Actually this position comes out even more tricky when you analyze it
further in gnubg:
- gnubg cannot even find the best move on 4-ply in this very simple
position.
- std. 0-ply 36 games rollout gives negative MWC for one of the moves (as I
reported yesterday)
- when rolling out on 0-ply 216 games, gnubg yields some very strange
results when using variance reduced and cubeful rollout:
- the move with the most single game wins (2,6%) is the hopeless 6/3(3)
5/2.
- the best moves (MWC) all have one thing in common: they undo the
6-point!
- using a 0-ply 216 games rollout without extras, gnubg's results seem more
"correct"
I suspect that it is the variance reduction functionality in a position
that the neural net cannot figure out, which is causing these problems.
Jim: could you be more specific as to which problems you found and
corrected in the rollout code?
And where can I read more about how variance reduction works (I do not have
the time to read it out of the source code, if some more readable texts is
available I would appreciate a link).
Thank you for a great contribution to modern backgammon! I'm very much
looking forward to the multi-processor version, which will enable to bot to
dig even deeper in, and probably give us more insight in this tricky game.
Best regards,
Henrik Bukkjaer
PS: I'm using 0.14 in the w32 version compiled by Öystein a couple of weeks
ago - with the Ortega/Kleinmann ME table.
address@hidden@gnu.org on 20-07-2003 18:52:47
Sent by: address@hidden
To: address@hidden
cc: address@hidden
Subject: [Bug-gnubg] GNUBG gammon bug
Hi,
I analysed a match my nephew had played on FIBS with GNUBG 0.14 (build
Jul 2 2003), analysis setting was supremo for checkerplay.
In the situation shown below GNUBG critised Spock's move as a huge blunder,
but I can't see why; to the contrary, isn't GNUBG's evaluation extremely
faulty ?
Thanks for any explanations !
melBOT (O, 1 pts) vs. Spock (X, 0 pts) (Match to 5)
Game number 2
Move number 72: X to play 33
GNU Backgammon Position ID: FgAAwJ8dRAACAA
Match ID : QYmtABAAAAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: melBOT (Cube: 2)
| X | | O O X | OOO 1 point
| | | O | OOO
| | | | OO
| | | | OO
| | | | OO
v| |BAR| | 5 point match
| | | 7 |
| | | X |
| | | X X |
| | | X X X | Rolled 33
| X | | X X X | 0 points
+12-11-10--9--8--7-------6--5--4--3--2--1-+ X: Spock
Pip counts: O 7, X 98
* Spock moves 14/5 11/8
Alert: very bad move (-9,664%)
Rolled 33 (+0,074%):
1. Cubeful 2-ply 11/8 6/3(3) MWC: 14,85%
0,000 0,000 0,000 - 1,000 0,419 0,235
2. Cubeful 2-ply 14/11 6/3(3) MWC: 11,51% ( -3,35%)
0,001 0,000 0,000 - 0,999 0,550 0,358
* 3. Cubeful 2-ply 14/5 11/8 MWC: 5,19% (
-9,66%)
0,002 0,000 0,000 - 0,998 0,799 0,589
4. Cubeful 0-ply 14/5 6/3 MWC: 3,83% (-11,03%)
0,007 0,000 0,000 - 0,993 0,865 0,209
5. Cubeful 0-ply 11/5 6/3(2) MWC: 3,76% (-11,09%)
0,004 0,000 0,000 - 0,996 0,861 0,173
Output generated Fri Jul 18 18:01:33 2003
by GNU Backgammon 0.14-devel (Text Export version 1.39)
I didnt manage to copy the position directly from GNUBG into Wordpad
so that the position is shown right formatted, so I`ll post the appropriate
html-file and the pos-file as attachments, sorry for any inconvenience.
Kindest regards
Hans-Jürgen Schäfer
_______________________________________________
Bug-gnubg mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/bug-gnubg
_______________________________________________
Bug-gnubg mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/bug-gnubg