Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: CCT-4 Sjeng Operator thoughts weekend #2

Author: José Carlos

Date: 13:11:26 01/31/02

Go up one level in this thread


On January 31, 2002 at 14:31:12, Dann Corbit wrote:

>On January 31, 2002 at 14:06:10, Chessfun wrote:
>
>>Sjeng.
>>
>>The week started badly as I collapsed at work on Tuesday only to wake up in
>>hospital to be told my gall bladder was bad. So, after they removed it they then
>>let me go home the same day.....modern health care?.
>>
>>I wasn't sure how much pain I was going to be in and whether I would be able to
>>operate Sjeng. As it was, sitting and being able to have something to focuss my
>>attention on during the weekend helped.
>
>Glad that the problem was found and solved.  Hope you are back to 100% very
>soon.
>
>>Sat 26 Jan.
>>Game 7.
>>Sjeng v AvernoX 0-1
>>
>>Seems Averno had done some homework on Sjeng's opening book. As black plays
>>7.....c5 a move I can't find in other opening book and then Sjeng proceeds to
>>stay in book till move 20?. Coming out of book at almost -3.00. Averno stayed in
>>book until move 25 and it's first move from book is in the following position.
>>As noted by the kibitzed notations following;
>>
>>[D]r5k1/pp1r1pp1/5n1p/2P5/P1P1p3/4B3/4BPPP/5RK1 b - - 0 25
>>
>>--- Game 554: Sjeng vs AvernoX ---
>>Sjeng(C) whispers: d13 -2.84 Be2 Rd7 h4 Kf8 a4 Rc8 a5 Rc6 g4 Ke7 Kg2 Ke6 Rd1 n:
>>77984732 qp: 22% fh: 87% c-x: 588476 r-x: 417704 1-x: 39905 t-x: 509 egtb: 0
>>time: 116.61 nps: 669109
>>
>>AvernoX(C) kibitzes: Depth:0  Ev:-0.01  Time:0.00  Nodes:30  PV: (0 knps)  *
>>EGTB: 0/0
>>
>>Sjeng(C) whispers: d13 -2.78 a4 Re8 h4 Kf8 a5 Re6 g4 Ke7 Kg2 Rc6 Rb1 h5 gxh5 n:
>>104951851 qp: 23% fh: 87% c-x: 563810 r-x: 595175 1-x: 34442 t-x: 1107 egtb: 0
>>time: 161.10 nps: 651915
>>
>>AvernoX(C) kibitzes: Depth:13  Ev:2.18  Time:183.00  Nodes:98700004  PV:a8e8
>>f1b1 e8e6 h2h4 e6a6 b1b4 g8h7 h4h5 a6e6 e3f4 e6c6 b4b5 * (539 knps)  * EGTB: 0/0
>>
>>Naturally the opening has already decided the outcome, although Sjeng did manage
>>to make it hard for Averno at one point getting it's minus score down to -1.7
>>but was checkmated on move 89.
>>
>>After this game I was in a bit of panic as Gian-Carlo wasn't available. I looked
>>at the two opening books I was using, normal.opn and nbook.bin. And it appears
>>the standard normal.opn follows this line, as when I removed it and simply used
>>nbook.bin Sjeng would not follow this line. Yet with normal.opn in place it
>>almost always followed the line. I deleted the few lines in normal.opn that
>>followed this line and although it would still prefer 4. Qc2 I never had time to
>>check if this was caused by normal.opn or nbook.bin.
>>
>>I am curious to know what part of the book was at fault? and whether Averno's
>>operator had set up this line, as naturally it was in his book.
>
>I don't remember GCP's book algorithm.  I suspect that both programmers used the
>same source for their opening book.  Many "automatically generated" opening
>books just consider that any move a GM has played must be good, and especially
>if it has been played twice.  Some opening books do not even consider if the one
>who made the book won or lost the game!
>
>I rather suspect that this was just an example of the difference in quality
>between an automatic book and a hand-tuned book.  I doubt very much if any of
>the amateur programs have any real hand work in their books.

  My book:

  - was based in pgn files, having winning side into account
  - had many lines from Carlos Pesce's excellent book
  - had some hand tuned lines, that I analyzed myself with four computers at the
same time (four different programs, one of which was Averno, just to make sure
it would 'understand' the positions), and with a bit of my intuition (not much,
though, as my ELO is only about 2150 FIDE)
  - had tons of automatically learned knowledge. I had a computer running test
matches with learning enabled for a week. Do you guess what the opponents
were...? :)

  I had been waiting for this tournament since, in CCT3, I didn't have my new
version ready and had to play with an obsolete version. I did my best
(considering my lack of time) to well in CCT4. And I believe I did not too bad
:)

  José C.

>In this long post:
>http://f11.parsimony.net/forum16635/messages/21642.htm
>I discuss how Pepito's opening book for 1.42 and earlier had Noah's Arc Trap in
>it, *AND* would play the losing side!
>
>Amateur opening books make for great amusement from time to time.  It's just
>another element of computer chess that needs more attention from the
>programmers.



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.