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

[Massimiliano Maini]

Well, that's exactly Misja's initial question:
why the cubeless data for
match is different from the one for money if they
both come from the 2s
bearoff db ?

He was expecting to see the same figures. I don't
think there's a reason
to have same figures, imagine whichever position you
want, at crawford
and at money: the cubeless figures shuld be different
due to the fact
that you play @ score in match (i.e. leader is in
gammon save mode).

But if you say that cubeless figures comes from the
db for money and for
match, then why are we seeing different figures ?

MaX.

Christian Anthon <address@hidden>
wrote on 26/06/2009 08:57:57:

> I believe that the cubeless data are extracted from the 2-sided db
as
> well. This results may well be different from that of an evaluation
> using the one-sided db. Try moving the 2-sided db out of the way and
> repeat the experiment.
> 
> Christian.
> 
> On Fri, Jun 26, 2009 at 8:52 AM, Misja Alma<address@hidden>
wrote:
> > Hi All,
> >
> > Thanks MaX for posting the question on this forum.
> >
> > So to Christian: You were saying that the cubeful bearoff db
is used
> > for moneygame but not for matches.
> > But the differences that were spotted here were for the cubeless
equities?
> >
> > Cheers,
> > Misja
> >
> > On Thu, Jun 25, 2009 at 9:00 PM, Christian
> > Anthon<address@hidden> wrote:
> >> The two-sided cubeful bearoff db is exact and used in preference
of
> >> evalutation, but only for moneygame.
> >>
> >> Christian.
> >>
> >> On Fri, Jun 19, 2009 at 3:56 PM, Massimiliano
> >> Maini<address@hidden> wrote:
> >>>
> >>> Hi all,
> >>>
> >>> on a forum I've found this: same position (in 2-sided
bearoff dB)
> >>> in money play and match play (3aw-3aw) shows different
cubeless
> >>> equities. I tried to explain it but I run into some confusion.
> >>>
> >>> That's money play:
> >>>
> >>>     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)
> >>>
> >>> Hint shows this:
> >>>
> >>> Cube analysis
> >>> 2-ply cubeless equity  +0.4971
> >>>   0.7485 0.0000 0.0000 - 0.2515 0.0000 0.0000
> >>> Cubeful equities:
> >>> 1. Double, take         +0.9158
> >>> 2. Double, pass         +1.0000  (
+0.0842)
> >>> 3. No double            +0.7661
 ( -0.1496)
> >>> Proper cube action: Redouble, take
> >>>
> >>> From eval (ctrl+E) I see (uninteresting parts removed):
> >>>
> >>> Position ID:        6AUAADoDAAAAAA
> >>> Match ID:        UQlgAAAAAAAA
> >>>
> >>> Evaluator:         Bearoff2
> >>>
> >>>              Player
      Opponent
> >>> Position          880  
        912
> >>>
> >>>         Win     W(g)  
 W(bg)   L(g)    L(bg)   Equity    Cubeful
> >>> static: 0.7485  0.0000  0.0000  0.0000
 0.0000   +0.4971   +0.7661
> >>>  1 ply: 0.7485  0.0000  0.0000  0.0000
 0.0000   +0.4971   +0.7661
> >>>  2 ply: 0.7485  0.0000  0.0000  0.0000
 0.0000   +0.4971   +0.7661
> >>>
> >>> Static, 1ply and 2ply figures are identical, which makes
me think they all
> >>> come from the 2s bearoff dB (eval output is a bit confusing
here).
> >>>
> >>> That's the same for match:
> >>>
> >>>     GNU Backgammon  Position ID: 6AUAADoDAAAAAA
> >>>                
    Match ID   : UQlgAAAAAAAA
> >>>     +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 3 point match
> >>>   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)
> >>>
> >>> Cube analysis
> >>> 2-ply cubeless equity  +0.4797 (Money:  +0.4797)
> >>>   0.7399 0.0000 0.0000 - 0.2601 0.0000 0.0000
> >>> Cubeful equities:
> >>> 1. Double, take         +0.9565
> >>> 2. Double, pass         +1.0000  (
+0.0435)
> >>> 3. No double            +0.7841
 ( -0.1724)
> >>> Proper cube action: Redouble, take
> >>>
> >>>         Win     W(g)  
 W(bg)   L(g)    L(bg)   Equity    Cubeful
> >>> static: 0.7485  0.0000  0.0000  0.0000
 0.0000   +0.4971   +0.8543
> >>>  1 ply: 0.7399  0.0000  0.0000  0.0000
 0.0000   +0.4797   +0.8308
> >>>  2 ply: 0.7399  0.0000  0.0000  0.0000
 0.0000   +0.4797   +0.7841
> >>>
> >>> The original question was: why the 2-ply cubeless GWC
are different
> >>> between money and match ?
> >>> First answer was: it's using bearoff dB from money and
not for match.
> >>> Objection was that gnubg could/should use the dB even
for match, at
> >>> least for the cubeless GWC figures.
> >>>
> >>> Here I tried to explain that cubefull figures from the
dB can only be
> >>> used in money play, but what about cubeless ? Are cubeless
figures from
> >>> the dB used as a base for cubefull eval in match play
?
> >>>
> >>> Looking at the figures I would say yes: the static cubeless
GWC for match
> >>> are exactly the ones you have in money. But they diverge
at 1 and 2 ply.
> >>>
> >>> Am I right if I say that doing 2ply eval in match, gnubg
needs to compute
> >>> a lot of 0ply cubeless evals and it takes them from the
dB. The figures
> >>> are converted into 0ply cubeful (Janowski formula) and
then 
> "backpropagated"
> >>> to get 2ply cubeful. 2ply cubeless is computed by inversing
the Janowski
> >>> formula on 2ply cubeful figures.
> >>>
> >>> If all the above is true, I do see the objection coming:
it's 
> true that what
> >>> matter are cubeful figures (and these cannot be taken
from the 
> dB in match),
> >>> but why gnubg isn't showing the "right" cubeless
GWC (the ones 
> from the dB)
> >>> in match ?
> >>>
> >>> Side question: I tried to use bearoffdump. I manage to
get the 
> info putting
> >>> in the position ID:
> >>>
> >>> D:\Documents\gnubg>bearoffdump.exe -p 6AUAADoDAAAAAA
gnubg_ts0.bd
> >>> Bearoff database: gnubg_ts0.bd
> >>> Position ID     : 6AUAADoDAAAAAA
> >>>
> >>> Information about database:
> >>>
> >>>  * On disk 2-sided bearoff database evaluator
> >>>    - generated by GNU Backgammon
> >>>    - up to 6 chequers on 6 points (924 positions)
per player
> >>>    - database includes both cubeful and cubeless
equities
> >>>    - number of reads: 0
> >>>
> >>>
> >>> Dump of position ID: 6AUAADoDAAAAAA
> >>>
> >>>              Player
      Opponent
> >>> Position          880  
        912
> >>>
> >>> Cubeless equity            
  : +0.4971
> >>> Owned cube            
       : +0.7661
> >>> Centered cube            
    : +0.7451
> >>> Opponent owns cube          
 : +0.4579
> >>>
> >>> But I had no success in using the -n option (enter directly
the
> >>> position indexes, 880 and 921): how is this supposed
to work ?
> >>>
> >>> MaX.
> >>> _______________________________________________
> >>> 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
> >>
> >>
> >