GnuBG rollout majorly disagrees 0-4 ply (am I doing it wrong?!)

[appleorange...@gmail.com]

GNU Backgammon Position ID: 7m8AANDWNgFQCA
Match ID : ARngAAAAAAAE
+12-11-10--9--8--7-------6--5--4--3--2--1-+ O: KobZero
| O | | O O O O O O | 0 points
| | | O O O O O |
| | | |
| | | |
| | | |
^| |BAR| |
| | | 7 |
| | | X |
| | | X X |
| | | X X X | X Cube offered at 4
| | X | O O X X X O | X 0 points
+13-14-15-16-17-18------19-20-21-22-23-24-+ X: You

I dropped this double, and GnuBG said it was a VERY BAD decision. It says that dropping is:

0 ply says -0.363
1 ply says -0.167
2 ply says -0.292
3 ply says -0.090
4 ply says -0.212

I still didn't believe it was that bad, so I pressed the rollout button. I've never done a rollout, So I didn't change any settings and I don't understand what all the statistics represent. When it finished and I pressed ok, the analysis now says that it's a pass and that TAKING is a -0.069 error.

First of all... Am I correct in thinking the rollout says this is a pass?

Second... If my interpretation of the rollout is correct, should I really believe Gnubg analysis if even 4 ply analysis is this far off from the rollout's?

Third...is there a guide to interpreting the rollout statistics. It's not explained in the Gammonlife article at http://www.gammonlife.com/gnu/index.htm

Thanks

[appleorange...@gmail.com]

By the way, it was a 7 point match.

[Walt]

On 5/16/2013 11:23 PM, appleorange...@gmail.com wrote:
> GNU Backgammon Position ID: 7m8AANDWNgFQCA
> Match ID : ARngAAAAAAAE
> +12-11-10--9--8--7-------6--5--4--3--2--1-+ O: KobZero
> | O | | O O O O O O | 0 points
> | | | O O O O O |
> | | | |
> | | | |
> | | | |
> ^| |BAR| |
> | | | 7 |
> | | | X |
> | | | X X |
> | | | X X X | X Cube offered at 4
> | | X | O O X X X O | X 0 points
> +13-14-15-16-17-18------19-20-21-22-23-24-+ X: You
>
> I dropped this double, and GnuBG said it was a VERY BAD decision. It says that dropping is:
>
> 0 ply says -0.363
> 1 ply says -0.167
> 2 ply says -0.292
> 3 ply says -0.090
> 4 ply says -0.212
>
> I still didn't believe it was that bad, so I pressed the rollout button. I've never done a rollout, So I didn't change any settings and I don't understand what all the statistics represent. When it finished and I pressed ok, the analysis now says that it's a pass and that TAKING is a -0.069 error.
>
> First of all... Am I correct in thinking the rollout says this is a pass?

I get similar results, for instance:


/////////////2-ply analysis says take//////////////
Cube analysis
2-ply cubeless equity +0.395 (Money: +0.400)
0.737 0.000 0.000 - 0.263 0.072 0.002
Cubeful equities:
1. Double, take +0.707
2. Double, pass +1.000 ( +0.293)
3. No double +0.670 ( -0.037)
Proper cube action: Redouble, take


///////////////rollout says pass //////////////////////////

Rollout cubeless equity +0.515 (Money: +0.507)

Cubeful equities:
1. Double, pass +1.000
2. Double, take +1.161 ( +0.161)
3. No double +0.885 ( -0.115)
Proper cube action: Redouble, pass

Rollout details:
Player O owns 2-cube:
0.795 0.000 0.000 - 0.205 0.078 0.005 CL +0.515 CF +0.885
[0.002 0.000 0.000 - 0.002 0.002 0.001 CL 0.003 CF 0.008]
Player X owns 4-cube:
0.826 0.000 0.000 - 0.174 0.078 0.010 CL +1.231 CF +1.161
[0.001 0.000 0.000 - 0.001 0.002 0.002 CL 0.006 CF 0.007]
Full cubeful rollout with var.redn.
1296 games, Mersenne Twister dice gen. with seed 682780187 and
quasi-random dice
Play: world class 2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0.16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]

//////////////////////////////////////////////////////////////////////

This is one of those positions where gnu analysis gets it wrong. They
are somewhat rare, but they happen. This is a rather extreme example,
with .45 equity difference between analysis and rollout. To answer your
question, yes this is a pass. gnu analysis gets it wrong.


