Queens vs Knights

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

Moderators: hgm, Harvey Williamson, bob

Uri Blass
Posts: 8010
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Need material for a speech

Post by Uri Blass » Mon Jul 17, 2017 12:25 pm

hgm wrote:Yes, QueeNy is about 1000 Elo weaker than Stockfish and Komodo (when run with normal piece values).

The point, however, is that it beats Komodo and Stockfish when playing with the Knights in this position. This is the hallmark of won positions; once you know how to win them, it doesn't matter how strong the opponent is. Won positions are won even against perfect play. TSCP beats Stockfish 100% of the time, in KQK. Even if Stockfish uses EGT.

In this case the knowledge is that you should not allow Q for 2N trading. Normal engines, including Komodo and Stockfish, might not know this, and frequently bungle the won game, even losing it. QueeNy has Q=9.5 and N=5, which makes it value 2 Knights above a Queen, and thus avoid conversion to 2Q vs 5N. Given enough depth to recognize the tactical threats for forcing such trades, it will be able to avoid them. Seven Knights is enough to keep all Knights protected twice virtually all the time. N=5 might even be too low, as it now would allow Q for 2N+P trades. This would only hurt when the opponent actively seeks such trades, though, which is probably not the case for Komodo.

Bishops cannot cooperate this well: they cannot protect each other when on different color, and they strongly hamper each other's mobility when the do protect each other. Knights and Bishops also do not mix well, because the protection they can offer is not mutual. So any Knight->Bishop substitution strongly weakens the side with the minors, and results of such games prove nothing about the pure Q vs N case.

So QueeNy is not "the measure of all things", but it does beat the top engines with the Knights in this position, with a much higher score than other top engines would achieve against the same opponent. This was tested with thousands of games. And that is the definition of "being better". Despite their strong search, most top engines just lack the evaluation knowledge to play this successfully. Better search is of no use if a poor evaluation doesn't let you search for the right thing.

All you do is offer (completely uninformed) talk, and, as they say, talk "cuts no wood". It is game results that define reality, and any reasoning that denies reality is know as 'delusion'...
Better search can help even if you have worse evaluation because chess is not an evaluation contest but the question is if the search of chess programs is really better in the relevant position.

Can latest stockfish beat programs that are 300 or 400 elo weaker in normal chess in the relevant position and also do not have specific knowledge for queens against knights thanks to better search?

The only way to know is by testing and
I do not know because I did not test it.

User avatar
Eelco de Groot
Posts: 3834
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Need material for a speech

Post by Eelco de Groot » Mon Jul 17, 2017 12:36 pm

Nordlandia wrote:Mark said that he couldn't find any extra queen penalty in SF8 code.

Komodo is better adjusted for imbalances, such as this.
Well, somebody indeed claimed on Fishcooking Googlegroup that Mark Lefler would have said this, but if it is true Mark must have missed a lenghty period of Stockfish code development (before Stockfish 8) during which there was some explicit code for both Rook and Queen major redundancy, and it included in a comment references to Larry Kaufman's article. It brought Elo so why would we have anbandoned it. The code was simplified into constants for the material imbalance table. That the references were also stripped is a pity. That the code could be improved is also possible. That Komodo specifically does this better is a claim made by Mark maybe, or said poster in Fishcooking, but I have not seen overwhelming evidence of this.

It is easy to show that Harm Geert Muller is right about Knights winning in the "Charge of the light brigade" position. I consider this proven beyond any doubt.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 1:12 pm

Eelco de Groot wrote:
Nordlandia wrote:Mark said that he couldn't find any extra queen penalty in SF8 code.

Komodo is better adjusted for imbalances, such as this.
Well, somebody indeed claimed on Fishcooking Googlegroup that Mark Lefler would have said this, but if it is true Mark must have missed a lenghty period of Stockfish code development (before Stockfish 8) during which there was some explicit code for both Rook and Queen major redundancy, and it included in a comment references to Larry Kaufman's article. It brought Elo so why would we have anbandoned it. The code was simplified into constants for the material imbalance table. That the references were also stripped is a pity. That the code could be improved is also possible. That Komodo specifically does this better is a claim made by Mark maybe, or said poster in Fishcooking, but I have not seen overwhelming evidence of this.

