New Houdini

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
S.Taylor
Posts: 8402
Joined: Thu Mar 09, 2006 2:25 am
Location: Jerusalem Israel

Re: re: win adjudication/Re: New Houdini

Post by S.Taylor » Fri Nov 04, 2016 3:14 pm

mehmet karaman wrote:
Nay Lin Tun wrote:When will new Houdini be released? 1st Nov, any news?
Houdini will be released in the second week of November. (Robert said that at Tcec Forum)
THAT'S no good! It won't be in time for the s uperfinal, which is beginning between 5th-7th of November!

Dirt
Posts: 2851
Joined: Wed Mar 08, 2006 9:01 pm
Location: Irvine, CA, USA

Re: re: win adjudication/Re: New Houdini

Post by Dirt » Fri Nov 04, 2016 3:21 pm

S.Taylor wrote:
mehmet karaman wrote: Houdini will be released in the second week of November. (Robert said that at Tcec Forum)
THAT'S no good! It won't be in time for the superfinal, which is beginning between 5th-7th of November!
There are some logistics involved in getting out a commercial version but that won't stop TCEC from getting it.
Deasil is the right way to go.

User avatar
Laskos
Posts: 10240
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: re: win adjudication/Re: New Houdini

Post by Laskos » Sun Nov 20, 2016 9:34 am

Laskos wrote:
The most striking peculiarity is that Houdini keeps more Queens on the board till the Win adjudication than Stockfish and Komodo. Average number of Queens at the Win adjudication:

Houdini: 1.06
Stockfish: 0.70
Komodo: 0.68

97.5% or 2 standard deviations that it's not a statistical fluke.
Here above I speculated after 58 games of Houdini in TCEC Rapid, a month ago. It seemed maybe a statistical fluke, but it is not. The result is confirmed by Terence Rattray database of 22,000 games between Houdini 5 and Stockfish dev:
http://talkchess.com/forum/viewtopic.php?t=62195

Error margins 95% confidence:

Average won game length:
Houdini 5: 83.3 +/- 0.6
Stockfish: 85.6 +/- 0.5


Average number of Queens at the adjudication of won games:
Houdini 5: 1.02 +/- 0.02
Stockfish: 0.92 +/- 0.02

Average material at the adjudication of won games (1-3-3-5-9):
Houdini 5: 15.5 +/- 0.2
Stockfish: 14.2 +/- 0.2


We already know that Stockfish is more aggressive, keeps more material for win and wins in shorter games than Komodo. But Houdini 5 is even more of all that than Stockfish. Thanks to Terence for the excellent database.

User avatar
Laskos
Posts: 10240
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: re: win adjudication/Re: New Houdini

Post by Laskos » Sun Nov 20, 2016 4:34 pm

Laskos wrote:
Laskos wrote:
The most striking peculiarity is that Houdini keeps more Queens on the board till the Win adjudication than Stockfish and Komodo. Average number of Queens at the Win adjudication:

Houdini: 1.06
Stockfish: 0.70
Komodo: 0.68

97.5% or 2 standard deviations that it's not a statistical fluke.
Here above I speculated after 58 games of Houdini in TCEC Rapid, a month ago. It seemed maybe a statistical fluke, but it is not. The result is confirmed by Terence Rattray database of 22,000 games between Houdini 5 and Stockfish dev:
http://talkchess.com/forum/viewtopic.php?t=62195

Error margins 95% confidence:

Average won game length:
Houdini 5: 83.3 +/- 0.6
Stockfish: 85.6 +/- 0.5


Average number of Queens at the adjudication of won games:
Houdini 5: 1.02 +/- 0.02
Stockfish: 0.92 +/- 0.02

Average material at the adjudication of won games (1-3-3-5-9):
Houdini 5: 15.5 +/- 0.2
Stockfish: 14.2 +/- 0.2


We already know that Stockfish is more aggressive, keeps more material for win and wins in shorter games than Komodo. But Houdini 5 is even more of all that than Stockfish. Thanks to Terence for the excellent database.
With the same database, castling patterns show Stockfish as playing more similarly to human GM castling patterns than Houdini 5.

GM2600.pgn file, 27202 GM ELO 2600+ games:
O-O: 84.9%
O-O-O: 7.1%
No Cast: 8.0%

Houdini 5:
O-O: 87.5%
O-O-O: 4.1%
No Cast: 8.4%

Stockfish dev:
O-O: 85.7%
O-O-O: 6.0%
No Cast: 8.3%

Stockfish in castling more queenside, in line with human games.

