[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] Bearoff dB position, few questions

From: Misja Alma
Subject: Re: [Bug-gnubg] Bearoff dB position, few questions
Date: Fri, 26 Jun 2009 13:41:03 +0200

Thx for the explanations, I think I understand now how the cubeless
chances are calculated by gnu. It's a bit confusing since they are
apparently not completely cubeless, but if you know what they are then
I think they are still meaningful.

About the side question:
The cubeless GWC are depending on the score as long as there are
gammons possible in the position.
Depending on the matchscore, the relative value of winning a gammon or
a single game can be different. To take advantage of this you can find
checkerplays to maximize or to avoid gammons, and because of that the
GWC of the same position can be different for certain scores.


On Fri, Jun 26, 2009 at 1:30 PM, Massimiliano
Maini<address@hidden> wrote:
> Massimiliano Maini
> Product Manager Availability and Schedules
> Availability and Shopping
> Amadeus IT Group SA
> T: +33.(0)
> address@hidden
> www.amadeus.com
> address@hidden wrote on 26/06/2009
> 12:58:56:
>> On Fri, Jun 26, 2009 at 12:37 PM, Misja Alma<address@hidden> wrote:
>> > Hi Alain,
>> >
>> > Yeah you're right that cubeless equities can well be different for
>> > matchplay and money play.
>> > But not only is the equity different, also the 'Win' percentage is
>> > different that is shown together with the W(g), W(bg) etc when you
>> > click the 'hint' button.
>> > The winning chance refers to the chance of just winning the game, so
>> > it should be the same for moneygame and matchplay. Or at least in
>> > cubeless bearoffs it should, because there are no gammons to save and
>> > all that.
>> >
>> > Misja
>> Keep in mind that cubeless evaluations depend on cubeless/cubeful
>> setting since the moves may depend on the score and cube value. So can
>> you show me the output without the 2 sided db and without cubeful
>> evaluation.
> I think it's simpler than that:
> - open the money position, ask for a hint at 0/1/2/3ply: the GWC will not
> change, as they always come from the 2s bearoff db.
> - open the match position, ask for a hint at 0/1/2/3ply: the GWC will
> change each time. I suspect this comes from the fact that gnubg takes
> 0ply GWC (for positions N plies ahead, if we are a N-ply) from the 2s
> db and then apply 1/2/3 ply on top. I think gnubg does this because he
> does the same for cubeful equities. It would probably be better/more
> logical to take cubeless GWC for the current position from the dB,
> without any ply applied on top.
> In other words, for match play: for a 2ply eval (cubeful and cubeless)
> gnubg needs 1ply eval of subsequent positions which needs 0ply eval of
> "subsequent subsequent" positions. I think 0ply cubeless GWCs of
> "subsequent subsequent" positions come from the dB, but then they are
> plied to get 2ply cubeless of the original position (just like 0ply
> cubeful, obtained via janowski'd formula, are then plied to get 2ply
> cubeful of the original position).
> Side question (purely theoretical): is it true that cubeless GWC
> (including g and bg %) for a given position are the same no matter the
> score (including money play) ?
> This is somehow similar to supposing that the gammon ratio or percentage
> is the same across the whole MET ...
> MaX.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]