Weak Opennings unearthed by Yorkman tourn?

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.
MikeGL
Posts: 901
Joined: Thu Sep 01, 2011 12:49 pm

Weak Opennings unearthed by Yorkman tourn?

Post by MikeGL » Wed Jan 17, 2018 4:55 am

From the Yorkman tournament thread, where 100 games were played between H6 and latest SF8 Dev at
http://chess.my2dollars.ca the tourn seems to have discovered some dubious black opennings
(assuming equal strength for both sides) because neither SF8 dev nor H6 can hold blacks position.

Image

Weak Opennings for black,
C16 French-Petrosian Variation
C96 Ruy Lopez-Keres Variation.

Both engines of almost equal strength, playing around 3400+ can't save black from these 2 opennings.
Maybe safe to label those 2 opennings 'dubious' for black. Out of nearly 50 opennings played, only these two popped up.

A quick check on Database also proves C16 and C96 don't show very good result for black on top level.
Ok, on lower rated games it's about equal with no sign that black is inferior.

Image

Image

Maybe playable at lower level, but at top level chess (super GM's, including eng-eng matches)
these 2 opennings looks dead.

BrendanJNorman
Posts: 1679
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Weak Opennings unearthed by Yorkman tourn?

Post by BrendanJNorman » Wed Jan 17, 2018 6:03 am

MikeGL wrote:From the Yorkman tournament thread, where 100 games were played between H6 and latest SF8 Dev at
http://chess.my2dollars.ca the tourn seems to have discovered some dubious black opennings
(assuming equal strength for both sides) because neither SF8 dev nor H6 can hold blacks position.

Image

Weak Opennings for black,
C16 French-Petrosian Variation
C96 Ruy Lopez-Keres Variation.

Both engines of almost equal strength, playing around 3400+ can't save black from these 2 opennings.
Maybe safe to label those 2 opennings 'dubious' for black. Out of nearly 50 opennings played, only these two popped up.

A quick check on Database also proves C16 and C96 don't show very good result for black on top level.
Ok, on lower rated games it's about equal with no sign that black is inferior.

Image

Image

Maybe playable at lower level, but at top level chess (super GM's, including eng-eng matches)
these 2 opennings looks dead.
I think you've made some false deductions here, Mike.

I have experimented a lot with the Petrosian variation, and with various types of engines and they simply do not understand black's strategy.

Engines are still far from understanding that black has to maneuver for some time in a cramped position and only slowly unravel.

Petrosian used to actually keep the black bishop on b7, block the c-pawn with ...Nc6 and 0-0-0 - followed by very patient unravelling play.

Try to explain that to an engine! :lol:

Engines invariably try traditional ideas, like to swap the bad bishop with ...Ba6 and play ...c5, sometimes even castling long.

When they realize that this is a little dangerous (structural and positional considerations aren't everything), it's too late.

This usually comes in the form of white quickly opening things up with something like dxc5 and c4, and when black plays ...d4 white gets e4 for his knight and a king attack develops.

I have seen my engines/personalities misplay the structure like this, and indeed myself lost like this in a tournament game back in 2009.

I rediscovered this when trying to tune the Petrosian personality for Rodent...

Often engines will do this ...Ba6/...c5 stuff and beat their opponent due to being simply stronger, but against an equal opponent there might be consequences.

I think with both this line, and the Keres line, the black side requires very human-like maneuvering and tolerating a lack of space, whilst on the other hand, the white side simply transfers pieces to the kingside and creates threats (easy for a computer).

This kind of reminds me of 15+ years ago when engines always wanted to prematurely play ...d5 in the Boleslavsky Sicilian structures, because they thought black had no space.

So this to me, is not so much the problem with the opening, but with the fact that engines still don't play these lines well.

At the Super-GM human level, I think they're fully playable.

User avatar
Ovyron
Posts: 3097
Joined: Tue Jul 03, 2007 2:30 am

Re: Weak Opennings unearthed by Yorkman tourn?

Post by Ovyron » Wed Jan 17, 2018 6:17 am

BrendanJNorman wrote:Try to explain that to an engine! :lol:
I have a feeling that Alpha Zero would just be fine on either of these :D

BrendanJNorman
Posts: 1679
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Weak Opennings unearthed by Yorkman tourn?

Post by BrendanJNorman » Wed Jan 17, 2018 6:18 am

Ovyron wrote:
BrendanJNorman wrote:Try to explain that to an engine! :lol:
I have a feeling that Alpha Zero would just be fine on either of these :D
But AZ isn't an engine, right? :wink:

User avatar
Ovyron
Posts: 3097
Joined: Tue Jul 03, 2007 2:30 am

Re: Weak Opennings unearthed by Yorkman tourn?

