Re: [Bug-gnubg] New Contact Net Results In Snowie Vs GNU Series:

[Joseph Heled]

Hi,

1296 is a good number. I use it too, but for 0ply rollouts. The noise is 
there, but it gives a good comparison - PROVIDED you have enough trials. 
150 is far too small. (My benchmark contains 100K positions).

Which ply was used for GNU and SW4.0? Also in cases of disagreement, it 
would be nice to see rollouts from more bots, not only SW3.2

-Joseph


address@hidden wrote:
> In a message dated 16/02/03 21:01:03 GMT Standard Time, address@hidden 
> writes:
>
>
>
> > Can I get more details?
> >   - Which positions?
> >   - Who is the Judge?
>
>
> Hi Joseph
> These positions came from a 500 game money session played between GNU15OCT 
> build and Snowie 3.2.
> They were analysed by GNU, Snowie 3.2, Snowie 4 and JellyFish 3.0.
>
> Any positions where any of the bots disagreed by >0.03 equity have been 
> rolled out using Snowie 3.2 1296 trials played 2 ply (GNU 1 ply) cube 3 ply.
>
> If the rollouts showed a BOT had erred by at least 0.04 they were included in 
> the series and have been posted on a daily basis on the GammonLine forum 
> where they have been analysed/discussed by the members.
>
> Although no rollout can ever give a definitive answer of course, they do seem 
> to be quite reliable and I think only one position so far has shown some 
> disagreement between SW rollout and GNU rollout. Although it was only rolled 
> out 144 trials using GNU.
>
> If you would like more info please email me.
>
>
> Michael