[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Bug-gnubg] Re: remote server code fixed I think
From: |
kvandoel |
Subject: |
[Bug-gnubg] Re: remote server code fixed I think |
Date: |
Thu, 4 Sep 2003 21:38:48 +0200 (CEST) |
One more data point: With everything set to expert all error rates are 0
when playing gnu-gnu 5pt match from one instance. [When one is external
BOTH players error rates become non-zero!]
Kees
On Thu, 4 Sep 2003 address@hidden wrote:
> On Thu, 4 Sep 2003 address@hidden wrote:
>
> > They way I test it is to set everything to expert level and do a match
> > analysis: all error rates should be 0. As it is with version 1.1343 if I
> > play like that, the external player makes "errors", meaning it's not
> > playing at expert level.
> >
> > I've tried 3pt matches like that, both with player 0 and player 1
> > external. The only errors are now chequerplay errors. The errors are
> > larger when player 0 is the external.
> >
> > The errors are very small (rating equiv. of 18 (player 0) and 40
> > (player 1)), yet they should be exactly 0.
>
> PS. I just ran another test on 5pt matches and I noticed that even for
> the local player sometimes the error rate is not 0! Difference is very
> tiny, equiv. of 2 rating points maximum.
>
> This is strange because a while ago when I did my rating/error
> simulations I did a series of "sanity tests", verifying that my scripts
> indeed spitted out "0" if played and analysed at the same level and it
> always did, so this is a problem recently introduced.
>
> To me this suggests a rounding problem. Perhaps the external code
> suffers from a more severe form of it?
>
> For the external player the errors of the external are sometimes as big
> as 150 rating points equiv.