simultaneous multiple key sorting algorithm

[Andrei Alexandrescu]

On 1/28/12 10:29 AM, H. S. Teoh wrote:
> Actually, the reason I chose hypercubes was because I was hoping that
> the convex hull with hypercubes in the original features' space would be
> equivalent (or easily massaged to be equivalent) to using simplexes in
> rank space.  (Although the part about the (1,1,1,...) vector may not
> work out quite that nicely in this case.) Otherwise it doesn't really
> give you a good algorithm since you'll have to precompute the ranks.

I understand. You're up to something; if you get to prove anything in 
that direction, I'd be interested. Thanks!

Andrei