[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] regression of recent tuning patches
From: |
Paul Pogonyshev |
Subject: |
Re: [gnugo-devel] regression of recent tuning patches |
Date: |
Mon, 11 Apr 2005 00:17:04 +0300 |
User-agent: |
KMail/1.4.3 |
Gunnar wrote:
> > This patch solves kgs:500. It also removes an invalid answer
> > from that test. Other regression impact is not tested.
>
> ego:8 PASS F12 [F12]
> ninestones:150 PASS C5 [C5|D5]
> kgs:500 PASS O15 [O15]
> 3 PASS
> Total nodes: 1675335803 3152538 13128014 (-0.037% -0.096% +0.022%)
>
> Looks fine!
I wrote:
> I also revised the owl tuning patch a bit: it is useless
> to specify which dragon runs to which as connection works
> in both directions...
>
> ...
>
>
> ego:8 PASS F12 [F12]
>
> Did not verify.
>
> nngs:510 FAIL P13 [G14]
>
> Don't understand it, probably accidental.
>
> auto04:2 FAIL 0 [!0]
>
> Mostly accidental. Valid move doesn't falls off because of
> branching cut off: new D1393 has a higher value. But the
> reading is not quite perfect anyway. This is a case where
> it is actually ``getting deep, looks lively'', but GNU Go
> doesn't think so. More often it is the way round.
>
> ninestones:150 PASS C5 [C5|D5]
>
> Accidental: owl node limit happens in a different reading
> branch now in reading of G15.
>
> kgs:500 PASS O15 [O15]
>
> The targeted test case, solved properly.
So, apparently my revision resulted in two FAILs. It still seems
logically correct to me and I think we should put the revised
version in CVS even if it accidentally gives worth results on
regressions. Also, in the long run this shouldn't matter much if
Arend finishes his algorithmic-owl-escape patch.
Any objections?
Paul