Queens vs Knights

[hgm]

I wonder why should there be engines showing score from the point of view of the side to move?

with SF showing score from white's perspective and Queeny showing score from the point of view of the side to move, it is almost impossible to make sense evaluation-wise what is happening.

Because that is the de-facto standard, for this format. Software processing the PGN will expect it, and flip signs when needed. E.g. in WinBoard's evaluation graph, where scoring is shown from white's POV.

[hgm]

Komodo 11.01 is a tougher nut (40 moves/5 minutes):

Well, looking at the game it seems QueeNy did itself in already in the early opening, probably because of the randomization. It almost immediately gives away a Pawn, for no apparent reason, and allows a Q-for-2N trade quickly after it. White cannot forcibly gain a Pawn so quickly in the initial position, as deep analysis will show; irrespective of how Komodo evaluates the imbalance, gobbling up a Pawn should always be good.

The advantage of the Knights is not enormous; delete one and the position is a hopeless loss. So the advantage is about 1-2 Pawns, and requires pretty accurate play to convert. Black cannot affort to waste any material. Especially not against an 800-Elo stronger player. Normally (i.e. without the evaluation adapted to Q-vs-N and the FIDE start position) even Knight odds would not be enough to overcome an 800-Elo gap. A certain minimum level of play is needed to avoid gross tactical blunders (which the opponent, searching some 10-20 plies deeper, would always see), and it is quite possible that at blitz speed QueeNy is still too weak to avoid them.

[hgm]

and the other 5 Queeny lost, right?

This was the only game I played (with SF 8), so far.

you seem not to dereading what I posted: you should be matching Queeny vs Queeny, or SF vs SF, to obtain reliable data about the position.

QueeNy vs QueeNy has already been done years ago, hundreds of games; The Knights always win, with an occasional draw due to needless conversion to KNNK, of which QueeNy is not aware that it is a draw. QueeNy was not changed, so there is no point in doing that again. It is Stockfish 8 that was never tried.

SF vs SF is pretty pointless, as obviously SF doesn't know how to play this. Its evaluation compells it to actually seek Q-for-2N trades with the Knights, and because of its superior position will be successful in that. Thus it will quickly turn a won position into a lost position. Games between clueless players tell you nothing about the position.

in a Queeny vs SF match all you is measuring is a looot of noise, namely superior evaluation for the specific position for one of the engines.

Not really. You are measuring how much better QueeNy handles the Knights than that Stockfish handles the Knights. You would never be able to measure the difference between two engines if you only played those against themselves.

So the proper procedure is:
First try different engines against each other, to see which does best with the Knights against the same opponents, and which one does best with the Queens. Ideally you would get an engine that performs better than any other both with the Knights and with the Queens, and then you would use that in self-play.

The problem is, that when QueeNy plays with the Knights at sufficiently long TC, it almost always wins. So you cannot really select the opponent that handles the Queens best based on that. But you can do 3 Queens vs 5 Knights, or handicap the Knights with Pawn odds, to get away from the 100% score, and use that to select the best Queen handler.

[Lyudmil Tsvetkov]

and the other 5 Queeny lost, right?

This was the only game I played (with SF 8), so far.

you seem not to dereading what I posted: you should be matching Queeny vs Queeny, or SF vs SF, to obtain reliable data about the position.

QueeNy vs QueeNy has already been done years ago, hundreds of games; The Knights always win, with an occasional draw due to needless conversion to KNNK, of which QueeNy is not aware that it is a draw. QueeNy was not changed, so there is no point in doing that again. It is Stockfish 8 that was never tried.

SF vs SF is pretty pointless, as obviously SF doesn't know how to play this. Its evaluation compells it to actually seek Q-for-2N trades with the Knights, and because of its superior position will be successful in that. Thus it will quickly turn a won position into a lost position. Games between clueless players tell you nothing about the position.

in a Queeny vs SF match all you is measuring is a looot of noise, namely superior evaluation for the specific position for one of the engines.

Not really. You are measuring how much better QueeNy handles the Knights than that Stockfish handles the Knights. You would never be able to measure the difference between two engines if you only played those against themselves.

So the proper procedure is:
First try different engines against each other, to see which does best with the Knights against the same opponents, and which one does best with the Queens. Ideally you would get an engine that performs better than any other both with the Knights and with the Queens, and then you would use that in self-play.

