Re: DM: CSI Visualizer Workstation Announcement

[Monte Hancock]

-----Original Message-----
From: AARB <ariguer@ctc.cl>
To: Rhonda Delmater <rhonda@csihq.com>
Cc: datamine-l@nautilus-sys.com <datamine-l@nautilus-sys.com>
Date: Thursday, January 08, 1998 7:31 PM
Subject: Re: DM: CSI Visualizer Workstation Announcement


>
>
>Rhonda Delmater wrote:
>
>> CSI's Visualizer Workstation Provides Important New Data Mining 
>Insights
>>                 Data can be viewed in multi-dimensional space.
>

AARB asks:
>
>What are the main differences of this software with Minesset (SGI)?
>What is the size of de data wich is possible analize?
>
>

   My understanding is that Mineset (a product of Silicon Graphics) 
uses
non-spatial representation (e.g., animation) to depict 
hyper-dimensions.
The CSI Visualizer uses spatial representation.  The principal 
operational
difference is the whether you are "lossy in space, or lossy in time":

  1)  Animation does not show all information simultaneously in time 
(lossy
in time, since at no particular
        time is all information necessarily visible).
  2)  Spatial representations(*) do not show all information 
simultaneously
in space (lossy in space,
        since objects may be occluded).

   Clearly, there is a domain-dependent trade here.  For complex data
mining, we think contemporaneous spatial depiction, though 
space-lossy, is
usually more intuitive.  To garner some of the benefits of time-lossy
representations, the Visualizer supports, among other things, viewer 
roam.
I've found smooth "coptering" around a cluster set with the 
Visualizer to be
very enlightening.  (Side note:  We've got a Visualizer prototype that
double-paints the display from two vantage points and uses the old 
1950's
style red/blue "3-D" glasses...not bad, but not ready for prime time!)

   There are, of course, non-spatial representations besides animation
(e.g., certain brushing techniques).  We have chosen to rely on 
spatial
representation because we think it provides the intuition best suited 
to our
target market.


   (*)  Parallel coordinate representations (A. Inselberg's 
ground-breaking
work in visualization) are a good example of a well-known spatial
representation for hyper-dimensional data.

--M. Hancock
  Chief Scientist, CSI