Re: [boost] metrics and tree stuctures

From: Jeff Garland (jeff_at_[hidden])
Date: 2005-07-03 10:53:03

On Sun, 3 Jul 2005 15:07:49 +0000 (UTC), Thorsten Ottosen wrote
> D. Kevin McGrath <dmcgrath <at> cs.indiana.edu> writes:
> 
> > 
> > I'm interested in adding support for arbitrary metrics (euclidean  
> > distance, taxi-cab, etc) to boost so that I can make use it, as well  
> > as some alternate tree structures, such as kd-trees.  I can't quite  
> > figure out where they would be most appropriately placed, and it  
> > doesn't seem as if there is enough (or enough similarity) to have a  
> > totally separate library.  Anyone have any thoughts?
> 
> It might make sense to put it Boost.Stat, ie, a statistics package 
> for boost.
> 
> There are some basic stuff in the sandbox in the "stat" directory, 
> but I have stalled development currently (for several reasons, most 
> notably lack of time). My next step would have been to add generic 
> matrix algorithms to implement stuff like inversion. Anyway, I feel 
> kd-trees and metrics would fit nicely into this lib.

The statistics algorithms fit into a the general concept of algorithms.  We
haven't done this consistently yet, but we've started putting these things
into groups under 'algorithms'.  For example, 'boost/algorithm/string' and
'libs/algorithm/string' for string algorithms. Also, as I recall Tom Brinkman
had some proposals for statistics around...which I don't think have gone
anywhere yet....

kd-trees seem like they should be a separate library to me.  They have uses
outside statistics.  There are also a couple other 'tree library' proposals
floating around somewhere.  It's an interesting question if we should consider
refactoring the boost tree to have something like 'libs/containers' for the
plethora of container types in boost...

Jeff