[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] How strong is the lookahead?
From: |
Erez |
Subject: |
Re: [Bug-gnubg] How strong is the lookahead? |
Date: |
Mon, 30 Mar 2009 19:44:30 +0300 |
Hi Boomslang,
This is quite remarkable - great work!
Thanks :)
Now I just need to think why did both machines crash when I tried to do what
I did...
----- Original Message -----
From: "boomslang" <address@hidden>
To: <address@hidden>
Sent: Monday, March 30, 2009 7:35 PM
Subject: Re: [Bug-gnubg] How strong is the lookahead?
Hi Erez,
About three years ago I let GNUBG play 26000 one pointers on FIBS at two
settings: Expert and Supremo. Based on its wins and losses, I estimated the
FIBS rating of the 2 settings using Maximum Likelihood. The results were:
Chequer play Point 95% Confidence
setting estimate interval
--------------------------------------------
expert 2034 (2009, 2059)
supremo 2098 (2055, 2142)
A difference of about 65 FIBS points (based on 1 pointers).
Greetings, boomslang
(And yes, even 0ply plays 2000+ level...)
--- On Mon, 30/3/09, Christian Anthon <address@hidden> wrote:
From: Christian Anthon <address@hidden>
Subject: Re: [Bug-gnubg] How strong is the lookahead?
To: "Erez" <address@hidden>
Cc: address@hidden
Date: Monday, 30 March, 2009, 1:25 PM
2009/3/30 Erez <address@hidden>:
> I wanted to know what is the true addition of the
lookahead to GNU so as a
> true lamer I set out a money game session between GNU
playing expert and GNU
> playing Supremo. I thought to let them play until 64
points and see Supremo
> mop the floor with expert.
> I was never able to get to 64 because the software
crashed somewhere in the
> middle (happened again on another computer), but still
I was very surprised
> to find out that Supremo got only about 20%
more points than Expert.
> So how much stronger really is Supremo than Expert?
>
Not much I expect. The strength of the higher plies are
mostly seen in
specific situations where the look-ahead is important. It
is more a
question of reliability than overall strength. 64 points
certainly
won't be enough to determine the difference with any
confidence. The
best approach is probably to let 0ply play 0ply for a long
money
session. Have supremo analyse the games for both players,
and then
rollout any disagreements. Then you'll have error
rates to compare
for the two playing strengths.
Christian.
> Erez
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>
_______________________________________________
Bug-gnubg mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/bug-gnubg
_______________________________________________
Bug-gnubg mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/bug-gnubg