This is why I've switched from gnubg to XG. XG tends to get these right
even at low analysis settings, for instance:


Analyzed in XG Roller+
Player Winning Chances: 80.29% (G:0.00% B:0.00%)
Opponent Winning Chances: 19.71% (G:7.16% B:0.28%)

Cubeless Equities: No Double=+0.526, Double=+1.161

Cubeful Equities:
No redouble: +0.770 (-0.230)
Redouble/Take: +1.094 (+0.094)
Redouble/Pass: +1.000

Best Cube action: Redouble / Pass

eXtreme Gammon Version: 2.10, MET: Kazaross XG2

>
> Second... If my interpretation of the rollout is correct, should I really believe Gnubg analysis if even 4 ply analysis is this far off from the rollout's

Most of the time, yes. gnu analysis is almost always pretty good, but
you have found one of the rare situations where it is not.

> Third...is there a guide to interpreting the rollout statistics. It's not explained in the Gammonlife article at http://www.gammonlife.com/gnu/index.htm

Zorba wrote a really nice intro to gnu at
http://www.gammonlife.com/gnu/gnu1.htm

The gnubg manual is here:
http://www.gnubg.org/documentation/doku.php?id=rollouts


Unfortunately, nothing to explain the results. Here's a quick tour:


Cubeful equities:
1. Double, pass +1.000
2. Double, take +1.161 ( +0.161)
3. No double +0.885 ( -0.115)
Proper cube action: Redouble, pass

This bit tells you the equity for each action. D/P of course gives O
1.0 equity since the game is over. D/T gives X 1.161 equity, meaning O
will win about 1.16 points for each game played. ND means O has .885
equity. Since O increases her equity from .885 by cubing, it's a
double. Then X has a choice of letting O have 1.0 equity or 1.16 equity
- the better choice for X is the smaller number, so it's a pass.


the next rows of numbers show winning percentages

0.795 0.000 0.000 - 0.205 0.078 0.005 CL +0.515 CF +0.885

Reading left to right, this says O wins 79.5% of games, 0% gammons 0%
backgammons - X wins 20.5% games. 0% gammons and O% backgammons

The last two numbers give the cubeless and the cubeful equity. if O
holds the cube at 2, her cubeless eqity is .515 and her cubeful equity
is .885

The row below it in [square brackets ] is the tolerance for the number
above - here it's .002, so interpret that as .795 +/- .002

Note that the rollout gives these percentages for both O holding the
cube and with X holding the cube at 4.

It's then followed by a bunch of details about the rollout settings.

Here's what an XG rollout looks like - the same statistics, just
presented a bit differently.

Analyzed in Rollout
No redouble
Player Winning Chances: 80.67% (G:0.00% B:0.00%)
Opponent Winning Chances: 19.33% (G:6.99% B:0.23%)
Redouble/Take
Player Winning Chances: 82.85% (G:0.00% B:0.00%)
Opponent Winning Chances: 17.15% (G:7.28% B:0.81%)

Cubeless Equities: No Double=+0.536, Double=+1.225

Cubeful Equities:
No redouble: +0.799 (-0.201)
Redouble/Take: +1.172 (+0.172)
Redouble/Pass: +1.000

Best Cube action: Redouble / Pass

Rollout:
1296 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG Roller
Confidence No Double: � 0.004 (+0.795..+0.803)
Confidence Double: � 0.009 (+1.163..+1.182)

Double Decision confidence: 100.0%
Take Decision confidence: 100.0%

Duration: 2 minutes 28 seconds

eXtreme Gammon Version: 2.10, MET: Kazaross XG2


--
//Walt