[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Bug-gnubg] bots roll out reliability
From: |
max-d |
Subject: |
[Bug-gnubg] bots roll out reliability |
Date: |
Tue, 30 Jul 2002 11:33:57 +0200 |
gnubg (O, 0 pts) vs. user (X, 1 pts) (Match to 3)
Move number 22: X on roll, cube decision?
GNU Backgammon Position ID: 2LYNAHDsdkYCCA
Match ID : UQlgAAAACAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: gnubg
| X O O O | O | O O O X | 0 points
| O O O | O | O O O |
| | O | |
| | | |
| | | |
v| |BAR| | 3 point match
| | | |
| | | |
| | | X X |
| X | | X X X X | On roll
| X X | | X X X X | 1 point
+12-11-10--9--8--7-------6--5--4--3--2--1-+ X: user (Cube: 2)
* user moves 11/9 6/3
Alert: bad move (-4,766%)
Rolled 23 (+3,685%):
1. Cubeful 2-ply 13/11 8/5 MWC: 54,39%
0,372 0,199 0,009 - 0,628 0,054 0,002
2. Cubeful 2-ply 13/11 4/1 MWC: 53,49% ( -0,90%)
0,373 0,190 0,008 - 0,627 0,086 0,003
3. Cubeful 2-ply 8/5 4/2 MWC: 53,15% ( -1,25%)
0,364 0,181 0,007 - 0,636 0,078 0,003
4. Cubeful 2-ply 13/11 6/3 MWC: 52,85% ( -1,54%)
0,351 0,179 0,008 - 0,649 0,057 0,002
5. Cubeful 2-ply 6/3 4/2 MWC: 52,50% ( -1,89%)
0,355 0,170 0,007 - 0,645 0,079 0,003
* 13. Cubeful 2-ply 11/9 6/3 MWC: 49,63% ( -4,77%)
0,306 0,148 0,006 - 0,694 0,060 0,002
Output generated Tue Jul 30 10:24:15 2002
by GNU Backgammon 0.12 (Text Export version 1.11)
in this position Gnubg 2plies and JF lv 7 do not agree at all !
I wanted to verify whether each bot still gave different results with
the same roll out settings .
if so ,bots's roll out reliability would be likely unreliable
(at least for one of them)
Gnubg (wc++) does not even consider Jf level 7 best move
4/1 11/9
For this move ,Jf gives eq (X) -0.370
For GNubg move (13/11 8/5) JF gives eq (X) =-0.480
I rolled the positions out 720 times (JF level 6 rollouts)
after Gnubg best move
W G/BG BG
O wins 75.5 13.1 0.4
x wins 24.5 9.7 1.2
eq(o)=0.537 sd 0.015
720 games equivalent to 6563
after JF best move
W G/BG BG
O wins 73.6 14 0.4
x wins 26.4 8.9 0.6
eq(o)=0.52 sd 0.015
720 games equivalent to 5724
seems JF evaluation is wrong since the roll out does not confirm the #eq.
I have problems to understand how to use Gnubg's roll out !
what are GNUbg's equivalent settings for jf level 6 roll out ?
what about GNUbg rolling this position out ?
Is there any interest ,for Gnubg team developpers ,
to report here such positions (main bots disagree by a lot )?
Thanks !
md.
- [Bug-gnubg] bots roll out reliability,
max-d <=