Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Node count explosion

Author: Vasik Rajlich

Date: 08:45:45 09/04/05

Go up one level in this thread


On September 04, 2005 at 05:46:03, Mridul Muralidharan wrote:

>On September 04, 2005 at 04:42:11, Vasik Rajlich wrote:
>
>>On September 03, 2005 at 15:41:48, Mridul Muralidharan wrote:
>>
>>>[D]r2q3r/1b1k1pbp/p4np1/2BP1pN1/p1B5/P1Q5/1PP3PP/R3K2R w KQ - 0 19
>>>
>>>My engine royally blows up when it comes to this position (arrived at a game
>>>against some engine a long while ago).
>>>I suspected excessive extensions , qsearch problems , eval inconsistencies ,
>>>bugs , etc.
>>>Could not find any bugs and there were no obvious others problems ... the
>>>position just kills the node count.
>>>Analysed it with other engines and most if not all had similar problems.
>>>
>>>I have not done an actual position analysis of the position with a strong player
>>>(that is tactics , positional aspects , etc - just analysed the stats that my
>>>engine dumps for this pos).
>>>
>>>Any comments ?
>>>
>>>Thanks,
>>>Mridul
>>
>>This reminds me of a funny position I saw the other day:
>>
>>[D] r1r3k1/1bq1bppp/p1npp1P1/2nB1N1Q/1p2P3/2N1B3/PPP2P1P/2KR2R1 b - -
>>
>>I was doing a q-search test which assumed that no q-search would go over 10k
>>nodes. This one crashed it :-)
>>
>
>Tough position this one too , thanks !
>I want to address positions like this upfront so that there is a lower bound to
>the worst case behaviour
>My idea is to play around and research about move ordering , extensions and
>qsearch/eval .. overall program strength/weakness is just a sideeffect as of now
>:)

Actually, this position is more for amusement. If every position was like this,
q-search would need to work differently - but it's not ..

>
>I did not understand "I was doing a q-search test which assumed that no q-search
>would go over 10k nodes" - what exactly did you mean ?
>

Just a funny debugging story. (Unfortunately I have a bit more of those than I'd
like.) I set up a test which just calls the q-search. There was a hard-coded
constant so that it could only handle a q-search of <10,000 positions. Seems
safe enough, right? :-)

Vas

>- Mridul
>
>>Here is the ridiculous source game:
>>
>>[Event "Engine Games 003 Min + X Sec"]
>>[Site "Internet"]
>>[Date "2005.01.17"]
>>[Round "?"]
>>[White "Deep Fritz 8"]
>>[Black "Fritz 8"]
>>[Result "1-0"]
>>[ECO "B89"]
>>[WhiteElo "2578"]
>>[BlackElo "2494"]
>>[PlyCount "89"]
>>[EventDate "2005.??.??"]
>>1. e4 c5 2. Nf3 e6 3. Nc3 d6 4. d4 cxd4 5. Nxd4 Nc6 6. Be3 Nf6 7. Bc4 a6 8. Qe2
>>Be7 9. Bb3 Qc7 10. O-O-O O-O 11. Rhg1 Nd7 12. g4 Nc5 13. Nf5 b5 14. Bd5 Bb7 15.
>>g5 Rfc8 16. Qh5 b4 17. g6 hxg6 18. Rxg6 fxg6 19. Qxg6 Bf6 20. Bxc5 Qd7 21. Rg1
>>Rf8 22. Bxd6 Rae8 23. Na4 exd5 24. Nc5 Qxf5 25. exf5 Ne7 26. Qh5 Bc6 27. Ne6
>>Rf7 28. Bxe7 Rfxe7 29. Qh6 Rf8 30. h4 Bb7 31. Qg6 Rff7 32. Rg5 Rxe6 33. fxe6
>>Bxg5+ 34. Qxg5 Rf6 35. Qe5 Bc6 36. Qb8+ Kh7 37. e7 Rxf2 38. e8=Q Bxe8 39. Qxe8
>>Rf1+ 40. Kd2 Rf5 41. Qe6 Rh5 42. Qg4 Re5 43. Qxb4 Re4 44. Qb6 Ra4 45. h5 1-0
>>
>>Vas



This page took 0 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.