Generally, a position can only be won or a draw. Any number for a won position is arbitrary.
@Sarg0n said in #11:
> Generally, a position can only be won or a draw. Any number for a won position is arbitrary.
The number is not arbitrary, it is a measure of how certain the engine is, that the position is actually won. Also if the engine gives a number, it is not 100% sure the position is actually won.
> Generally, a position can only be won or a draw. Any number for a won position is arbitrary.
The number is not arbitrary, it is a measure of how certain the engine is, that the position is actually won. Also if the engine gives a number, it is not 100% sure the position is actually won.
@Sarg0n said in #10:
> Two remarks:
>
> -a big number which doesn't change means nothing. It's a 0.00 basically. A position is only good when it can be improved.
> -when several moves yield the same evaluation it is a strong evidence that a fortress has been built.
>
> PS: check it here lichess.org/forum/game-analysis/do-you-believe-in-fortresses#1
A big number means a high probability for a win, and a high probability for a win means it can likely be improved/converted. I don't know about this fortress of yours, but a game is not over until it is really over. Stockfish gives a quite high probability, that it can be converted, but maybe you know better.
> Two remarks:
>
> -a big number which doesn't change means nothing. It's a 0.00 basically. A position is only good when it can be improved.
> -when several moves yield the same evaluation it is a strong evidence that a fortress has been built.
>
> PS: check it here lichess.org/forum/game-analysis/do-you-believe-in-fortresses#1
A big number means a high probability for a win, and a high probability for a win means it can likely be improved/converted. I don't know about this fortress of yours, but a game is not over until it is really over. Stockfish gives a quite high probability, that it can be converted, but maybe you know better.
@NewtOrange said in #13:
> A big number means a high probability for a win, and a high probability for a win means it can likely be improved/converted. I don't know about this fortress of yours, but a game is not over until it is really over. Stockfish gives a quite high probability, that it can be converted, but maybe you know better.
Sargon gives the probabilty of converting (after the rook swap): 0.00%
> A big number means a high probability for a win, and a high probability for a win means it can likely be improved/converted. I don't know about this fortress of yours, but a game is not over until it is really over. Stockfish gives a quite high probability, that it can be converted, but maybe you know better.
Sargon gives the probabilty of converting (after the rook swap): 0.00%
@NewtOrange said in #12:
> The number is not arbitrary, it is a measure of how certain the engine is, that the position is actually won. Also if the engine gives a number, it is not 100% sure the position is actually won.
This is wrong. The number has nothing to do with probability of winning. It is meant for internal use of the engine, where (traditionally) an evaluation of 3 means the engine values it as advantage equivalent to 3 pawns. It is used by the engine to make decisions using min-max algorithm or its variations. In the old days the engine simply valued a pawn = 1, knight = 3, ... and used these values for decision making. Then the evaluation got more and more sophisticated, so, for example, the engine started evaluating more space or king safety in terms of approximate value in terms of number of pawns. This was often based on programmer's experience or if the programmer took help of strong players, based on their judgment, and experimentation. Then they made it more sophisticated by saying, e.g., in the endgame knight values so much and in the opening it is so many points, and so on. This is sometimes called 'hand crafted evaluation function'.
It is us who are now showing these evaluations to spectators and now spectators are thinking +3 should mean so much chance to win and +8 should definitely win. Engine is not saying that. Engine is just using that number to make decisions.
Some modern engines do Monte Carlo analysis to estimate probability of winning, and some of them may be crudely translating these probabilities to 'pawn units'. But such a translation is not particularly meaningful.
> The number is not arbitrary, it is a measure of how certain the engine is, that the position is actually won. Also if the engine gives a number, it is not 100% sure the position is actually won.
This is wrong. The number has nothing to do with probability of winning. It is meant for internal use of the engine, where (traditionally) an evaluation of 3 means the engine values it as advantage equivalent to 3 pawns. It is used by the engine to make decisions using min-max algorithm or its variations. In the old days the engine simply valued a pawn = 1, knight = 3, ... and used these values for decision making. Then the evaluation got more and more sophisticated, so, for example, the engine started evaluating more space or king safety in terms of approximate value in terms of number of pawns. This was often based on programmer's experience or if the programmer took help of strong players, based on their judgment, and experimentation. Then they made it more sophisticated by saying, e.g., in the endgame knight values so much and in the opening it is so many points, and so on. This is sometimes called 'hand crafted evaluation function'.
It is us who are now showing these evaluations to spectators and now spectators are thinking +3 should mean so much chance to win and +8 should definitely win. Engine is not saying that. Engine is just using that number to make decisions.
Some modern engines do Monte Carlo analysis to estimate probability of winning, and some of them may be crudely translating these probabilities to 'pawn units'. But such a translation is not particularly meaningful.
<Comment deleted by user>
@kajalmaya said in #15:
> This is wrong. The number has nothing to do with probability of winning. It is meant for internal use of the engine, [...]
>
> It is us who are now showing these evaluations to spectators and now spectators are thinking +3 should mean so much chance to win and +8 should definitely win. Engine is not saying that. Engine is just using that number to make decisions.
>
> Some modern engines do Monte Carlo analysis to estimate probability of winning, and some of them may be crudely translating these probabilities to 'pawn units'. But such a translation is not particularly meaningful.
That's a very good point, I didn't consider that, as I should have.
> This is wrong. The number has nothing to do with probability of winning. It is meant for internal use of the engine, [...]
>
> It is us who are now showing these evaluations to spectators and now spectators are thinking +3 should mean so much chance to win and +8 should definitely win. Engine is not saying that. Engine is just using that number to make decisions.
>
> Some modern engines do Monte Carlo analysis to estimate probability of winning, and some of them may be crudely translating these probabilities to 'pawn units'. But such a translation is not particularly meaningful.
That's a very good point, I didn't consider that, as I should have.
"if only he'd explain the explanation" --BYRON
This topic has been archived and can no longer be replied to.