bug-gnubg
[Top][All Lists]
Advanced

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

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


From: Robert-Jan Veldhuizen
Subject: Re: [Bug-gnubg] Bearoff dB position, few questions
Date: Sun, 28 Jun 2009 23:39:32 +0200



On Fri, Jun 19, 2009 at 3:56 PM, Massimiliano Maini <address@hidden> wrote:
 

    GNU Backgammon  Position ID: 6AUAADoDAAAAAA
                    Match ID   : UQkAAAAAAAAA
    +24-23-22-21-20-19------18-17-16-15-14-13-+  O: gnubg
 OO |          O  O  O |   |                  |  0 points
 OO |             O    |   |                  |  
 OO |             O    |   |                  |  
 OO |             O    |   |                  |  
  O |                  |   |                  |  
    |                  |BAR|                  |v
  X |                  |   |                  |  
 XX |                  |   |                  |  
 XX |          X       |   |                  |  
 XX |          X     X |   |                  |  On roll
 XX |       X  X     X |   |                  |  0 points
    +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: MaX (Cube: 2)

The original question was: why the 2-ply cubeless GWC are different
between money and match ?

First of all, this is a cube decision. The breakdowns you gave seem to come from the No Double evaluation only. It's very possible that the Double/take breakdown would look different. After all, in match play it's almost a drop already, so if you don't double this now, it could easily lead to different checker play (f.i. a less volatile play, so your cube next turn will be more efficient).

So I'd check that first.

Apart from that:

For money, the 2-sided database is perfect, so there's no lookahead needed. 0-ply gives the final answer already.

For matchplay, the 2-sided database is not perfect. Obviously cubeful equities from the db are of no use, therefore cube actions also need to be determined using normal lookahead. But moves can be different too, and GnuBG plays according to score (and cube!). So, 0-ply uses the db cubeless figures, applies "match-janowski"to get cubeful equities, and then uses the best CUBEFUL move. This is NOT always the best move for cubeless equity, nor is it always the same move that the database says is best for money cubeful.

Some plays with lower GWC will actually give a higher cubeful equity, because of better cube efficiency for the player or worse cube efficiency for the opponent.

The 2-sided database for money is cubeful and therefore knows even the best cubeful play, when it's different that the best cubeless play (this happens).

But not for matchplay. Then, a different best play can pop up when the bot does cubeful lookahead, even when using this database. The bot will choose the best CUBEFUL play at the score, and uses the cubeless values from the database FOR THAT PLAY.

Conclusion: from 1-ply on, GnuBG is making some different play(s) in this bearoff at -3,-3 than for money; at least it seems to dothis with teh cube still owned on 2.

--
Robert-Jan Veldhuizen

reply via email to

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