Author: Christophe Theron
Date: 21:22:31 05/26/03
Go up one level in this thread
On May 26, 2003 at 02:27:16, Jonas Bylund wrote:
>On May 26, 2003 at 01:34:10, Christophe Theron wrote:
>
>>On May 25, 2003 at 19:52:14, Jonas Bylund wrote:
>>
>>>On May 25, 2003 at 17:34:13, Christophe Theron wrote:
>>>
>>>>On May 25, 2003 at 13:33:14, Andre van Ark wrote:
>>>>
>>>>>Conditions: Time = 90 min blitz.
>>>>>Ponder = on.
>>>>>Hastables: King = 32 MB
>>>>>Tiger 15.0 = 128 MB
>>>>>Tablebases = NO
>>>>>GUI = Hiarcs 8
>>>>>PC = Dual Athlon 1400, 1000 MB ram.
>>>>>Book = None (Nunn 2 database)
>>>>>
>>>>>PGN comes when the match has finished,
>>>>>
>>>>>Ciao,
>>>>>Andre
>>>>
>>>>
>>>>
>>>>Wouldn't it be better to give The King the same amount of hash tables? Each
>>>>program should be given the same computing resources. I'm sure that more hash
>>>>table for The King will not hurt it.
>>>>
>>>>
>>>>
>>>> Christophe (author of Chess Tiger)
>>>
>>>Actually tests indicate (both my own, Kurt's and others) that 32 Mb is the
>>>optimal amount of hash for the king, i guess that is why he chose to use that
>>>amount as well.
>>>
>>>Regards
>>>Jonas
>>
>>
>>
>>If you want my opinion it doesn't make any sense. Or I need a good explanation,
>>and some evidence.
>>
>>
>>
>> Christophe
>
>I did say "indicate" i have no proof of this, i only tried to explain why he
>chose 32 Mb for The King. If memory serves me right The King clears hash between
>moves and therefore dosn't need as much ram as other engines.
>
>Evidence... computerchess leaves more questions than answers, if i played 10.000
>games with 32 Mb for the king against various opponents and repeated the 10.000
>games with 64 Mb hash and the 32 Mb setup had the better score, some people
>would still want evidence :)
>
>I, like Thorsten am a big believer in watching the games you test, in order to
>get a "feeling" for strenght's and weaknesses instead of just looking at the
>results, both have to be combined IMO. Therefore i can't give you any evidence,
>but just my opinion along with other peoples equal observations regarding
>optimal hash settings for the king.
>
>Jonas
You are right that it is going to be extremely difficult to give evidence by
playing games and comparing the results.
My comment comes from the fact that from a mathematical or algorithmic point of
view, assuming the hash table management is not pathological in The King, and
even if there is a short delay to clear them before the search starts, using
bigger hash tables should not hurt, and in a few cases really help (in the
endgame maybe).
Even if there is no significant increase in playing strength or no significant
shorter solution times for test suites, I think it would be fair to simply give
each program the same amount of memory.
Unless it can be shown by some mean that giving The King more hash table will
actually hurt him. But I don't think it is the case.
Christophe
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.