It is easy to show that Harm Geert Muller is right about Knights winning in the "Charge of the light brigade" position. I consider this proven beyond any doubt.
code is getting shorter, man.

more deletions than additions, that is what warms up developers' hearts. :)

minor pieces do gain value in respect to the queen, but the question is in what precise measure.

Harm's assessment is definitely too much exaggerated.

for example, 2 Qs vs 5 Ns is almost certainly won for the queens.

only thing that matters is what works in code, for example, if you bonise minors vs queen, and then additionally penalise redundancy of major pieces, in a QRR vs RRBNN, the queen will get 2 penalties, maybe a bit too much.

but you are right, SF has still some 100 elo to gain only on imbalances. (current mainframers would not quite agree, I know, judging by the difficulty of a patch passing, but it is true)

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 1:22 pm

hgm wrote:Yes, QueeNy is about 1000 Elo weaker than Stockfish and Komodo (when run with normal piece values).

The point, however, is that it beats Komodo and Stockfish when playing with the Knights in this position. This is the hallmark of won positions; once you know how to win them, it doesn't matter how strong the opponent is. Won positions are won even against perfect play. TSCP beats Stockfish 100% of the time, in KQK. Even if Stockfish uses EGT.

In this case the knowledge is that you should not allow Q for 2N trading. Normal engines, including Komodo and Stockfish, might not know this, and frequently bungle the won game, even losing it. QueeNy has Q=9.5 and N=5, which makes it value 2 Knights above a Queen, and thus avoid conversion to 2Q vs 5N. Given enough depth to recognize the tactical threats for forcing such trades, it will be able to avoid them. Seven Knights is enough to keep all Knights protected twice virtually all the time. N=5 might even be too low, as it now would allow Q for 2N+P trades. This would only hurt when the opponent actively seeks such trades, though, which is probably not the case for Komodo.

Bishops cannot cooperate this well: they cannot protect each other when on different color, and they strongly hamper each other's mobility when the do protect each other. Knights and Bishops also do not mix well, because the protection they can offer is not mutual. So any Knight->Bishop substitution strongly weakens the side with the minors, and results of such games prove nothing about the pure Q vs N case.

So QueeNy is not "the measure of all things", but it does beat the top engines with the Knights in this position, with a much higher score than other top engines would achieve against the same opponent. This was tested with thousands of games. And that is the definition of "being better". Despite their strong search, most top engines just lack the evaluation knowledge to play this successfully. Better search is of no use if a poor evaluation doesn't let you search for the right thing.

All you do is offer (completely uninformed) talk, and, as they say, talk "cuts no wood". It is game results that define reality, and any reasoning that denies reality is know as 'delusion'...
you are completely mad.(yes, it is you who called me deluded first)

I do not know when you played Queeny vs SF for the last time, maybe 5? years ago, but current SF will smash the hell out of Queeny, at least what concerns the 2Qs vs 5Ns imbalance, with all pawns on the board.

you claim this is won or very close to being won for the knights too, but Queeny will lose 30-70 or so against SF, please check again with latest SF, I already did that and know what I am talking about.

the 2 queens win vs the 5 knights.

it is not true that bishops would perform worse than knights, it is quite the opposite: for example, 7 bishops quite convincingly beat 7 knights, I checked that too in the past, you might like to do your homework too.

I think it is reasonable to measure the effect of a mix of minors, to know how in general minors relate to queens; why would we like to know how 7 knights relate to 3 queens?

what purpose serves that and how frequently does this appear in standard chess games?

when was the last time you saw such a game in TCEC?

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 1:24 pm

I would have tested 2Qs vs 5 Ns, but for that, while keeping all pawns at the same time, I have to resort to unpleasant guis like arena, which I would not quite like to do.

Fritz does not accept the imbalance, anyway.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 1:27 pm

Uri Blass wrote:
hgm wrote:Yes, QueeNy is about 1000 Elo weaker than Stockfish and Komodo (when run with normal piece values).

