Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Some explanations about SmarThink

Author: Drexel,Michael

Date: 07:17:53 06/06/03

Go up one level in this thread


On June 06, 2003 at 09:41:51, Uri Blass wrote:

>On June 06, 2003 at 08:58:17, Drexel,Michael wrote:
>
>>On June 06, 2003 at 07:54:07, Uri Blass wrote:
>>
>>>On June 06, 2003 at 07:06:41, Drexel,Michael wrote:
>>>
>>>>On June 06, 2003 at 06:15:36, Ryan B. wrote:
>>>>
>>>>>Making a strong opening book may not be so high on his to-do list at this time.
>>>>>I don?t see how where he is from has anything to do with his opening book.  The
>>>>>best of the free programs not named rebel 12 I have seen just generate an
>>>>>opening book from pgn files such as gm2600.pgn  This way of making an opening
>>>>>book requires no actual chess skill and defiantly in no way discriminates based
>>>>>on ones home land.
>>>>
>>>>And such books are definitely weak. The Ruffian book is relatively weak for
>>>>example.
>>>
>>>Ruffian book seems to be very good based on Ruffian results in Leo tournament.
>>>
>>>Ruffian got 79% against programs that include 2 proffesional programs(Gandalf
>>>and Nimzo)
>>>
>>>Ruffian scored 3.5-.5 against Gandalf in that tournament.
>>>
>>>It may be interesting if some program can take the first place from Ruffian.
>>>
>>>I am sure that Deep sjeng is going to try but I even doubt if it can get 79%
>>>against weaker programs in the 1st division.
>>>
>>>It has now 19.5/24 that is more than 79% but it still did not play against the
>>>best programs of the first division(Aristarch and delfi).
>>>
>>>If you claim that Ruffian's book is bad for Ruffian then you need to prove that
>>>Ruffian can get better results when it plays in conditions that are similiar to
>>>the nunn match against opponents.
>>
>>I found enough proves for it when I saw the games played by SSDF.
>>I will give you some examples from the Ruffian-Fritz 8 match.
>>You can be sure, I could find many more examples. I just dont have the time to
>>search for them.
>
>I know that Ruffian did not use book learning in the ssdf games so I think that
>using these games may lead to wrong conclusions.

Book learning will not prevent the first loss in a particular variation.
>
>I also see that most of the moves that you say that they are mistakes are not
>bad moves in Ruffian's book but simply cases when Ruffian is out of book and
>played a bad move and maybe it is better to blame the engine in these cases.
>
>Maybe Ruffian for some reasons performs better under winboard and I think
>that it will be a good idea if someone checks if the same moves in the games can
>be reproducable under winboard.
>
>We need to check that no one of the following problems do not happen:
>
>1)There are book moves that Ruffian knows not to play under winboard but does
>not know not to play in the ssdf conditions.
>
>2)Ruffian needs more time to find the right move in ssdf condition and
>practically could not do it in the game when under winboard it could find better
>move and avoid weak moves that you describe.

Do we know exactly how SSDF has tested Ruffian?
I have set up Ruffians as UCI engine and with the winboard adapter in Chessbase
and ran a testsuite with both some time ago. The result was that in about 50% of
the Testpositions Winboard Ruffian found the solution faster and in 50% UCI
Ruffian. The difference between the solution times for a position however was
almost always < 3s.

>
>3)The moves that you describe as weak are not weak(at least for computers in
>comp-comp games) and Ruffian lost because of later mistake(It is possible in
>order to check it to check if Fritz8 can beat other programs after the moves
>that you describe as weak)

The moves are weak. Especially for comp-comp games. That does not mean that they
are losing by force with best play. However programs dont play perfect.
They are sometimes weak in defending a worse position.

Michael



>
>
>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.