Agreed. A fuzzy selection mechanism consists of a few floating point operations to
determine where, on a curve from 0.0 to 1.0 percent probability, a given value lies.
A lot of fuzzy systems use multiple sets, which require this selection mechanism to be
invoked many times, however, for a single set (i.e. a 'match' or 'no match') only one
is
required. Perhaps a complex set was used which caused the significant overhead you
saw?
I could do some benchmarks on a beta curve selection mechanism (the curve I prefer,
since
probability is never zero, just very low, and requires a fair amount of affinity to
optimal value before probability reaches >75%).
However, this operation should take no more than a few dozen microseconds at absolute
worst.
Ian Clarke wrote:
>
> On Sun, Jun 03, 2001 at 10:52:46AM +0100, Adam Langley wrote:
> > On Sat, Jun 02, 2001 at 10:47:01PM -0700, Ian Clarke wrote:
> > > It is the popular conception that fuzzy searching in Freenet would not
> > > be possible, yet early-on we discussed a proposal which could achieve
> > > just this in an efficient and elegant manner.
> >
> > Thought about it. Made up a simple simulator. Found the fuzzy matching
> > functions were so damm slow it wasn't viable to simulated it. Gave up.
>
> News to me. What fuzzy matching functions did you use? I have found
> some pretty efficient ones.
>
> Ian.
>
>
>------------------------------------------------------------------------------------------
> Part 1.2Type: application/pgp-signature
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
