Understanding tuning results

[Greg Louis]

On 20030605 (Thu) at 1311:51 -0400, David Relson wrote:

> That looks much nicer !  I'd choose 1e-2, 0.475, 0.503000 from the equally 
> good results and see how it goes over the next few days.  Since I've used 
> 1e-2 for robs for a while, it'd be comfortable to continue using that value.
> 
I'm very distrustful of min_dev that high.  It's the bogus ledge that
I'm fairly sure is an artefact, though I haven't yet worked out how
it's caused.  A lot of the time, with min_dev up there, you're going to
be deciding on the basis of one or two tokens.  I'd be happier
sacrificing a couple of tenths of a percent and going with a lower
min_dev for the sake of reliability.

I ran for a while with 0.44 and that was more or less ok, but even then
I got the occasional anomaly.

-- 
| G r e g  L o u i s          | gpg public key: finger     |
|   http://www.bgl.nu/~glouis |   glouis at consultronics.com |
| http://wecanstopspam.org in signatures fights junk email |