Congratulations to the Stockfish team

Discussion of chess software programming and technical issues.

Moderator: Ras

Dann Corbit
Posts: 12792
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Congratulations to the Stockfish team

Post by Dann Corbit »

The latest incarnation of Stockfish is even good at openings analysis -- which previously was a "Rybka only" domain.

I have a database of a quarter million popular opening positions that I am analyzing. I already know the right moves, but I am performing mechanical verification by computer. Previously, only Rybka was even worth using because all other engines faired very badly. However, the current incarnation of Stockfish is as good, or possibly even better, than Rybka.
User avatar
michiguel
Posts: 6401
Joined: Thu Mar 09, 2006 8:30 pm
Location: Chicago, Illinois, USA

Re: Congratulations to the Stockfish team

Post by michiguel »

Dann Corbit wrote:The latest incarnation of Stockfish is even good at openings analysis -- which previously was a "Rybka only" domain.

I have a database of a quarter million popular opening positions that I am analyzing. I already know the right moves, but I am performing mechanical verification by computer. Previously, only Rybka was even worth using because all other engines faired very badly. However, the current incarnation of Stockfish is as good, or possibly even better, than Rybka.
Opening play is a difficult beast to tame. I am working on Gaviota's eval and I estimate (based on self play of super ultra fast games) that it is ~100 elo points stronger than the released version. However, based on observation, it plays horrible in the opening w/o book.

Miguel

This is embarrassing:

Code: Select all

        20   1:      0.0    +0.32  1.Nf3
        79   2:      0.0    +0.10  1.Nf3 Nf6
       321   3:      0.0    +0.32  1.Nf3 Nf6 2.Nc3
      1250   4:      0.0    +0.10  1.Nf3 Nf6 2.Nc3 Nc6
      4500   5:      0.0    +0.22  1.Nf3 Nf6 2.Nc3 Nc6 3.d3
      7889   6       0.0    +0.10  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6
     15933   6:      0.1    +0.10  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6
     26866   7       0.1    +0.21  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6 4.Be3
     62100   7:      0.2    +0.21  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6 4.Be3
     88080   8       0.3    +0.10  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6 4.Be3 Be6
    167625   8:      0.5    +0.10  1.Nf3 Nf6 2.Nc3 Nc6 3.d3 d6 4.Be3 Be6
    279440   9       0.9    +0.07  1.Nf3 Nf6 2.Nc3 Nc6 3.e3 e6 4.d4 Bd6
                                   5.Bd3
    622929   9:      1.8    +0.07  1.Nf3 Nf6 2.Nc3 Nc6 3.e3 e6 4.d4 Bd6
                                   5.Bd3
   1023834  10       3.1    +0.10  1.Nf3 Nf6 2.Nc3 d6 3.a3 Nc6 4.d3 a6
                                   5.Be3 Be6
   1803066  10:      5.4    +0.10  1.Nf3 Nf6 2.Nc3 d6 3.a3 Nc6 4.d3 a6
                                   5.Be3 Be6
   4516644  11      14.0    +0.19  1.Nf3 Nc6 2.Nc3 d5 3.d3 e5 4.h3 Nf6
                                   5.Bg5 h6 6.Be3
   7505272  11      23.3    +0.19  1.Nc3 Nf6 2.Nf3 d6 3.e4 Nc6 4.d4 h6
                                   5.d5 Ne5 6.Nxe5 dxe5
   7582804  11:     23.6    +0.19  1.Nc3 Nf6 2.Nf3 d6 3.e4 Nc6 4.d4 h6
                                   5.d5 Ne5 6.Nxe5 dxe5
  15909147  12      49.9    +0.10  1.Nc3 Nc6 2.d4 d5 3.Bf4 Bf5 4.Nf3 Nb4
                                   5.Rc1 Nf6 6.a3 Nc6
  38534673  12     120.9    +0.14  1.d4 Nf6 2.Nc3 d5 3.Bf4 h6 4.h4 a6
                                   5.Nf3 Nc6 6.e3 Bf5
  41707489  12:    130.0    +0.14  1.d4 Nf6 2.Nc3 d5 3.Bf4 h6 4.h4 a6
                                   5.Nf3 Nc6 6.e3 Bf5
  73997874  13     232.3    +0.18  1.d4 Nf6 2.Nc3 Nc6 3.Nf3 d5 4.Bf4 Nh5
                                   5.Be3 Bf5 6.Nh4 Bg4 7.g3
  90009426  13:    280.7    +0.18  1.d4 Nf6 2.Nc3 Nc6 3.Nf3 d5 4.Bf4 Nh5
                                   5.Be3 Bf5 6.Nh4 Bg4 7.g3
Tord Romstad
Posts: 1808
Joined: Wed Mar 08, 2006 9:19 pm
Location: Oslo, Norway

Re: Congratulations to the Stockfish team

Post by Tord Romstad »

michiguel wrote:Opening play is a difficult beast to tame. I am working on Gaviota's eval and I estimate (based on self play of super ultra fast games) that it is ~100 elo points stronger than the released version. However, based on observation, it plays horrible in the opening w/o book.
I think we tend to be too critical about how chess engines play the opening. We forget that humans have quite literally spent more than a hundred years analyzing the standard opening position and coming up with good plans and ideas for both sides. A hundred years of analysis and practical experience are hard to match, even for a computer that searches millions of positions per second.

To get a more realistic impression of the strength of chess engines in the early opening, try playing a few Chess960 games against one of the top programs. I've done that a few times myself against Stockfish, and usually find myself under heavy pressure before move 10, even when playing white. I think the top human players would score even worse against computers in Chess960 than in regular chess.
QED
Posts: 60
Joined: Thu Nov 05, 2009 9:53 pm

Re: Congratulations to the Stockfish team

Post by QED »

Tord Romstad wrote:To get a more realistic impression of the strength of chess engines in the early opening, try playing a few Chess960 games against one of the top programs.
I have no clear understanding of how the static evaluation is tuned, but has anyone tried do it only using matches of 1920 games (960 FRC positions, both colors)? I think it will result in better looking opening play, comparing to engine tuned (for example) to Sicilian Najdorf middlegame.
Testing conditions:
tc=/0:40+.1 option.Threads=1 option.Hash=32 option.Ponder=false -pgnin gaviota-starters.pgn -concurrency 1 -repeat -games 1000
hash clear between games
make build ARCH=x86-64 COMP=gcc
around 680kps on 1 thread at startposition.