@EmaciatedSpaniard stockfish needs competition. Send me a game when you beat :)
Its not SF14+. I am not referring to starting a game with SF14+ on lichess via the "Play with the computer" button. That is very strong.
I am talking about the "play with computer" button in puzzles of endgames. I think that's part of the confusion here. That button on puzzles should probably be renamed to "practice with computer". I think it is the same as Study chapters which have Analysis mode set to "Practice with computer".
In endgames it does not play to preserve its best chances to obtain mate (or avoid mate if its in a losing position). It does nonsensical things like abandoning pawns that could be promoted.
Try any one of the examples I have posted and you will find it is not playing the best moves. They are outright blunders.
I am talking about the "play with computer" button in puzzles of endgames. I think that's part of the confusion here. That button on puzzles should probably be renamed to "practice with computer". I think it is the same as Study chapters which have Analysis mode set to "Practice with computer".
In endgames it does not play to preserve its best chances to obtain mate (or avoid mate if its in a losing position). It does nonsensical things like abandoning pawns that could be promoted.
Try any one of the examples I have posted and you will find it is not playing the best moves. They are outright blunders.
Hmmm, i have to revise my statement in my last post. If I go to the play main menu of lichess and click "play with the computer", select "play from position" and use the fen "8/8/8/8/4k3/8/3KP3/8 w - - 0 1" and play as black, i get the same behaviour. using level 8 of SF.
I guess all the play with computer options are using SF and its all related to what ProgrammerAngrim described.
Here is my game:
I guess all the play with computer options are using SF and its all related to what ProgrammerAngrim described.
Here is my game:
@EmaciatedSpaniard said in #20:
> I just started a study of Capablanca's classic endgame lessons (KP vs K, drawn) and discovered that "practice with computer" is even worse than i thought. Even when white still has a chance to promote a pawn if black makes an error, white throws the pawn in front of the black king and allows capture. This moves makes no sense even when consulting a tablebase does it? I mean white should make moves that preserve winning chances, shouldn't it?
>
This is a weakness of tablebases, they make no provision of enemies making errors. They make perfect moves assuming that the enemy is also perfect, but sometimes astonishingly dumb moves against an enemy who is flawed.
One issue that has come up on the programming discussion boards is what to do when a position is a "mate" in more than 50 moves without a conversion. In theory it will be a draw due to 50 move rule, but if you simply treat all such as a draw then the engine will make no distinction between a win in 51(which a flawed enemy will certainly lose, it only take any tiny error) and a loss in 51(which will be a draw, but look really bad to observers). What I do in my engine is when a position is in endgame tables as a draw, I delete all losing moves from it's list, then turn off endgame tables and use the regular search to select between the remaining drawing moves. I don't think that stockfish does this though, since it won't give any elo gain against an opponent with the same size endgame tables.
> I just started a study of Capablanca's classic endgame lessons (KP vs K, drawn) and discovered that "practice with computer" is even worse than i thought. Even when white still has a chance to promote a pawn if black makes an error, white throws the pawn in front of the black king and allows capture. This moves makes no sense even when consulting a tablebase does it? I mean white should make moves that preserve winning chances, shouldn't it?
>
This is a weakness of tablebases, they make no provision of enemies making errors. They make perfect moves assuming that the enemy is also perfect, but sometimes astonishingly dumb moves against an enemy who is flawed.
One issue that has come up on the programming discussion boards is what to do when a position is a "mate" in more than 50 moves without a conversion. In theory it will be a draw due to 50 move rule, but if you simply treat all such as a draw then the engine will make no distinction between a win in 51(which a flawed enemy will certainly lose, it only take any tiny error) and a loss in 51(which will be a draw, but look really bad to observers). What I do in my engine is when a position is in endgame tables as a draw, I delete all losing moves from it's list, then turn off endgame tables and use the regular search to select between the remaining drawing moves. I don't think that stockfish does this though, since it won't give any elo gain against an opponent with the same size endgame tables.
Very interesting. I like your approach of ranking the remaining nonlosing lines using regular search. Even if this gains no ELO, it can't hurt to include it - so i don't see why SF doesn't do the same.
But the discussion generally points out an issue with using SF as a trainer. Applying the 50 move draw rule to a situation where you are just trying to train a standard procedure to draw (or take advantage of mistakes which convert a draw into a win) doesn't really work.
But the discussion generally points out an issue with using SF as a trainer. Applying the 50 move draw rule to a situation where you are just trying to train a standard procedure to draw (or take advantage of mistakes which convert a draw into a win) doesn't really work.
Hm ... for practicing on our level it would be better to use no tablebase at all if that is the reason.
I still wonder: why not use the same rules as in "analysis board" mode. As I and others showed in that older thread mentioned: In practice mode the opponent makes different moves than those suggested in the analysis mode. Why not just apply the same algorithm to choose a move? I understand the engine's assumption on optimal play of both sides. But the analysis arrays surely show better defense.
I still wonder: why not use the same rules as in "analysis board" mode. As I and others showed in that older thread mentioned: In practice mode the opponent makes different moves than those suggested in the analysis mode. Why not just apply the same algorithm to choose a move? I understand the engine's assumption on optimal play of both sides. But the analysis arrays surely show better defense.
Is this something that got attention by developers? Its unfortunate. Many endgames require a very methodical approach to win them and train with computer would be a good way to train those, but i find its pretty useless for such endgame situations.
SF doesn't put up much of a fight if the situation is losing which is really too bad.
SF doesn't put up much of a fight if the situation is losing which is really too bad.
@EmaciatedSpaniard said in #27:
> SF doesn't put up much of a fight if the situation is losing which is really too bad.
It is not a matter of Stockfish! Trying to bump your thread no matter what answers you get, are you?
If you use the analysis board SF shows the best move of your opponent, which in that late stadium of the game differs from the move "practice with the computer" suggests.
> SF doesn't put up much of a fight if the situation is losing which is really too bad.
It is not a matter of Stockfish! Trying to bump your thread no matter what answers you get, are you?
If you use the analysis board SF shows the best move of your opponent, which in that late stadium of the game differs from the move "practice with the computer" suggests.
This topic has been archived and can no longer be replied to.