Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Ferret vs Gerbil

Author: Uri Blass

Date: 17:29:21 03/27/02

Go up one level in this thread


On March 27, 2002 at 19:24:37, Bas Hamstra wrote:

>On March 26, 2002 at 08:42:50, Uri Blass wrote:
>
>>On March 26, 2002 at 08:37:28, Uri Blass wrote:
>>
>>>On March 25, 2002 at 19:49:40, Tom Kerrigan wrote:
>>>
>>>>On March 25, 2002 at 12:50:57, Uri Blass wrote:
>>>>
>>>>>Gerbil is clearly more complicated than tscp and gerbil
>>>>>is also clearly better than tscp.
>>>>>Gerbil is using hash tables and ponder when tscp does not.
>>>>
>>>>Searching faster/deeper via hash tables and pondering does not necessarily make
>>>>it better. Last I checked, Gerbil relied entirely on piece/square tables. It
>>>>wouldn't surprise me if TSCP does enough evaluation to beat piece/square table
>>>>programs (unless it's getting ridiculously outsearched).
>>>>
>>>>-Tom
>>>
>>>Piece square table with better search rules and hash tables
>>>and pondering is enough to beat tscp.
>>>
>>>The truth is that I have a piece square table program that
>>>is better than tscp without hash tables, pondering or null move
>>>
>>>It is using futility pruning and better time management and
>>>better order of moves than tscp and few extensions that
>>>tscp does not use but I have no doubt that
>>>the search rules can be improved significantly and not only
>>>by null move pruning and more pruning rules but by better extensions).
>>>
>>>
>>>My latest program has some more knowledge in the evaluation
>>>but based on tests that I did it is probably
>>>only slightly better than the piece square table version.
>>>
>>>I believe that tscp at depth x is weaker then modified tscp
>>>at depth x+2 when the modified version has only
>>>piece square table evaluation and x is not important.
>>>
>>>Maybe someone can test both versions against different programs
>>>in order to find out.
>>>
>>>Uri
>>
>>I can add that claiming that gerbil is clearly better than tscp was based on
>>results of the tournament of Leo
>>see the history pages of
>>http://home.hccnet.nl/leo.dijksman/index.html
>>
>>Gerbil scored 21.5/28 when tscp scored only 13/28
>>
>>Uri
>
>Don't forget blitz/rapid  is different from real games. You say X is not
>important, but it is...

My piece square table program could beat tscp more easily at long time control
thanks to better branching factor.

I believe that the difference in depth is not more than 2 plies even at long
time control inspite of the fact that I cannot do fair comparison because of the
fact that my program also does more extensions than tscp and I also believe that
I use better piece square table than tscp's piece square table.

I did not say that x+2 with piece square table evaluation always beat every
evaluation function with depth x for every x that you can practically use but I
said it only about tscp.

Maybe things become different when x=10 but tscp usually does not get depth 10
in the middle game even at 120/40 time control.

When x<10 I believe that depth x+2 for piece square table that is the same as
tscp is better than depth x for tscp full evaluation.

Maybe somebody can volunteer to check it(it is easy to modify tscp to piece
square table program)

Uri



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.