Alpha Zero's 21. Bg5! revisited

[mmt]

It's a bad test position if you only consider one move to be the right solution (which is the case for most test positions and was an assumption for this one). If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?

[peter]

It's a bad test position if you only consider one move to be the right solution (which is the case for most test positions and was an assumption for this one). If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?

You don't want to understand my point, do you?
I don't doubt b4 to win, yet I think Bg5 clearly better.
If you don't do so, you'll have to prove b4 winning in less moves on corr.- games of some quality. I know, you don't say so neither, but a third choice was only this one, simply to say, you don't know, which one is better.
Is it that, what you mean? Guess not neither, or am I really wrong about that main question?

If you don't want to take Bg5 as single best move in test suite judging single best moves only, don't do so, neither would I.
But if want to take it as a test, which engine gives which one of the both candidates sooner the more correct eval and output-line, just do so, and I wont' argue against that, why would you?
If it's not a good test position for "your kind of testing" try another kind of testing (which you do all the time as for positional testing and or eng-eng-match-testing and quoting evals to argue with them) or forget about the position fully, no matter to me neither, you see?

Still not quite sure, what you're after at regards

[mmt]

If you don't do so, you'll have to prove b4 winning in less moves on corr.- games of some quality.

No, I don't have to because that's not what makes a move better when playing a normal game of chess where winning is all that matters. All winning moves are equally good. I'll grant you that there is a tiny chance that Bg5 is winning but b4 is not but there is also a tiny chance that b4 is winning and Bg5 is not.

It's very simple what I'm saying: if you're doing any sort of testing where you want to determine if one engine (version) is better than another, this is not a good test position. That's what people were testing. If you want to do other sorts of testing for most beautiful moves or something then fine.

[peter]

If you don't do so, you'll have to prove b4 winning in less moves on corr.- games of some quality.

No, I don't have to because that's not what makes a move better when playing a normal game of chess where winning is all that matters. All winning moves are equally good. I'll grant you that there is a tiny chance that Bg5 is winning but b4 is not but there is also a tiny chance that b4 is winning and Bg5 is not.

It's very simple what I'm saying: if you're doing any sort of testing where you want to determine if one engine (version) is better than another, this is not a good test position. That's what people were testing. If you want to do other sorts of testing for most beautiful moves or something then fine.

You insist on "your way of testing" is the only one that matters, but you don't even say, what your kind of testing means.
Guess you mean the way by "single best moves", indeed meaning single one game changing moves, but those are at the utmost the best ones for single best move tactical test suites.
What would you give opening positions for test-matches any other names then test positions? ("Opening position" isn't the counting criterium here neither, opening positions can be bad test positions, as well for game playing as for positional testing too, can't they, so the point of a good test positions isn't "opening position" as well as it isn't "single best move" position, is it?)
They are most of the time of very similar height of rather equal eval, you don't know, which one's better for sure then the other one for which side, and yet they are the classical test postions at all, aren't they? Whether you let them be played out or judge them in positional test suites, who cares?

The one and only criterium, that really doesn't count for test positions, is "beauty", which you always start to talk about being asked several times already, which one of the two moves here in discussion you think to be the better one. It's the only one criterium of no meaning at all, cause it's the only one you can't measure at all. You really think Bg5 only more beautiful and not better in any measurable way? Well, than don't come along with 3.59 for the one and 4.59 for the other one to prove, they are both equal. If they are, why then the difference in eval, if the difference doesn't matter (to you), why than does it prove from your side, the moves are equal?

So, if you really think b4 exactly as good as Bg5, it's on your side to prove it, if it doesn't matter to you, what are you doing so much writing about that (ok, that's the same question as for my part of the latest pages ) and if you just want to give your exact definition of a good testing position than you have to stick to a certain kind of testing of your choice, if I may help you out with this, to show you, I really got your point about that since quite a while, I alreaedy gave it (your definition) too: single_one_game_changing_winning_move_postion.

Once again, it's not the only one definition of a testing position, not even the only one of a tactical test position, but I take notice, if it's that, what you want, that it's the only one definition to you.
Most of the testers prefer equal test positions, not knowing the outcome of play outs from them and anything else is just "position dependent" testing for those testers, no problem to me neither, cause to me any kind of testing in any way is always position dependent anyhow.

