DM: Fwd: Parallel Coordinates -- previous message attached

[Dorothy Firsching]

>Date: Fri, 7 Aug 1998 20:17:08 +0300 (GMT+0300)
>From: Alfred Inselberg <aiisreal@math.tau.ac.il>
>Reply-To: Alfred Inselberg <aiisreal@math.tau.ac.il>
>To: owner-datamine-l@crosslink.net
>Subject: Parallel Coordinates -- previous message attached
>
>
>To: owner-datamine-l@crosslink.net 
>    Subject: Re: DM: Parallel coordinate plots 
>    From: bwallet@nswc.navy.mil (Brad Wallet) 
>    Date: Wed, 13 Aug 1997 23:31:50 -0400 
>
>
>
>While Inselberg may have have written the first paper on parallel
>coordinates, Wegman is the one who really brought this method to
>usefulness in EDA.  Wegman (90, JASA) is the defining piece.  His
>more recent work, particularly involving R^d->R^d grand tours as
>well as saturation for massive data sets makes parallel coordinates
>a must for anyone serious about data exploration.
>
>Brad
>
>===========================================================================
===
>
>Dear Brad,
>
>Starting in 1985 one of my students T. Chomut at the UCLA Comp.Sc. 
>Dept for
>his M.Sc. wrote a software package called EDA which used Parallel
>Coordinates to do EDA and also did a case study using data on 
>European
>Food Consumption. This work is documented and referenced and Prof. 
>Wegman is
>well aware of it. In the AT&T Workshop on Data Visualization in 
>Statistics
>this June 1998, also attended by Prof. Wegman, Prof. Luke Tierney 
>from the 
>Univ. of Min. pointed out that in his book (1990?) that the Grand 
>Tour 
>using Parallel Coordinates is given as an exercise.
>
>Most often Parallel Coordinates (||-coords) are used as "glyphs"; 
>that is
>without taking notice or advantage of the neat properties of the
>COORDINATE SYSTEM. In general multidimensional relations can NOT be 
>seen
>directly from their points (data) but ONLY by RECURSIVELY building 
>up higher
>dimensional components FROM the points. Of course that requires 
>study like
>all good things see the elegant paper Gennings, CARTŪ er et al 
>(Biometrics 90) for example. In simplitic uses of ||-coords people
complain of
>"overplotting", "distance nonlinearities" etc and some decide to 
>"study"
>density without understanding that in ||-coords "density" can not be
>measured linearly, ditto for "saturation". As for Grand Tours there 
>is a
>translation <---> rotation duality which greatly simplifies things 
>i.e. there is no need to compute the rotations for the Grand Tour. 
>
>Still I am happy that you recommend ||-coords and hope that
>more people would take advantage of their more powerful properties.
> 
>BTW since you mentioned "precedences", I proposed ||-coords in a
>graduate Topology course I was taking at the Univ. of Ill. in 1959. 
>My
>thought was that in geometry the fundamental property is parallelism,
>rather than orthogonality which requires a notion of angle (while
>parallelism does NOT -- i.e. lines without points on common). It was 
>that
>and the fact that orthogonality "used-up" the plane very fast that 
>led 
>to the idea.
>
>Best regards
>
>Al Inselberg
> 
Dorothy Firsching
CEO
Nautilus Systems, Inc.
3867 Alder Woods Court
Fairfax, VA  22033
http://www.nautilus-systems.com/
nautilus-info@nautilus-systems.com