Re: [Bug 3821] scores are overoptimized for training set

[Justin Mason <jm...@jmason.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Loren Wilton writes:
> > BTW, this is the "rule reliability tflag" idea again; basically provide a
> way to
> > hint that this rule is reliable, and this rule should not be considered
> reliable
> > -- no matter what their hit-rates in mass-checks were.
> >
> > I agree it may have good effects as a hint to the Perceptron, so it may
> now be
> > time to do this.  what d'you think, Henry?
> 
> Note that Bob M. has a hint comment of his own that gives several levels of
> hint, not just a binary value.  He uses this for his own scoring tool with
> good results.
> 
> I think that the idea of a multi-level hint is a good one and should be
> considered.  I don't know if that concept will fit in tflags.  If not,
> perhaps some other ("scorehint") could be cconsidered.

yeah -- definitely -- I was thinking that, although I didn't mention
it. ;)   imo a new config command (I was thinking "reliability"
or similar) would be good.

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFBWMCZQTcbUG5Y7woRAu8KAKDvZuLSPDziv73jJ0vuB6tJckagwQCgk4cI
QtCGKENa11sgPI9zme5ma3M=
=Wvfm
-----END PGP SIGNATURE-----