Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Big Book Of Combinations - Ikarus 0.18

Author: Ed Schröder

Date: 06:41:10 09/18/01

Go up one level in this thread


On September 18, 2001 at 08:08:23, Jouni Uski wrote:

>On September 18, 2001 at 07:23:42, Ed Schröder wrote:
>
>>On September 18, 2001 at 06:26:05, Uri Blass wrote:
>>
>>>On September 18, 2001 at 04:38:32, Steve Maughan wrote:
>>>
>>>>The much underestimated Ikarus 0.18 does well with 27 / 42 at 10 m / m.  Not bad
>>>>at all!
>>>
>>>I believe that the test is not a good test to test tactical ability because I
>>>see often cases when programs find the move for positional reasons when there is
>>>almost no difference in the score between the move that the programs find and
>>>the previous best move.
>>>
>>>I do not say that there is no tactics but the tactics is simply too deep in part
>>>of the cases and programs may find the right move in few seconds when they need
>>>a long time to find a big difference between the evaluation of the best move and
>>>the evaluation of the second best move.
>>>
>>>I read that Hiarcs is more materialistic than other programs and I also know
>>>that it is considered to be weak in king sefety's evaluation and it can explain
>>>the fact that it could not solve a lot of positional problems (from computer's
>>>point of view) in the test suite.
>>>
>>>Uri
>>
>>
>>It depends of course what you want to test. This set is about king safety and
>>search. In theory everything can be solved by search. If you can search 200
>>plies deep there is no need to build chess knowledge into a chess program.
>>Since it is impossible to search 200 plies you need chess knowledge. That's
>>why for me this set was a welcome contribution to test Rebel's king satefy as
>>it offered me new (and good) king safety material.
>>
>>There is no final test-set, never will.
>>
>>Ed
>
>Hi Ed. Can You explain how You can use this set to king safety? I think in most
>positions King is already doomed - do You check Rebel's evaluation without
>search or what?
>
>Jouni


Hi Jouni,

After Maastricht I have changed Rebel's king saftey and tested it first with
a large set of positions with the topic king safety only. I know most of these
positions by head and while watching the screen I immediately can see if the
changes made to the program have the wished effect (or not).

An annoying fact of this approach is that positions get out-dated, out-dated
in the sence that nowadays software + hardware will find the key move of the
positions I used in (say) 1994 within a second (or so) and thus I frequently
have to update my test positions and memorize them again. So this set posted
by you was very welcome for me.

Another aspect I am always very keen on when the subject is king safety is
the issue of "balance", the balance between material sacrifices and the
positional compensation because of that. In the case of your test-set I have
also ran it using Century3 and it only gets 23 wheras the latest version gets
33.

When I look at the differences I see many positions are found 1-4 plies earlier
due to the fact of the new king safety and the rest of the improvements is
simply found due to a faster search.

I am now looking for unjustified sacrifices, so far I have found only one:

[d]r2q1r2/pb3pk1/1p3bpp/3pN3/2nP4/P1N5/1P3PPP/1BRQR1K1 w - -

Here Rebel at the first 7 plies wants to play 1.Nxg6 but later is replaced
by more careful moves like 1.Nxc4 or 1.Qf3. The move to find in this position
is 1.Qg4! Rebel wants to play 1.Qg4 at 2 minutes but unfortunately is replaced
later by 1.Nxc4 by a deeper search. King safety... a never ending story.

Ed



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.