The neural network of the Stockfish chess engine

Re:The neural network of the Stockfish chess engine by dan-robertson on Wednesday January 13, 2021 @04:25AM (#25759906)

Chess is not a solved game (though I think the stalemate rules make it finite, there are far too many positions). All computer chess programs have to try to determine if positions will be good without playing out all possible ways the game could go. Usually this is some combination of search strategies and scoring positions (the article describes how stockfish scores positions).

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by V-2 on Wednesday January 13, 2021 @04:45AM (#25760037)

I agree it's finite, but why do you think it's specifically owed to the stalemate rule, of all things?

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by elcomet on Wednesday January 13, 2021 @04:49AM (#25760060)

I believe it's the repetition rule that makes the game finite, not the stalemate rule, since number of positions is finite.

There is also the 50-move rule of course which limits more drastically the length of a game.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by V-2 on Wednesday January 13, 2021 @04:54AM (#25760089)

Right, that would make more sense to me. Although it should be stressed that neither threefold repetition nor the 50-move rule work automatically. They need to be claimed by one of the players. If neither player does, the game can go on.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by jstanley on Wednesday January 13, 2021 @05:42AM (#25760377)

The 75-move rule does work automatically though, so the game is still finite.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by V-2 on Wednesday January 13, 2021 @09:45AM (#25762554)

Frankly, I wasn't aware of the 75-move rule! This indeed caps the length of possible games.

The threefold repetition rule doesn't have an equivalent (there's no, say, "fivefold repetition" draw that could be enforced despite the players' will), but it's really irrelevant from the perspective of solving chess, because there's no point in analyzing what happens if game goes through the same loop 1, 10 or 100 times.

And the 75-move rule sets a hard limit anyway, so you couldn't loop infinitely, even if it mattered.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by harryposner on Wednesday January 13, 2021 @11:27AM (#25764192)

> there's no, say, "fivefold repetition" draw that could be enforced despite the players' will

There is exactly such a rule. See the FIDE Laws of Chess [0], rule 9.6.1.

[0] https://handbook.fide.com/chapter/E012018

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by anandoza on Wednesday January 13, 2021 @11:06AM (#25763846)

Lichess.org actually does have a 5fold repetition rule that applies without players claiming it -- https://lichess.org/faq#threefold

Kind of funny that your hypothetical example was exactly reality in this case.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by dan-robertson on Wednesday January 13, 2021 @06:32AM (#25760785)

I think this is what I was referring to.

[ Parent | on HN ]

Re:The neural network of the Stockfish chess engine by V-2 on Wednesday January 13, 2021 @09:24AM (#25762269)

OK, so that's not a stalemate. A stalemate is a situation in which one of the players has no legal moves (but isn't in check, which would be checkmate).

It's a typical mistake to equate chess draws with a stalemate, possibly due to the broader, idiomatic meaning of "stalemate" in common speech.

[ Parent | on HN ]