Post by Ovyron » Wed Jan 17, 2018 6:23 am

BrendanJNorman wrote:
Ovyron wrote:
BrendanJNorman wrote:Try to explain that to an engine! :lol:
I have a feeling that Alpha Zero would just be fine on either of these :D
But AZ isn't an engine, right? :wink:
Well, it's chess playing software, so it counts as an engine in my book :)

MikeGL
Posts: 901
Joined: Thu Sep 01, 2011 12:49 pm

Re: Weak Opennings unearthed by Yorkman tourn?

Post by MikeGL » Wed Jan 17, 2018 7:08 am

Thanks. Will check those games a bit deeper, especially how Tigran Petrosian himself handles this French line with black..

Tobber
Posts: 379
Joined: Fri Sep 28, 2012 3:53 pm
Location: Sweden

Re: Weak Opennings unearthed by Yorkman tourn?

Post by Tobber » Wed Jan 17, 2018 8:06 am

Have a look at the TCEC Superfinal. Same thing, neither engine could hold as black in the Petrosian variant.

/John

jdart
Posts: 3866
Joined: Fri Mar 10, 2006 4:23 am
Location: http://www.arasanchess.org

Re: Weak Opennings unearthed by Yorkman tourn?

Post by jdart » Wed Jan 17, 2018 12:42 pm

I have the ChessBase Correspondence Database, and I always find it helpful to see what correspondence players are choosing in terms of openings. They have days to make a move and can use computer assistance. These days there is a very high draw rate in correspondence chess. What you see pretty often are lines that look lopsided at first but eventually lead to a draw score. So you can't just rely on the engine evals out of book, at least if they are not based on very deep search. And I think the top players do a lot of preparation outside games, as OTB players do.

-- Jon

MikeGL
Posts: 901
Joined: Thu Sep 01, 2011 12:49 pm

Re: Weak Opennings unearthed by Yorkman tourn?

Post by MikeGL » Fri Jan 19, 2018 4:50 am

BrendanJNorman wrote: This usually comes in the form of white quickly opening things up with something like dxc5 and c4, and when black plays ...d4 white gets e4 for his knight and a king attack develops.

I have seen my engines/personalities misplay the structure like this, and indeed myself lost like this in a tournament game back in 2009.

I rediscovered this when trying to tune the Petrosian personality for Rodent...

Often engines will do this ...Ba6/...c5 stuff and beat their opponent due to being simply stronger, but against an equal opponent there might be consequences.

This kind of reminds me of 15+ years ago when engines always wanted to prematurely play ...d5 in the Boleslavsky Sicilian structures, because they thought black had no space.

So this to me, is not so much the problem with the opening, but with the fact that engines still don't play these lines well.

I just checked
[d]rnb1k1nr/p1pq1ppp/1p2p3/3pP3/3P4/P1P5/2P2PPP/R1BQKBNR w KQkq - 0 7

Engine-Engine matchs High Quality (only rated 2800 and above) from random sources CCRL/CEGT/SSDF/Talkchess Tourn Forum etc
Out of 621,000 games there were 74 French Petrosian
Ba6 line: 64
Bb7 line: 9

MillionBase by Ed Schroder (with very few engine-engine games)
out of 2.91M games only 614 French Petrosian
Ba6 line: 570
Bb7 line: 44

Bb7 (not Ba6) may be the modern way to handle this position as shown by Caruana on his latest French, but it was rapid or blitz.
I rechecked and noticed even ex world champion Tigran Petrsian himself seems to
prefer Ba6 move, so looks like engines are playing French Petrosian correctly.

MikeGL
Posts: 901
Joined: Thu Sep 01, 2011 12:49 pm

Re: Weak Opennings unearthed by Yorkman tourn?

Post by MikeGL » Fri Jan 19, 2018 4:57 am

Tobber wrote:Have a look at the TCEC Superfinal. Same thing, neither engine could hold as black in the Petrosian variant.
/John
Thanks for the info, already checked those high quality games the previous day.

jdart wrote:I have the ChessBase Correspondence Database, and I always find it helpful to see what correspondence players are choosing in terms of openings. They have days to make a move and can use computer assistance. These days there is a very high draw rate in correspondence chess. What you see pretty often are lines that look lopsided at first but eventually lead to a draw score. So you can't just rely on the engine evals out of book, at least if they are not based on very deep search. And I think the top players do a lot of preparation outside games, as OTB players do.

-- Jon
Would be interesting to see the stats on wins/losses and the numbers of Ba6 lines played.

Post Reply