Re: [help-GIFT] Nearest-Neighbour Vector space retrieval?

[Wolfgang Müller]

On Tuesday 28 September 2004 09:59, David Squire wrote:
> Hi folks,
>
> If I wanted to add a module to to nearest-neighbour,
> Euclidean-distance-in-feature-space retrieval to GIFT, which would be
> the source files/directories to go looking in - if it doesn't already
> exist?  :)

Oh, I surely have programmed it somewhere, but not as part of the GIFT, and as 
the Plugin part is the part you need time on, it would take me more time to 
dig that out than for you to write it. 

I would try using the PluginMaker. If you've got problems with that thingy, I 
would be VERY interested in bug reports and getting it running together with 
you (this effect is known as: nobody's used it in a long while).

> PS. I am just thinking of dumb exhaustive search at the moment. Smart
> indexing structures can wait for now.

Experience shows that in highdimensional spaces you need quite large 
collections in order to be faster than dumb exhaustive search. Intelligent 
exaustive search a.k.a. VA-files and related ideas are faster than trees on 
large collections.

Cheers,
Wolfgang

P.S.: Do yourself a favour and use mmap(2).

-- 
Dr. Wolfgang Müller
LS AI 1
Universität Bayreuth