 
found
this method
of direct "hands-on" interaction with data they are analyzing to be a 
useful
adjunct
to standard "blind" analysis techniques.  The primary consideration 
in the
design of
the visual representation is "user intuition".

>
>These examples [mentioned in a previous communication] are not 
>challenging.
>What happens with nontrivial examples such as those in Furnas & Buja
(1994),
>"Prosection views:  dimensional inference through sections and 
>projections,
J. of
>Computational and Graphical Statistics, 3, 323-385?

>

   Quite so.  The current release of the CSI Visualizer does not 
render
manifolds,
so the F&B prosections have not been benchmarked.  However, as noted 
in the
previous communication, CSI has just completed a prototype app which 
does
raytraced graphics of manifolds in Rn.  This may be added to the 
Visualizer
if
there is market interest.

>
>Since the original announcement was posted to
>datamine-l@nautilus-sys.com, it would be nice to have an explanation
>posted to that list of what the marketese really means.
>

   Sorry, didn't mean to offend.  Our target market is not the 
research
community.