[Bug-gnubg] Multi-bot rollouts of backgame position

[Albert Silver]

Hi,

        The following was posted by Michael Depreli in the GoL forum. It
is the same position I asked about earlier however it comes with massive
rollouts by all the top known bots (MANY thanks Michael!), including
both the older and newer versions of GNU. I hope the text formatting
comes out ok.

        I have to agree with him in remarking how amazingly precise
GNU's 0-ply rollouts have shown themselves to be. 

                                                Albert


GNU Backgammon  Position ID: vHsXAADCNhQCNw
                 Match ID   : MAEAAAAAAAAA
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     O: White
 | O           O  O |   | O           O    |     0 points
 |             O  O |   | O                |     On roll
 |                  |   |                  |     
 |                  |   |                  |     
 |                  |   |                  |    
^|                  |BAR|                  |     (Cube: 1)
 |                  |   |                  |    
 |                  |   |    X     X       |     
 |                  |   | X  X  X  X  O    |     
 |                  |   | X  X  X  X  O  O |     
 | O           O  X |   | X  X  X  X  O  O |     0 points
 +13-14-15-16-17-18------19-20-21-22-23-24-+     X: Blue
Pip counts: O 203, X 69


Re: Backgame Bot Benchmark Position
Posted By: Michael Depreli <address@hidden>
Date: Saturday, 8 March 2003, at 5:47 a.m. 


Here are the results of the rollouts.
For the sake of uniformity I've listed Snowie's ply as 1 ply less that
actually used in the rollout.
I used SW's strongest settings and set GNU to the same filters. 


------------------------------------------------------------------------
--------


Is nuGNU 0-ply really that strong? Or is it losing equity defending
against the backgame?
GNU & nuGNU 1-ply seems to have gone haywire.
Although SW4' seems to evaluate back games very well I expected the
difference in playing strength to be be more noticeable over SW3.2.
nuGNU's 2-ply evaluation is truly shocking!



  
        Trials Ply      Eval       R/O  W       Wg      Wbg     L
Lg      Lbg 

JF      32000     0     -0.640   -0.672  40.2   0.9     0.0     59.8
35.0    13.6 

GNU     32000   0       -0.664   -0.582  43.5   0.9     0.0     56.5
34.6    11.5 

nuGNU 32000   0         -0.835   -0.442  48.7   0.9     0.0     51.3
31.5    11.0 

SW3.2 32000   0         -0.738   -0.669  40.4   1.0     0.0     59.6
35.7    13.0 

SW4     32000   0       -0.425   -0.636  42.1   1.0     0.0     57.9
35.7    13.1 


JF      1296vr  1       -0.732   -0.607  42.4   1.1     0.0     57.6
33.6    12.9 

GNU     1296vr  1       -0.682   -0.878  35.6   0.6     0.0     64.3
41.7    18.0 

nuGNU 1296vr  1         -0.477   -0.797  37.9   0.9     0.0     62.1
40.8    15.6 

SW3.2   1296vr  1       -0.773   -0.545  44.1   1.6     0.1     55.9
34.0    10.3 

SW4     1296vr  1       -0.475   -0.552  44.8   1.1   0.0       55.2
33.6    12.3 


GNU     1296vr  2       -0.705   -0.497  46.5   0.8     0.0     53.5
32.0    11.5 

nuGNU 1296vr  2         -1.056   -0.469  47.4   1.3     0.0     52.6
31.8    11.2 

SW3.2 1296vr  2         -0.784   -0.498  45.7   1.4     0.1     54.3
32.7    9.9 

SW4     1296vr  2       -0.487   -0.496  46.4   1.1     0.0     53.6
32.4    11.2