[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] quasi random dice and 2ply rollouts
From: |
Jim Segrave |
Subject: |
Re: [Bug-gnubg] quasi random dice and 2ply rollouts |
Date: |
Thu, 15 Apr 2004 19:48:14 +0200 |
User-agent: |
Mutt/1.4i |
On Thu 15 Apr 2004 (13:19 +0200), Achim Mueller wrote:
> Hi folks,
>
> at Nordic Open I bought the BG Enyclopedia by Kit Woolsey and
> started comparing gnubg with the results from Snowie Rollouts (I
> guess, Kit used Snowie 3, 360 3ply full rollouts with variance
> reduction).
>
> A technical question now. When I activate quasi-random dice, is it
> better to set up a multiple of 36 (i.e. 324) or doesn't this matter
> much and I can also use 360 rollouts?
Quasi random dice gives a "fairly random" set of rolls with the added
property that you get a full set of possible first rolls every 36
games, a full set of every possible pair of first roll/second roll
every 1296 games, every possible first/second/third roll every ...
I suppose, when rolling out a position, very volatile situations,
where there are big winning/losing rolls for each side in the next
pair of rolls would suggest that 1296 is needed. Many situations
probably aren't that volatile and you will see the equities converge
on a value with little fluctuation. After that occurs, it's unlikely
that you are missing anything by not extending the rollouts.
--
Jim Segrave address@hidden