bug-gnubg
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] 0ply doubles early


From: Joseph Heled
Subject: Re: [Bug-gnubg] 0ply doubles early
Date: Sat, 16 Dec 2006 07:59:51 +1300

I understand your frustration with gnubg not handling the cube as well
as you think it should at those "simple" or "straightforward"
situations. Yet I do not think the situation is simple. race is
simpler than contact, still someone may make great improvements if she
is willing to do the proper research.

Personally I do not believe any of the numbers below except the
cubeless winning percentage. 2ply, 4ply and rollout cubfull numbers
are all based on a large number of 0ply cubefull decisions, and if
this is suspect, why would they be any good. 2ply play will not be any
better than 0ply if your 0ply is awful. If I was to start somewhere, I
would start with doubling on the very last stage of bearoff - where
you first get the true actions by brute force. This requires a large
database since you need the result for each score. In addition I am
not sure I agree with the doubling code in gnubg. I always used my own
code which is part of the fibs2html or gnubg-nn, which I think is
better (but I may be wrong). If someone want to take this code and
integrate it into gnubg, where one can choose which method to use
would be a great start as well.

-Joseph

On 12/16/06, Robert-Jan Veldhuizen <address@hidden> wrote:
Here's a very simple, unspectacular example:

    GNU Backgammon  Position ID: sO3uAACwu3sAAA
                     Match ID   : cAkAAAAAAAAA
    +24-23-22-21-20-19------18-17-16-15-14-13-+  O: Kit
Woolsey
     |             O  O |   | O  O  O  O       |  0 points
    |             O  O |   | O  O  O  O       |
    |                  |   |    O  O  O       |
     |                  |   |                  |
    |                  |   |                  |
     |                  |BAR|                  |v (Cube: 1)
    |                  |   |                  |
    |                  |   |       X          |
     |                X |   | X  X  X          |
    |             X  X |   | X  X  X          |  On roll
     |             X  X |   | X  X  X          |  0 points
    +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: GOL
Readers

Rollouts indicate it's not quite a double yet, with 71.21% GWC.

2-ply and 4-ply cube see 71.18 and 71.32% GWC and also say not quite a
double.

0-ply cube sees LESS winning chances, only 70.74%, yet it doubles (0.0243
difference even).

So 0-ply doubles because it overestimates volatility, whereas it actually
underestimates GWC here.

 Could it be that some parameters used for 0-ply cube decisions need a bit
more fine-tuning? This is a very simple long race position where you'd
expect volatility to be fairly predictable just by the pipcounts.

Here's the GNUBG output, note that I use a 1-sided DB to the ninepoint so
your output may differ slightly, but it seems irrelevant for this issue.

Cube analysis
0-ply cubeless equity  +0.4149
   70.74%   0.00%   0.00% -  29.26%   0.00%   0.00%
Cubeful equities:
1. Double, take         +0.6290
2. Double, pass         +1.0000  ( +0.3710)
3. No double            +0.6047  ( - 0.0243)
Proper cube action: Double, take

Cube analysis
2-ply cubeless equity  +0.4235
   71.18%   0.00%   0.00% -  28.82%   0.00%   0.00%
Cubeful equities:
1. No double            +0.6528
2. Double, pass         +1.0000  ( +0.3472)
3. Double, take         +0.6520  ( -0.0008)
Proper cube action: No double, take (0.2%)

Cube analysis
4-ply cubeless equity  +0.4265
   71.32%   0.00%   0.00% -  28.68%   0.00%   0.00%
Cubeful equities:
1. No double            +0.6654
2. Double, pass         +1.0000  ( +0.3346)
3. Double, take         +0.6637  ( -0.0017)
Proper cube action: No double, take (0.5%)
=======================================
 Cube analysis
Rollout cubeless equity  +0.4251

Cubeful equities:
1. No double            +0.6660
2. Double, pass         +1.0000  ( +0.3340)
3. Double, take         +0.6573  ( -0.0087)
Proper cube action: No double, take ( 2.5%)
Rollout details:
Centered 1-cube:
   71.26%   0.00%   0.00% -  28.74%   0.00%   0.00% CL  +0.4251 CF  +0.6660
 [  0.01%   0.00%   0.00% -   0.01%   0.00%   0.00% CL   0.0003 CF   0.0031]
Player Kit Woolsey owns 2-cube:
   71.21%   0.00%   0.00% -  28.79%   0.00%   0.00% CL  +0.8484 CF  +0.6573
 [  0.01%   0.00%   0.00% -   0.01%   0.00%   0.00% CL   0.0006 CF   0.0031]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 1017692672 and quasi-random
dice
Play: 0-ply cubeful prune [expert]
Cube: 0-ply cubeful prune [expert]

Cube analysis
Rollout cubeless equity  +0.4252

Cubeful equities:
1. No double            +0.6685
2. Double, pass         +1.0000  ( +0.3315)
3. Double, take         +0.6527  ( -0.0158)
Proper cube action: No double, take (4.6%)
Rollout details:
Centered 1-cube:
   71.26%   0.00%   0.00% -  28.74%   0.00%   0.00% CL  +0.4252 CF  +0.6685
 [  0.01%    0.00%   0.00% -   0.01%   0.00%   0.00% CL   0.0003 CF
0.0031]
Player Kit Woolsey owns 2-cube:
   71.21%   0.00%   0.00% -  28.79%   0.00%   0.00% CL  +0.8485 CF  +0.6527
 [  0.01%   0.00%   0.00% -   0.01%   0.00%   0.00% CL   0.0006 CF   0.0034]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 1017692672 and quasi-random
dice
Play: 0-ply cubeful prune [expert]
Cube: 2-ply cubeful prune [world class]

--
Robert-Jan Veldhuizen
_______________________________________________
Bug-gnubg mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/bug-gnubg







reply via email to

[Prev in Thread] Current Thread [Next in Thread]