ShashChess

Discussion of anything and everything relating to chess playing software and machines.

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Ozymandias
Posts: 1222
Joined: Sun Oct 25, 2009 12:30 am

Re: ShashChess

Post by Ozymandias » Fri Sep 11, 2020 8:26 pm

I suspect that most of the Elo gains for SF won't come from improving neither the classical eval, nor the new one, but rather from tuning when to use one or the other. We know search was already excellent, so I don't expect that to contribute much growth either.

amchess
Posts: 148
Joined: Tue Dec 05, 2017 1:42 pm

Re: ShashChess

Post by amchess » Thu Sep 17, 2020 4:39 pm


User avatar
MikeB
Posts: 4412
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: ShashChess

Post by MikeB » Fri Sep 18, 2020 1:36 am

amchess wrote:
Thu Sep 17, 2020 4:39 pm
Added linux versions:
https://github.com/amchess/ShashChess/releases/tag/14
Good work - your work inspire me to get Shashin's book. I noticed that your T, P and C options are set to false, I'm assuming that is for best play. Was wondering how much Elo loss ( or perhaps gain) have you measured with T, P & C set to true?
Image

carldaman
Posts: 2004
Joined: Sat Jun 02, 2012 12:13 am

Re: ShashChess

Post by carldaman » Fri Sep 18, 2020 10:29 am

viewtopic.php?f=2&t=49282&p=533571&hili ... in#p533571

I just realized that seven years ago I was the first to post about Shashin's book and its potential for aiding engines and not just humans. :o

amchess
Posts: 148
Joined: Tue Dec 05, 2017 1:42 pm

Re: ShashChess

Post by amchess » Fri Sep 18, 2020 10:42 am

Every position can be of type
Tal, Capablanca, Petrosian or mixed type: Tal-Capablanca or Capablanca-Petrosian:
As rule of thumb, if you set the contempt to 0 and reset any dynamic contempt,
if on a position the engine like stockfish shows a score
  • <-0.25 the position is of Petrosian type
  • =-0.25 the position is of Petrosian-Capablanca type (the so called caos positions in terms of Shashin theory)
  • between -0.25 and 0.25 the position is of Capablanca type
  • =0.25 the position is of Capablanca-Tal type (the other type of so called caos positions in terms of Shashin theory)
  • >0.25 the position is of Tal type
There is also, strictly speaking a final caos positions type: Tal, Capablanca and Petrosian.
This is for positions where even the above rule of thumb can't be applied.
Those positions can be reocognized because the engine's score swing very frequently between the indicated ranges.
Anyway, if, for a position, you know a priori the Shashin's type, if you check the corresponding checkbox(es), the analysis will be better, but surely the default option is the best for match playing, because the initial position is even (capablanca type also equivalent to unchecked boxes).

Paloma
Posts: 977
Joined: Thu Dec 25, 2008 8:07 pm

Re: ShashChess

Post by Paloma » Sat Sep 19, 2020 2:55 am

carldaman wrote:
Fri Sep 18, 2020 10:29 am
viewtopic.php?f=2&t=49282&p=533571&hili ... in#p533571

I just realized that seven years ago I was the first to post about Shashin's book and its potential for aiding engines and not just humans. :o
The Link to Chessvibes don't exist anymore. :(
I got it via Wayback. :D

http://web.archive.org/web/201309090544 ... -best-play

btw. is chessvibes.com home page dead?

BrendanJNorman
Posts: 2291
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: ShashChess

Post by BrendanJNorman » Sat Sep 19, 2020 3:18 am

Paloma wrote:
Sat Sep 19, 2020 2:55 am
btw. is chessvibes.com home page dead?
Chessvibes was acquired by chess.com a few years ago, I believe.

One of the big chess sites anyway...pretty sure it was chess.com.

User avatar
Ozymandias
Posts: 1222
Joined: Sun Oct 25, 2009 12:30 am

Re: ShashChess

Post by Ozymandias » Sun Sep 20, 2020 3:41 pm

amchess wrote:
Fri Sep 18, 2020 10:42 am
Every position can be of type
Tal, Capablanca, Petrosian or mixed type: Tal-Capablanca or Capablanca-Petrosian:
As rule of thumb, if you set the contempt to 0 and reset any dynamic contempt,
if on a position the engine like stockfish shows a score
  • <-0.25 the position is of Petrosian type
  • =-0.25 the position is of Petrosian-Capablanca type (the so called caos positions in terms of Shashin theory)
  • between -0.25 and 0.25 the position is of Capablanca type
  • =0.25 the position is of Capablanca-Tal type (the other type of so called caos positions in terms of Shashin theory)
  • >0.25 the position is of Tal type
There is also, strictly speaking a final caos positions type: Tal, Capablanca and Petrosian.
This is for positions where even the above rule of thumb can't be applied.
Those positions can be reocognized because the engine's score swing very frequently between the indicated ranges.
Anyway, if, for a position, you know a priori the Shashin's type, if you check the corresponding checkbox(es), the analysis will be better, but surely the default option is the best for match playing, because the initial position is even (capablanca type also equivalent to unchecked boxes).
I'm not familiar with the book, but from what I recall about the theory, it was about evaluating positions. If the evaluation is in the nnue file, how does Shashchess apply Shashin's theory? Trough search alone? Does it apply Tal, Capablanca, Petrosian values afterwards to guide it?

amchess
Posts: 148
Joined: Tue Dec 05, 2017 1:42 pm

Re: ShashChess

Post by amchess » Sun Sep 20, 2020 9:30 pm

Hi.
Many thanks for the interesting question.
Clearly, Shashin's theory must be adapted to a chess engine.
If you sees the code, the position's nature is etablished at each depth from the previous depth static/quiescent position.
The initial position is "bootstrapped" by the three Shashin's options.

amchess
Posts: 148
Joined: Tue Dec 05, 2017 1:42 pm

Re: ShashChess

Post by amchess » Sat Oct 03, 2020 11:19 am


Post Reply