Author: Tim Foden
Date: 00:34:14 09/24/03
Go up one level in this thread
On September 24, 2003 at 01:58:35, Johan de Koning wrote:
>On September 23, 2003 at 03:42:18, Tim Foden wrote:
>
>>On September 23, 2003 at 03:03:12, Johan de Koning wrote:
>>
>>>On September 22, 2003 at 02:32:34, Tim Foden wrote:
>>>
>>>>I think these positions might behave like smaller versions of the full thing...
>>>>
>>>>[d] 8/1pppppp1/8/8/8/8/1PPPPPP1/8 w - -
>>>>
>>>>GLC thinks this is a win... d4 Mate in 15.
>>>
>>>Hmmm..., my program (named JJ.EXE :-) proved it a draw.
>>>Then I added en passent.
>>>Now it says it's a mate in 14.
>>>
>>>098.81 75M346 D21 +4 : d4,d5 f3,b6 e4,dxe4 fxe4,b5 b4,c6 g3,f6 g4,g6 e ...
>>>184.33 140M189 D23 +212 : d4,d5 f3,g6 e4,dxe4 fxe4,f5 exf5,gxf5 b4,e6 b5, ...
>>>279.30 212M443 D25 +9973 : d4,d5 f3,b6 e4,dxe4 fxe4,b5 c3,e6 e5,c6 b4,f6 e ...
>>>377.94 286M689 D27 +9973 : d4,d5 f3,b6 e4,dxe4 fxe4,b5 c3,c6 b3,g6 c4,bxc4 \
>>> bxc4,f6 c5,e6 d5,exd5 exd5,cxd5
>>>0000011111 1111222222 22222222:: :::::::::: :::::::::: :::::::::: ::::::::::
>>>0357912456 7899011222 23222210:: :::::::::: :::::::::: :::::::::: ::::::::::
>>>
>>>So by adding e.p. I removed the 1 known bug, but we still don't agree. :-(
>>>Unfortunately we don't know the number and nature of the unknown bugs. :-)
>>>
>>>.. Johan
>>
>>Well, we can probably track the bugs down...
>>
>>After d4 d5 GLC agrees that f3 is a mate in 13...
>>
>>[d]8/1pp1ppp1/8/3p4/3P4/8/1PP1PPP1/8 w
>>
>> 22 6.229 +Mate13 6249564 f3 g5 2. e4 e5 3. dxe5 d4 4. b4 c6 5. g3 b6 6. f4
>> gxf4 7. gxf4 c5 8. f5 d3 9. cxd3 cxb4 10. e6
>> fxe6 11. fxe6 b3
>>
>>GLC prefers g5 after f3 rather than b6.
>>
>>But what does yours say after d4 g5?
>>
>>[d]8/1ppppp2/8/6p1/3P4/8/1PP1PPP1/8 w
>>
>>GLC says:
>>
>> 22 22.202 +Mate14 21765k c4 f5 2. g3 f4 3. e4 fxe3 4. fxe3 e6 5. g4 d6 6.
>> d5 exd5 7. cxd5 c6 8. dxc6 bxc6 9. e4 c5 10. b3
>> d5 11. exd5 {ht}
>
>002.86 3633954 D19 +102 : c4,f5 g3,f4 g4,b6 b4,c6 e3,f3 c5,bxc5 bxc5,d5 ...
>003.46 4415449 D19 +103 : e4,b5 c3,g4 b3,d6 f4,e5 fxe5,dxe5 d5,g3 c4,b4 ...
>004.29 5417281 D21 +105 : e4,b5 c3,c6 b3,g4 c4,bxc4 bxc4,d6 d5,cxd5 exd ...
>005.00 6327052 D21 +9977 : c4,f5 g3,f4 g4,b6 b4,c6 e3,f3 e4,d5 exd5,cxd5 ...
>005.33 6766094 D23 +9977 : c4,f5 g3,f4 g4,b6 b4,c6 e3,f3 e4,d5 exd5,cxd5 \
> c5,bxc5 bxc5,e6 c6,e5
This is the position after 6. e4 above.
[d]8/3pp3/1pp5/6p1/1PPPP1P1/5p2/5P2/8 b - - 0 7
Here GLC thinks that 6... d5 is a blunder. It thinks that 6... b5 draws.
5 0.010 +0.000 515 b5 2. c5 e6 3. e5 d5
After 6... d5:
8 0.000 +Mate06 772 exd5 cxd5 2. c5 bxc5 3. dxc5 e5 4. c6 d4 5. c7 e4
6. c8
>
>Evaluation didn't exactly help here. :-)
>Anyway, that's Mate12, so it considers 1...g5 worse then 1...d5.
>
>>or for a 1 move longer PV...
>>
>> 25 25.336 +Mate14 25311k c4 f5 2. g3 f4 3. e4 fxe3 4. fxe3 e6 5. g4 d6 6.
>> d5 exd5 7. cxd5 c6 8. dxc6 bxc6 9. e4 c5 10. b3
>> d5 11. exd5 c4
>
> . . . . . . . . This is the end of your PV.
> . . . . . . . . O-to-move has a runner at d5, which wins in 5 ply.
> . . . . . . . . So that adds up to Mate14 all right.
> . . . o . . x . BUT ...
> . . x . . . o . That Mate14 is a heurstic lower bound, not a perfect score.
> . o . . . . . . The true score is of course bxc4#.
> . . . . . . . . In other words, you can use heurstic bounds to prune against
> . . . . . . . . beta (or alpha), but not inside the window.
Yes, I was only using it as a limit... but I'd got the logic a little wrong.
I've corrected it now. It makes the program a little slower though. :)
>
>
>
>BTW, my program is much faster than yesterday. Mostly because of using the
>TT move for sorting. But I think I've implemented all the standard tricks that
>come with AlphaBeta, and I'm running out of non-standard ideas as well.
Looks a lot faster than GLC now. :)
>I'm sure the code (and data!) can easily be optimized to run 2x or 3x faster,
Yes. I feel the same about GLC's code... it was just a real hack and slash job.
There's still lots of stuff I could remove and/or change which would probably
make it much faster.
>but that's rather pointless as long as there are no tournamtents to play.
>It would still be fun though. :-)
Well, glc300-p.exe is still a winboard engine that I've just hacked. It SHOULD
still be able to play a game. Would you like a copy of the exe?
>But for today I'll let it burn some CPU on the d4d5 and 7-7 positions.
>
>... Johan
I tried the full (8-8) start position the night before last, but it still had no
result after 9 hours, so I stopped it.
On the 7-7 position, I decided that the lack of left/right symmetry was hurting,
so I stopped that too. :)
I thought of trying the 8-8 position after d4 d5, but I forgot to leave it
running last night. Maybe I'll try it tonight.
Cheers, Tim.
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.