The point, however, is that it beats Komodo and Stockfish when playing with the Knights in this position. This is the hallmark of won positions; once you know how to win them, it doesn't matter how strong the opponent is. Won positions are won even against perfect play. TSCP beats Stockfish 100% of the time, in KQK. Even if Stockfish uses EGT.

In this case the knowledge is that you should not allow Q for 2N trading. Normal engines, including Komodo and Stockfish, might not know this, and frequently bungle the won game, even losing it. QueeNy has Q=9.5 and N=5, which makes it value 2 Knights above a Queen, and thus avoid conversion to 2Q vs 5N. Given enough depth to recognize the tactical threats for forcing such trades, it will be able to avoid them. Seven Knights is enough to keep all Knights protected twice virtually all the time. N=5 might even be too low, as it now would allow Q for 2N+P trades. This would only hurt when the opponent actively seeks such trades, though, which is probably not the case for Komodo.

Bishops cannot cooperate this well: they cannot protect each other when on different color, and they strongly hamper each other's mobility when the do protect each other. Knights and Bishops also do not mix well, because the protection they can offer is not mutual. So any Knight->Bishop substitution strongly weakens the side with the minors, and results of such games prove nothing about the pure Q vs N case.

So QueeNy is not "the measure of all things", but it does beat the top engines with the Knights in this position, with a much higher score than other top engines would achieve against the same opponent. This was tested with thousands of games. And that is the definition of "being better". Despite their strong search, most top engines just lack the evaluation knowledge to play this successfully. Better search is of no use if a poor evaluation doesn't let you search for the right thing.

All you do is offer (completely uninformed) talk, and, as they say, talk "cuts no wood". It is game results that define reality, and any reasoning that denies reality is know as 'delusion'...
Better search can help even if you have worse evaluation because chess is not an evaluation contest but the question is if the search of chess programs is really better in the relevant position.

Can latest stockfish beat programs that are 300 or 400 elo weaker in normal chess in the relevant position and also do not have specific knowledge for queens against knights thanks to better search?

The only way to know is by testing and
I do not know because I did not test it.
but I did, and 2Qs win more or less convincingly against a mix of 5 minors(why should I test against 5 knights, when knights are weaker than bishops, and most guis do not accept similar oddities anyway)

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 1:33 pm

Guenther wrote:
Lyudmil Tsvetkov wrote:
Guenther wrote:
Lyudmil Tsvetkov wrote: so, you really suggest Queeny is stronger than SF and Komodo?
Yes for certain imbalances, but you know this already since 4 years,
see the link I gave before.
that is simply not true.

no one ever demonstrated Queeny is better than either SF or Komodo in such imbalances.

it is very much the other way around, of course, as SF finds tactical queen moves Queeny does not.

with its poor search, Queeny can do almost nothing the right way, therefore, by playing games with it you might get the wrong conclusions.

only thing that has been demonstrated is that Queeny evaluates the position more relevantly, but that has nothing to do with playing it also better.
Either you have a problem with your memory or you are not the same Tsvetkov as 4 years before.
Of course other reasons could be: plain trolling / insane?
All this was demonstrated in lengthy exercises especially for a certain L. Tsvetkov in 2013.

http://talkchess.com/forum/viewtopic.php?t=49857

http://talkchess.com/forum/viewtopic.php?t=49808
precisely, I demonstrated many ways to beat Queeny, taking the queen side, but you seem to only read Harm's messages.

besides, sometimes, engine tests with such positions might lead to false positives, precisely because one of the opponents, the one unaware of the imbalance intricacies, does not quite handle the position the right way.

it is more or less obvious 2 Qs are stronger than a mix of 5 minors and stronger or equal than a mix of 7 minors.

I still have not investigated with a renewed perspective, 4 years on, the specific queen vs only-knights imbalance, will do that at the first available moment, or, even better, when Fritz starts allowing such mad fens.

I do not know why are you talking, when you yourself did not test?

