Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How to judge?

Author: Christophe Theron

Date: 23:10:18 12/28/99

Go up one level in this thread


On December 28, 1999 at 15:24:04, Ed Schröder wrote:

>>Posted by John Warfield on December 28, 1999 at 07:02:31:
>>
>>>Thanks for the compliment in the first place. Your observations are very
>>>interesting for me. What in your opinion was it that made Rebel9 so much
>>>better than version 8? Same question for Rebel10 -> Century. I know these
>>>questions are hard to answer because the subjective nature. Keep in mind
>>>you maybe got a faster PC during Rebel9 and Century?
>>>
>>
>>
>>  Well as you said my impression is mostly subjective. But after playing hundres
>>of games with 6, , 8, 9, 10, and now century , one gets to get used to the
>>playing styles.  A subtle differnece I noticed between version 9 and 10, is that
>>Rebel 10 tended to play more aggressive. Also I felt more pressure on my game
>>out of the opening. Don't get me wrong I thought rebel 9 was extremely strong,
>>infact I have to say, I believe the difference between rebel 10 and century is
>>larger than the differences in rebel 9 and 10.  It just seems to me that rebel
>>10 possessed just a little more finesse than version 9.  Another notable
>>diffence in Rebel 10 from nine, was that version 10 seemed to be alot less
>>materialistic, Positional considerations seemed to play a larger role in it's
>>evaluation. It would not hungrily snatch a pawn as 9 would occasionally do.
>> I think the main improvemnt of century over Rebel 10, is that Century now has
>>much more tactical vision then 10, not that I was able to play tactically with
>>version 10, but in my observations in computer vs computer play, i noticed
>>century doesn't get tactically outplayed by the likes of chessmaster and fritz
>>as was the case with rebel 10. Also Century seems to be even more agressive
>>than 10, it has a very nice human feel to it, my games are never boring. it makes
>>beautiful quiet moves that are often seen in grandmaster play.  (good example
>>is the Kg6!! in the rebel russek game!)  After a game with Century I get a feeling
>>of chess art and apprecatation.  I am not claiming any of these observations are
>>facts, i can only say they are crystal clear and evident to me personally.
>>Although Many may not be able to articulate these differences precisely, I am
>>sure that anyone who is a relatively strong chess player, will immediately
>>notice the differences in strength and playing style.
>
>Hundreds of games, wow. One plausible reason seems to be the improvements
>made to selective search. Every release (year) at least one month (sometimes
>3 months) are spend to work on selective search. Failures in selective search
>can hurt a chess program a lot. Here is a (tactical) example:
>
>1nr1b1k1/r4p2/1p1q2p1/pPppN3/3P4/2RBP3/P1Q2PP1/2R3K1 w - - bm dxc5;
>
>00:00:06  8.00  0.16   1.Nc6 Nxc6 2.dxc5 bxc5 3.Rxc5 Rac7
>                       4.bxc6 Rxc6 5.Rxc6 Rxc6 6.Qb2 Rxc1+
>                       7.Qxc1  (6)
>
>00:00:14  8.33  0.18   1.dxc5 Qxe5 2.cxb6 Rxc3 3.Qxc3 Qxc3
>                       4.Rxc3 Rb7  (13)
>
>00:00:15  9.00  1.42   1.dxc5 Qxe5 2.cxb6 Rxc3 3.bxa7 Rxc2
>                       4.Bxc2 Qb2 5.axb8=Q  (15)
>
>00:00:22 10.00  1.51   1.dxc5 Qxe5 2.cxb6 Rxc3 3.bxa7 Rxc2
>                       4.Bxc2 Qb2 5.axb8=Q  (20)
>
>Try this with previous Rebel's and 1.dxc5! isn't found at all because of a
>pruning failure.
>
>
>A positional example:
>
>1k6/3K4/1P3p1p/3B2pP/3b2P1/8/8/8 w - - bm b7;
>
>00:00:00  9.00  1.07   1.Kc6 Kc8 2.Bc4 Be5 3.Be6 Kb8 4.Bd5
>00:00:00  9.01  1.07   1.b7
>00:00:01  9.01  1.59   1.b7 Bc5 2.Ke6 f5 3.gxf5 g4 4.f6
>                       Ka7 5.Kf7 Bd4  (0)
>
>00:00:01 10.00  1.72   1.b7 Be5 2.Ke6 Kc7 3.Kf5 Kb6 4.Kg6
>                       Kc5 5.Bf3 f5 6.gxf5  (1)
>
>00:00:02 11.00  1.79   1.b7 Be5 2.Ke6 Kc7
>00:00:04 12.00  2.03   1.b7 Be5 2.Be4
>00:00:07 13.00  2.07   1.b7 Be5 2.Be4 Bc3+
>
>Key move found op ply 9. Rebel10 needs 13 plies.
>
>
>Last example:
>
>3rrb2/1pq2ppk/p1p1p2p/4P3/3P4/2P3QR/PP4PP/2B2R1K w - - bm Rf6 or Bxh6;
>
>00:00:02  7.00  1.32   1.Qg4 Qe7 2.Bg5 f6 3.exf6 Qf7 4.Qe4+
>                       Kg8  (2)
>
>00:00:04  7.10  1.32   1.Bxh6
>00:00:05  7.10  3.36   1.Bxh6 Kg8 2.Bg5 Be7 3.Qh4 f5 4.exf6
>00:00:07  8.00  3.19   1.Bxh6 gxh6 2.Rf6 Qe7 3.Qg5 Qxf6
>                       4.Qxf6 Rd7 5.Rg3 Red8  (6)
>
>Rebel10 needs 11 plies to see 1.Rf6!
>
>These so called "selective search failures" hurt chess programs a lot
>especially in the positional area as I had to learn the hard way as
>these mistakes in the selective search are "invisible". Your program
>plays an inferior move (no blunder) and after hours of debugging its
>reason is not the present chess knowledge but selective search.
>
>Tactical failures of selective search are easy to recognize as your
>program misses a pre-defined key-move and these are (luckily) not so
>many. But these positional errors in selective search are real killers
>for a program's positional play and often a lot of chess knowledge
>present in the program is simply killed because of that resulting
>in minor positional mistakes (no big blunders).
>
>Since version 9 I have made it an important topic for Rebel. Especially
>Rebel10 and Century have many improvements here. There is quite a loss
>in ply-depth (factor 2 between version 9 and Century compensated by new
>speed-up ideas) but the quality in the positional area improved a lot not
>because of the added new chess knowledge only but most of the time by not
>pruning "good moves" any more as now Rebel's chess knowledge is not killed
>any more by selective search failures.
>
>Errors in selective search can make an intelligent chess program sometimes
>look very dumb while actually there is no reason for that. The program has
>the knowledge to play the right move but during the search the knowledge is
>killed (lost).


I totally agree, Ed.

As I said in another post several weeks ago, it is very easy for a search
algorithm to spoil the positional knowledge of a program.

It sounds counter-intuitive, but a better search does not bring tactical
strength only.

It mainly brings more positional accuracy.

Good search is not for tactics.

Good search is for positional play!


    Christophe



>I consider these selective search improvements as the main reason Rebel
>plays better chess. Other reasons are of course new added chess knowledge
>but I do think the selective search improvements are dominant. Also Century's
>new king safety has done some good too to Rebel's tactical and positional
>understanding. It was (is) definitely needed to survive against GM's.
>
>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.