The problem is, that when QueeNy plays with the Knights at sufficiently long TC, it almost always wins. So you cannot really select the opponent that handles the Queens best based on that. But you can do 3 Queens vs 5 Knights, or handicap the Knights with Pawn odds, to get away from the 100% score, and use that to select the best Queen handler.

you don't know what you are talking.

2Qs vs 5Ns, with all pawns on the board, is easily won for the queen side.

3Qs vs 7Ns is, if not won, at least measurably better for the queen side.

maybe you played those Queeny games in the past millenium, but, due to its very poor search, such games would be indicative of nothing.

Queeny will be missing almost all tactical solutions, involving deep lines with checks, capturing, etc., and that is what queens do best.

my shootouts with SF and Komodo are quite convincing, queens easily win 2Qs vs 5Ns, and at least draw 3Qs vs 7Ns, but the queen side only misses early sacs, that would boost its score, because of lack of knowledge.

[hgm]

you don't know what you are talking.

2Qs vs 5Ns, with all pawns on the board, is easily won for the queen side.

It seems more that you don't know what I am talking. What's this obsession with 2Q vs 5N? I never claimed it was won for the Knight side, and whether it is or not is totally irrelevant, as 3Q vs 7N should never convert to it.

BTW, the "most reasonable position accepted by Fritz" that you used to test 2Q-vs-3N with (1nnnknn1/p1pp1pp1/8/8/8/8/P1PP1PP1/2QQK3 w - - 0 1) is ridiculously biased to favor the Queens: 1. Qb2 immediately forks a hanging Knight and a hanging Pawn, and thus gains a Pawn. If you use silly positions like that for testing, no wonder you never arrive at a correct conclusion...

3Qs vs 7Ns is, if not won, at least measurably better for the queen side.

Coming from you that makes it certain that it is lost for the Queens...

maybe you played those Queeny games in the past millenium, but, due to its very poor search, such games would be indicative of nothing.

Queeny will be missing almost all tactical solutions, involving deep lines with checks, capturing, etc., and that is what queens do best.

Apparently Stockfish 8 misses those too, because it still loses with the Queens against QueeNy...

my shootouts with SF and Komodo are quite convincing, queens easily win 2Qs vs 5Ns, and at least draw 3Qs vs 7Ns, but the queen side only misses early sacs, that would boost its score, because of lack of knowledge.

Doesn't mean a thing, if you cannot prove first that Komodo doesn't play like an idiot with the Knights. Beating an idiot is no great achievement, and doesn't prove his position was worse.

[Evert]

I wonder why should there be engines showing score from the point of view of the side to move?

For UCI this is easy: the spec says that theis is what they should do. For CECP it's not mandatory, but sufficiently standard that it's annoying when engines don't do it (*cough*crafty*cough*).

[hgm]

Indeed, that applies to the engines. But that doesn't mean, of course, that they have to be included that way in the PGN. It is the GUI that decides on the PGN format, not the engine.

[hgm]

Pawel has an elephantiasis correction for the queen in Rodent III:

If # of own queens > 1
then correction = -constant * (# of opp knights + # of opp bishops)

The default value of the constant is 4, which may be too low for this situation. I used 60 in the game above, though I have no idea if that was a good value.

This doesn't seem to work. I just played Rodent III against Stockfish 8, and they seem to agree that the score of the initial position is around +8 for the Queens.

[Evert]

Pawel has an elephantiasis correction for the queen in Rodent III:

If # of own queens > 1
then correction = -constant * (# of opp knights + # of opp bishops)

The default value of the constant is 4, which may be too low for this situation. I used 60 in the game above, though I have no idea if that was a good value.

I just had some fun adding the same term to Stockfish (with constant=60cp), which should be about right. I only had time to let it play a few games, but they were rather decisive. The queens are toast.

[Guenther]

with this one:

[d]nn1nkn1n/p3p2p/1n3n2/8/8/8/P3P2P/2QQKQ2 w - - 0 1

the closest possible in Fritz,
5 white wins, 5 black wins, 10 draws

not a shadow of knight superiority.

equal score, but more natural position with all pawns will only favour queens, of course, as pawns will restrain knights.

This opinion is also completely wrong. Basic chess knowledge already should tell you otherwise.
More pawns of course don't restrain the Knights. It's the opposite, the more pawns and the more closed the position the stronger are the Knights.
Also more pawns enable more pawn defenders for dangerous Knight outposts.
The Queens need space for their power and too much pawns prevent that.