RE: [Bug-gnubg] Odd rollout results of classic reference position

[Ian Shaw]

Øystein wrote:

>I think this position may be misplayed in the rollouts. If it keeps the cube 
>for a roll, X may be able to get a incorrect take from O. I guess that what's 
>happening in the rollouts.


> Can you try to play this position a couple of times against GNU and see if it 
> does something funny in subsequent cube actions? (I guess chequer play will 
> be really simple in this position.)


You are correct, Øystein.
After No Double, X 62 6/0 6/4, Dance, O takes if X doubles. A rollout shows 
this to be a 0.03 pass. There are others, no doubt.

It's no big deal that GnuBg gets this position wrong. It won't be the last time 
a bot gets a well known position wrong.

But there must be a bug here:

> > Cubeful equities:
> > 1. Double, pass         +1.000
> > 2. Double, take         +1.061  ( +0.048)
> > 3. No double            +1.013  ( +0.000)
> > Proper cube action: Redouble, pass

How can No Double be both +1.013 and ( +0.000)? It ought to be ( +0.013).
The proper cube action can't be Redouble, pass

If the 1.013 is accurate, as seems likely from my further test, then the cube 
action should be Too Good to double, pass. This is clearly ridiculous with no 
gammons possible.
If the ( +0.000) is correct, then it is Optional Redouble, pass. (Does GnuBg 
have that?)

Looking at it another way. Perhaps GnuBg IS indeed correct to say, "Redouble, 
pass". It is clearly not Too Good, so it must be a double. The interpretation 
is: "It is technically a redouble and a pass. However, X can expect to gain 
equity by holding on to the cube and getting a bad take later." After all, if a 
bot is going to get it wrong, then we can expect humans to err, too.

So where does that leave us. Possibly the only problem is the "( +0.000)". If 
the rollout is making bad takes I guess that can't be helped, except by 
improving the net.

I have also learnt that a rollout isn't necessarily more accurate than an 
evaluation. 0 through 4-ply all get this problem right.

--Ian


> >
> >     GNU Backgammon  Position ID: tw0AANp2dwAAAA
> >                     Match ID   : UQkAAAAAAAAA
> >     +24-23-22-21-20-19------18-17-16-15-14-13-+  O: gnubg
> >   O | O  O  O  O       | O |                  |  0 points
> >   O | O  O  O  O       |   |                  |
> >   O | O                |   |                  |
> >   O |                  |   |                  |
> >   O |                  |   |                  |
> >     |                  |BAR|                  |v
> >     |                  |   |                  |
> >     |                  |   |                  |
> >     |          X  X  X |   |                  |
> >     | X  X  X  X  X  X |   |                  |  On roll
> >     | X  X  X  X  X  X |   |                  |  0 points
> >     +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: user (Cube: 2)
> >
> > Cube analysis
> > Rollout cubeless equity  +0.584
> >
> > Cubeful equities:
> > 1. Double, pass         +1.000
> > 2. Double, take         +1.061  ( +0.048)
> > 3. No double            +1.013  ( +0.000)
> > Proper cube action: Redouble, pass
> > Rollout details:
> > Player user owns 2-cube:
> >   0.792 0.000 0.000 - 0.208 0.000 0.000 CL  +0.584 CF  +1.013
> >  [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.001 CF  +0.003]
> > Player gnubg owns 4-cube:
> >   0.795 0.000 0.000 - 0.205 0.000 0.000 CL  +1.181 CF  +1.061
> >  [0.001 0.000 0.000 - 0.001 0.000 0.000 CL   0.002 CF  +0.003]
> > Full cubeful rollout with var.redn.
> > 1296 games, seed 1206724736 using quasi-random dice and 
> Mersenne Twister
> > Play and cube: 2-ply cubeful 100% speed [world class]