Guess I've really said enough about so little and so well known content.
And as for the position in question: Bg5 is better than b4.
Period regards

[mmt]

Well, than don't come along with 3.59 for the one and 4.59 for the other one to prove, they are both equal. If they are, why then the difference in eval, if the difference doesn't matter (to you), why than does it prove from your side, the moves are equal?

You didn't answer when I asked "If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?" I asked this because if you answer, you should see what these scores really mean and see why it makes no sense to try to distinguish them from the testing perspective. If b4 is winning, like SF is almost certain it is, then if you penalize a program for choosing it you're doing it wrong.

[peter]

Well, than don't come along with 3.59 for the one and 4.59 for the other one to prove, they are both equal. If they are, why then the difference in eval, if the difference doesn't matter (to you), why than does it prove from your side, the moves are equal?

You didn't answer when I asked "If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?" I asked this because if you answer, you should see what these scores really mean and see why it makes no sense to try to distinguish them from the testing perspective. If b4 is winning, like SF is almost certain it is, then if you penalize a program for choosing it you're doing it wrong.

I said several times I see b4 as an almost certain winner too (would mainly depend on players, humans, engines, which ones, hardware-tc), it's you, who still didn't answer my as often as this repeated question, which of the two moves you think to be better.

To be even more clear once in a while again, I expect Bg5 to get a shorter win with blunder- free play then b4, eg. in corr.- games of some quality, if you want to find out, if I'm right and for how much moves the difference might count that way of testing, let's have two games togehter.

You even wrote a mate solving program of your own, you know dtm matters, you give exact examples of evals you seem to trust in, so you too admit also,that eval matters (of course only regarding the corresponding output- lines and the engines you're testing as for their evals).

Eval matters in the position given even more than in others regarding the context we are talking about here, NNUE- eval and classical SF- eval, so did you get that too? Who said, Bg5 was the only one game changing winning move? I didn't.
So, let's play it out or trust the evals, you seem to trust in, you say 1 pawn difference isn't enough for you for a test position of your criteria, I say to me it is, not as for a single best move tactical test suite, but to judge output- lines and evals accordingly with it. A Pawn is a Pawn is a Pawn, if it isn't more or less then a Pawn. I'm interested in single Pawns in chess, even in lesser eval- differences than single Pawns, of course always only position- dependent. From which eval- differences eval starts to matter to you? From two Pawns, a Knight, a King? A King is nothing without a kingdom, not in a certain position on the board, a King isn't a King in chess, you see?

It doesn't make sense to you to distinguish the two moves, because none of them is a single one game changing, so you don't like to call it a test position. I tell you there are other ways of test positions than single game changins ones, you know so, but you won't call that the way of testing you mean.
Call it however you want, don't use it for testing, yet be so kind and don't try to forbid me doing so, will you?

Still happy to have chances enough to practice my English with you, please don't stop now regards

[mwyoung]

Well, than don't come along with 3.59 for the one and 4.59 for the other one to prove, they are both equal. If they are, why then the difference in eval, if the difference doesn't matter (to you), why than does it prove from your side, the moves are equal?

You didn't answer when I asked "If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?" I asked this because if you answer, you should see what these scores really mean and see why it makes no sense to try to distinguish them from the testing perspective. If b4 is winning, like SF is almost certain it is, then if you penalize a program for choosing it you're doing it wrong.

I said several times I see b4 as an almost certain winner too (would mainly depend on players, humans, engines, which ones, hardware-tc), it's you, who still didn't answer my as often as this repeated question, which of the two moves you think to be better.

I will answer this question.

If Bg5 wins, and b4 also wins. Then they are both equal. A engine evaluation is meaningless.
Are you trying to claim one may win faster, and then it still does not hold water. Chess gives no extra credit for shorter wins.
All positions are scored 1, =, -1. That is the only true evaluation in chess.

"If you assume that +4.6 and +3.6 are valid evals from SF at depth 66, what average results do you expect after Bg5 and b4 for 100 correspondence games each starting at these positions?

I did not see the engine score this as a win, loss, or draw here. So that is just funny in itself that you want to argue this point. 3.6 vs 4.6 when I have seen scores much higher then this be draws from the best engines in the world.

[peter]

If Bg5 wins, and b4 also wins. Then they are both equal. A engine evaluation is meaningless.
Are you trying to claim one may win faster, and then it still does not hold water. Chess gives no extra credit for shorter wins.

It would have been enough to write, that you are of mmt's opinion, I answered to these his arguments often enough I guess, not having to start same discussion all over again, just because it's you now coming along with the same tune.

