Drawkiller Openings Project

[Michel]

It should also be pointed out that there is nothing inherently wrong with unbalanced openings as long as one plays games in pairs with reversed colors and furthermore the pentanomial model is used to compute the variance. Under the BayesElo model (using the advantage parameter to model bias) unbalanced openings lead to an increase in sensitivity in some scenarios as was for example discussed here

http://talkchess.com/forum3/viewtopic.php?f=7&t=61245

Needless to say that one should take all elo models with a grain of salt.

[Dann Corbit]

Re:"It should also be pointed out that there is nothing inherently wrong with unbalanced openings as long as one plays games in pairs with reversed colors and furthermore the pentanomial model is used to compute the variance."

Consider the extreme example of an opening book that contains:
1. f3 e5 2. g4 *
It will produce a win for white for both sides. Fine if the opponents are equal. It produces an absurd result if there is a significant difference in strength.

The outcome of using that opening bends the results because this one trial indicates that the two opponents are equal after both games are played, which is only true when the opponents are actually equal in strength.

In short, I do not believe that highly polar openings do not damage the correctness of the results, even with both players starting as white and black, unless the players have the exact same strength.

Any badly biased opening will introduce some kind of bias in the outcome of the calculation, giving an increase in apparent strength to the weaker side.

[Michel]

Re:"It should also be pointed out that there is nothing inherently wrong with unbalanced openings as long as one plays games in pairs with reversed colors and furthermore the pentanomial model is used to compute the variance."

Consider the extreme example of an opening book that contains:
1. f3 e5 2. g4 *
It will produce a win for white for both sides. Fine if the opponents are equal. It produces an absurd result if there is a significant difference in strength

Of course that would be stupid. I never said the bias should be chosen to be ridiculously large. According to Kai's computations the optimal bias should be comparable to the value of draw_elo (under the BayesElo model). This has the effect that one engine is fighting for a draw and the other engine is fighting for a win.

[Dann Corbit]

Re:"It should also be pointed out that there is nothing inherently wrong with unbalanced openings as long as one plays games in pairs with reversed colors and furthermore the pentanomial model is used to compute the variance."

Consider the extreme example of an opening book that contains:
1. f3 e5 2. g4 *
It will produce a win for white for both sides. Fine if the opponents are equal. It produces an absurd result if there is a significant difference in strength

Of course that would be stupid. I never said the bias should be chosen to be ridiculously large. According to Kai's computations the optimal bias should be comparable to the value of draw_elo (under the BayesElo model). This has the effect that one engine is fighting for a draw and the other engine is fighting for a win.

I think I agree with you then. I was just pointing out that large (probably inadvertent) bias for some openings does introduce a defect.

I think an equal or greater mistake is created by choosing openings from drawn games (or mostly from drawn games). Quite often, the reason that the game is drawn from that position is that a draw is the best outcome for that ending book position. And so this also introduces a bias in that engines that would not draw from most positions will be forced to draw against weaker opponents.

Now, the hard part is figuring out which openings are hard-wired big advantages and which are hard-wired to draw verses those openings which have a lot of variance in possible outcomes depending upon how they are played. It sounds trivial but I do not think it is trivial.

[Nordlandia]

A possible "Transcendental Chess" position. A quick-fix for the silly opening moves is to give cutechess transcendental chess support and find the two out of 8294400 possible starting position. Finding the needle in the haystack. What does Stefan Pohl say about this. Also the pawn at the starting square looks much better.

[d]knbrqbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQRBNK w - - 0 11

[pohl4711]

What I say about this? That makes no sense, because it is not classical chess. That position cannot be reached with legal moves from normal starting position of chess.
That is one of the huge advantages of Drawkiller, that all endpositions of Drawkiller are reached by playing legal moves from the starting position of classical chess.
So each chess engine on the planet can play with Drawkiller openings and in each GUI, opening books can be built out of Drawkiller. That is impossible with Chess960 or such "Trancendental Chess"-nonsense. Drawkiller is classical chess. Chess960 or other chess-variants are not.

Stefan (SPCC)

[Nordlandia]

Drawkiller is classical chess. Chess960 or other chess-variants are not. Stefan (SPCC)

The reason why I address this issue is because many find the opening move silly (they criticise the opening rather than the starting position with the kings in the corner). Apparently few pay more attention to the necessary moves to reach the corner for the kings, so they disregard it simply because of the weird opening. I talked with few tcec chat members yesterday and they responded this way, arguing no GM would play something like this but i simply replied that does it matter what GM's would play when it is engines that play the games.

[pohl4711]

Of course, the Drawkiller opening moves are silly. Their only purpose is to reach a (non silly) Drawkiller-endposition, which is the starting position for more interesting and less drawish engine-chess.
If anybody rejects Drawkiller because of it's silly moves, he doesnt understand the whole concept. The concept is, to let engines start playing out of a non-drawish starting-position, which is not a chess-variant, but classical chess. Not more, not.less.
And all Drawkiller Openings-Sets are as EPD-files in the download-package, which means a collection of FEN-codes of all Drawkiller endpositions. Without any (silly) moves. Anybody, who doesnt like the moves, can use these EPD-files...

[Ovyron]

If anybody rejects Drawkiller because of it's silly moves, he doesnt understand the whole concept. The concept is, to let engines start playing out of a non-drawish starting-position, which is not a chess-variant, but classical chess.

The problem is not that the moves are silly, the problem is that they're game losing blunders, and nobody in their right mind would agree to play them. And if they did, nothing in the rules say I should comply with my promise of play those moves in the end (say, we "agree" to do it, then as we're playing the moves I deviate by surprise and beat you, because your moves are bad.)

Nothing in classical chess forces both players to agree to reach some position and do it, in fact, some tournaments forbid players to arrange playing a drawn variation in advance (they can still do it, of course, buy they can't just play the same drawn variation, as that'd raise suspicion, anyway.)

In classical chess you're supposed to play to win, or at least find the best moves in each postition, and play it. These silly moves are against the spirit of the game.

Now, I get the concept, but the concept doesn't require that these possitions are reachable from chess's opening position, because how you get there is irrelevant, because there positions are not reachable by opponents playing the game (unless they play game losing blunders for no reason, but the whole point of chess is to avoid playing blunders, so a player reaching these positions is expected to find improvements and punish the silly moves instead.)

How you got there is arbitrary, so just add to the EPD the best positions to serve this purpose, nobody should care if it's reachable from classical chess, because reaching it from classical chess is impossible anyway.

[pohl4711]

Of course, nobody would play those moves in a game of chess. But the point is, that all chessengines and all chessGUIs can handle Drawkiller and the moves. Because they are classical chess moves (even no engine would play them on its own). And that is a huge advantage compared to any chess-variant, like Chess960 etc., which are having differences in rules or starting positions (compared to classical chess).
The primary goal of Drawkiller is, to prevent computerchess from dying the death-by-draw in the future. And for that, it is very important, that each engine and each GUI on the planet can handle Drawkiller...