Re: [Bug-gnubg] Gnu 2ply seems too optimistic

[Albert Silver]

There are other areas where it is much much worse. For example
evaluating cube situations in containment games, I have positions
(several), where the 2-ply might say No Double/take, and a rollout
will say Too Good to Double/Pass.

GNU Backgammon  Position ID: 73sAAAyYM7RCBQ
                Match ID   : UQkAAHAAGAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+     O: V72222
| X  X  X          |   | X  X  X  O  O  O |     7 points
| X                |   |          O  O  O |
|                  |   |          O  O  O |
|                  |   |          O  O  O |
|                  |   |             O    |
v|                  |BAR|                  |
|                  |   |                  |
|                  |   |                  |
|                  |   | X                |
|             X    |   | X     X  O       |     On roll
 | X           X    |   | X     X  O       |     3 points
+12-11-10--9--8--7-------6--5--4--3--2--1-+     X: AlmostDecent (Cube: 2)


Cube analysis
2-ply cubeless equity  +0.4984
  71.86%  17.30%   0.09% -  28.14%  11.22%   0.05%
Cubeful equities:
1. No double            +0.8127
2. Double, pass         +1.0000  ( +0.1873)
3. Double, take         +0.8073  ( -0.0054)
Proper cube action: No redouble, take (2.8%)


Cube analysis
Rollout cubeless equity  +0.6281

Cubeful equities:
1. No double            +1.1187
2. Double, take         +1.2742  ( +0.1555)
3. Double, pass         +1.0000  ( -0.1187)
Proper cube action: Too good to redouble, pass (43.3%)
Rollout details:
Player AlmostDecent owns 2-cube:
  76.11%  20.29%   0.37% -  23.89%  10.02%   0.05% CL  +0.6281 CF  +1.1187
[  0.16%   0.24%   0.05% -   0.16%   0.13%   0.01% CL   0.0052 CF   0.0124]
Player V72222 owns 4-cube:
  77.65%  22.01%   0.64% -  22.35%   9.88%   0.05% CL  +1.3608 CF  +1.2742
[  0.17%   0.30%   0.06% -   0.17%   0.15%   0.02% CL   0.0111 CF   0.0154]
Full cubeful rollout with var.redn.
1298 games, Mersenne Twister dice gen. with seed 1038057470 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]


Albert


On 11/8/06,  address@hidden
<address@hidden> wrote:
> Mochy wrote:
>
> >Recently I found 3 positions where Gnu 2ply evaluate the
>  >equity too optimistic.After the rollout or in Snowie evaluation (and
> >Rollout) , Those are actually huge pass.
> >Not only 2ply, 4ply also make similar mistakes.
> >
> >I dont know if it is because of new build or not.
> >Any comments are welcome.
> >
> >Here are 3 positions (Position ID / Match ID)
> >ZtvAAWBjN4MIDA / cAmgAQAAAAAA
> >bFsIOEhxOwcDIA / MAGgAVAAMAAA
> >xs7BAURwb4OFAA / MAHgADAAAAAA
> >
> >Mochizuki Masayuki (mochy)
> >address@hidden
>
> Hi Mochy,
>
> as stated by others, a bug has been introduced in some important evaluation
> code
> affecting evaluation of positions with a player having 2 men on the bar.
>
> I checked your poistions with the latest available source code and they all
> are
> pass at 2-ply world-class.
>
> I'm currently unable to put a new Win distrib (or just the new executables)
> on
> the  www.gnubg.org site : due to a local network problem, I'm unable to do
> anye
> external FTP.
>
> I can mail you the new executables (1.2Mb) so you can check they behave "as
> expected". Let me know if you're interested.
>
> MaX.
>
>
>
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
>  http://lists.gnu.org/mailman/listinfo/bug-gnubg