Also: the cute g6 move as White is made by:

Houdini: 9.4% of White games
Stockfish: 11.7% of White games

Tdunbug
Posts: 47
Joined: Sun Mar 06, 2016 11:46 pm

Re: re: win adjudication/Re: New Houdini

Post by Tdunbug » Sun Nov 20, 2016 6:39 pm

Interesting analysis Kai Laskos. I am glad the database proved useful for showing tendencies of different engines. Would be cool to know other things like handling of the bishop pair, knight pair, center dominance, space adantages and attacking the king etc!

User avatar
Laskos
Posts: 10240
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: re: win adjudication/Re: New Houdini

Post by Laskos » Mon Nov 21, 2016 4:29 am

Tdunbug wrote:Interesting analysis Kai Laskos. I am glad the database proved useful for showing tendencies of different engines. Would be cool to know other things like handling of the bishop pair, knight pair, center dominance, space adantages and attacking the king etc!
Some of these things I am unable to filter out. I am using a text editor with powerful search function and Scid. PGN-Extract and 40H-PGN don't seem to help, or at least I am unable to use them for that. For now two of the more striking differences between Houdini 5 and Stockfish dev I get:

*** Wins with Queens on the board later in the game:

Houdini: 1147/3605 = 31.8%
Stockfish: 1478/5948 = 24.8%

Houdini likes to win with the Queens on the board more.


*** Bishop pair until late in the game:

Houdini: 1195 games
Stockfish: 795 games

Houdini likes the Bishop pair more overall (wins, draws, losses).

User avatar
Laskos
Posts: 10240
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: re: win adjudication/Re: New Houdini

Post by Laskos » Mon Nov 21, 2016 5:47 am

Laskos wrote:
Tdunbug wrote:Interesting analysis Kai Laskos. I am glad the database proved useful for showing tendencies of different engines. Would be cool to know other things like handling of the bishop pair, knight pair, center dominance, space adantages and attacking the king etc!
Some of these things I am unable to filter out. I am using a text editor with powerful search function and Scid. PGN-Extract and 40H-PGN don't seem to help, or at least I am unable to use them for that. For now two of the more striking differences between Houdini 5 and Stockfish dev I get:

*** Wins with Queens on the board later in the game:

Houdini: 1147/3605 = 31.8%
Stockfish: 1478/5948 = 24.8%

Houdini likes to win with the Queens on the board more.


*** Bishop pair until late in the game:

Houdini: 1195 games
Stockfish: 795 games

Houdini likes the Bishop pair more overall (wins, draws, losses).
One of the most striking:
*** Rook against minors and pawns later in the game:

Houdini: 292 games
Stockfish: 618 games

Also:
*** Rook pair until late in the game:

Houdini: 3436 games
Stockfish: 3876 games

Stockfish likes the Rook more.

User avatar
Laskos
Posts: 10240
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: re: win adjudication/Re: New Houdini

Post by Laskos » Mon Nov 21, 2016 12:16 pm

Laskos wrote:
Laskos wrote:
Tdunbug wrote:Interesting analysis Kai Laskos. I am glad the database proved useful for showing tendencies of different engines. Would be cool to know other things like handling of the bishop pair, knight pair, center dominance, space adantages and attacking the king etc!
Some of these things I am unable to filter out. I am using a text editor with powerful search function and Scid. PGN-Extract and 40H-PGN don't seem to help, or at least I am unable to use them for that. For now two of the more striking differences between Houdini 5 and Stockfish dev I get:

*** Wins with Queens on the board later in the game:

Houdini: 1147/3605 = 31.8%
Stockfish: 1478/5948 = 24.8%

Houdini likes to win with the Queens on the board more.


*** Bishop pair until late in the game:

Houdini: 1195 games
Stockfish: 795 games

Houdini likes the Bishop pair more overall (wins, draws, losses).
One of the most striking:
*** Rook against minors and pawns later in the game:

Houdini: 292 games
Stockfish: 618 games

Also:
*** Rook pair until late in the game:

Houdini: 3436 games
Stockfish: 3876 games

Stockfish likes the Rook more.
One more striking difference:

*** Pawn at g6 as White against castled Black Kg8. Often a King attack:

Houdini: 191 games
Stockfish: 498 games

User avatar
Nordlandia
Posts: 2631
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: re: win adjudication/Re: New Houdini

Post by Nordlandia » Fri Dec 02, 2016 3:21 pm

How does Houdini 5.01 Pro pluss 6-Men Nalimov assess this position ->

One individual active on this forum seem to have 6-men nalimov installed on SSD (can't remember username) Please share this puzzle if you know.

[d]6K1/4p3/3b4/8/b6P/2Bk4/8/2R5 b - - 0 1

Post Reply