User avatar
Guenther
Posts: 2246
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Contact:

Re: Need material for a speech

Post by Guenther » Mon Jul 17, 2017 1:44 pm

Lyudmil Tsvetkov wrote:
Guenther wrote:
Lyudmil Tsvetkov wrote:
Guenther wrote:
Lyudmil Tsvetkov wrote: so, you really suggest Queeny is stronger than SF and Komodo?
Yes for certain imbalances, but you know this already since 4 years,
see the link I gave before.
that is simply not true.

no one ever demonstrated Queeny is better than either SF or Komodo in such imbalances.

it is very much the other way around, of course, as SF finds tactical queen moves Queeny does not.

with its poor search, Queeny can do almost nothing the right way, therefore, by playing games with it you might get the wrong conclusions.

only thing that has been demonstrated is that Queeny evaluates the position more relevantly, but that has nothing to do with playing it also better.
Either you have a problem with your memory or you are not the same Tsvetkov as 4 years before.
Of course other reasons could be: plain trolling / insane?
All this was demonstrated in lengthy exercises especially for a certain L. Tsvetkov in 2013.

http://talkchess.com/forum/viewtopic.php?t=49857

http://talkchess.com/forum/viewtopic.php?t=49808
precisely, I demonstrated many ways to beat Queeny, taking the queen side, but you seem to only read Harm's messages.

besides, sometimes, engine tests with such positions might lead to false positives, precisely because one of the opponents, the one unaware of the imbalance intricacies, does not quite handle the position the right way.

it is more or less obvious 2 Qs are stronger than a mix of 5 minors and stronger or equal than a mix of 7 minors.

I still have not investigated with a renewed perspective, 4 years on, the specific queen vs only-knights imbalance, will do that at the first available moment, or, even better, when Fritz starts allowing such mad fens.

I do not know why are you talking, when you yourself did not test?
The thread started about 3Q vs. 7N, but we know already that if you lose an argument you start modifying the topic and pretend your topic is suddenly the true one until your answer fits...
Guenther Simon
http://rwbc-chess.de/

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Need material for a speech

Post by Lyudmil Tsvetkov » Mon Jul 17, 2017 2:55 pm

tested 2Qs vs 5Ns with the most reasonable position acceptable by Fritz:

[d]1nnnknn1/p1pp1pp1/8/8/8/8/P1PP1PP1/2QQK3 w - - 0 1

as expected, queen side won, +9 -3 =8

larger pawn span with all pawns on the board will only favour queen side, of course.

conclusions(to the attention of Harm, Guenther, and also Eelco):
- queen side easily wins(much opposite to what Harm claims)
- queen side performs better against 5Ns than against a mix of 5 minors, as also expected by me, so Harm's claims are basically just blabbering

I also expect the queen side to perform worst against 5 bishops.

Harm tested Queeny from 5 or 10 years ago vs SF from 5 or 10 years ago, hence the results.

He also tested 2 different engines, one knowing about the specific imbalance, and the other one not.

in that case, the results might have been due to 2 reasons:
1. Queeny plays better than SF the specific imbalance, or
2. the position is objectively won for the knight side

of course, in our case, the culprit is number 1.

So, Harm leap-jumped to the wrong conclusion, and many other gullible people promptly believed.

only way to test the imbalance is Queeny against Queeny, or SF vs SF.

if SF and Komodo knew about the imbalance, they would easily crush Queeny 90-10 of course, with both colours.

I have no doubts now at all the 2Qs vs 5Ns imbalance, with all pawns on the board, is won for the queen side.

no way to reasonably check the larger imbalance with 7 knights, but am certain it is won or close to won for the queen side too.

Sometimes I wish my interlocutors knew what they are talking about, at least from time to time.

PS. under standard conditions queen gets some 60-80cps penalty vs 3 minors.
How could this rise to over 300cps by just adding another queen and 2 minors?

User avatar
hgm
Posts: 22195
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Contact:

Re: Need material for a speech

Post by hgm » Mon Jul 17, 2017 3:19 pm

Well, as Lyudmil has a perfect track record of never having claimed anything about Chess that was even remotely true, we now even know for certain that 5 Knights beat the hell out of 2 Queens, which I so far had never tested. (Because with good play 7 Knights vs 3 Queens can never convert to it anyway.)

Thanks, Lyudmil, for this final conformation. :lol:

Post Reply