Of course chess gives extra credit for shorter wins, it's called distance to mate.
Whole lot of composed problems deal with this by you ignored position- immanent feature. If you can't tell distance to mate in a position of interest doesn't mean it doesn't count, you can ignore it, but you can ignore whole game of chess too, it doesn't qualify you as a better player ignoring the only one really exact eval (dtm). Any other evals given by engines you of course don't have to take notice of neither, but why then argue with evals like 4.59 and 3.59, taking them as a prove for you (if you keep on sticking to mmt's arguments) that both moves would have all the same outcomes in 100 corr.-games? Who tells you, if you don't believe in engine- evals, that distance to mate wil be the same or about the same for both moves?
If you don't take notice of engine- evals at all, what are your personal human evals in the positions of interest?

Before you continue repeating mmt's pov onwards, would you be able to say, which one move is better for your personal pov?

If you are, would you tell us so, which one it was (at least in this regard maybe willing and able to give more information than he did till now) and what's your way to find out?
Dont you think you'd have to have any way to find out, if you don't have any here, what's your way of playing games of chess in other one positions?
Or is this the only one position you can't decide for a certain move?
You have to choose a move in a game of chess, when it's your turn, you can't tell the opponent, you can't decide, because there are too many moves seeming equal to you. Not true neither, you can tell so, but nobody will be interested in your reasons not to play, you'd simply lose on time.

Repeating regards

[mwyoung]

If Bg5 wins, and b4 also wins. Then they are both equal. A engine evaluation is meaningless.
Are you trying to claim one may win faster, and then it still does not hold water. Chess gives no extra credit for shorter wins.

It would have been enough to write, that you are of mmt's opinion, I answered to these his arguments often enough I guess, not having to start same discussion all over again, just because it's you now coming along with the same tune.

Of course chess gives extra credit for shorter wins, it's called distance to mate.
Whole lot of composed problems deal with this by you ignored position- immanent feature. If you can't tell distance to mate in a position of interest doesn't mean it doesn't count, you can ignore it, but you can ignore whole game of chess too, it doesn't qualify you as a better player ignoring the only one really exakt eval. Any other evals given by engines you of course don't have to take notice of neither, but why then argue with evals like 4.59 and 3.59, taking them as a prove for you (if you keep on sticking to mmt's arguments) that both moves would have all the same outcomes in 100 corr.-games? Who tells you, if you don't believe in engine- evals, that distance to mate wil be the same or about the same for both moves?
If you don't take notice of engine- evals at all, what are your personal human evals in the positions of interest?

Before you continue repeating mmt's pov onwards, would you be able to say, which one move is better for you personal pov?
If you are, would you tell us, which one it was and what's your way to find out?
Dont you think you'd have to have any way to find out, if you don't have any here, what's your way of playing games of chess in other one positions?
Or is this the only one position you can't decide for a certain move?
You have to choose a move in a game of chess, when it's your turn, you can't tell the opponent, you can't decide, because there are too many moves seeming equal to you. Not true neihter, you can tell him, but he won't be interested in your reasons not to play, you'd simply lose on time.

Repeating regards

Nice try, but I am your huckleberry!

"Whole lot of composed problems deal with this by you ignored position- immanent feature. If you can't tell distance to mate in a position of interest doesn't mean it doesn't count, you can ignore it, but you can ignore whole game of chess too, it doesn't qualify you as a better player ignoring the only one really exakt eval."

You ignore all of what I have said.

A composed problem has only "1" solution.

It does not matter what has come before, it alone matters what the position currently is a Win, Loss, or Draw.

[peter]

You ignore all of what I have said.

You really talk about me?

Which one move is better to you in the given position?
Doesn't it matter to you?
Why than all that fuzz?

And not to ignore your other one "argument", only composed problems would have only one soution (which isn't even true for composed or other problems neither, you just call it dual then, if there are two solutions, yet you have to judge those then too, if they are really equal, if they are Major or Minor Duals e.g), that depends on your definition of "solution".
I tell you once again in case you didn't read or get this one I said before (or ignored it), any position in chess, you want to play onwards from, has to have a single solution, if you want to play and if it's your turn. You have to decide for a single one move for your side, if you can't decide, you have to resign, not having found the "solution" of the position.
So you once in a while just have to decide, if you want to play or not. Lay your bets please or leave the table (board) for good for this time.

And now, my huckleberry friend (which you tell me to be, btw you know Moon River, don't you?), what's your next move regards