Moderator: Ras
Can't wait to hear the results.JuLieN wrote:My deep Stockfish analysis of what would have happened if White had played 42.Qb1 is still going on. It's a Stockfish-Stockfish autoplay and takes times, because I use a fixed depth analysis set to depth 28, which result in thinking times varying from 15mn up to more than 4 hours, depending on the position (I have a Core i5 CPU).
Up to now, SF is thinking for its 59th move (so 32 half moves have been played yet). I won't tell u the result until the analysis is over, but I can yet tell you that it WILL be interesting.... and that the evaluation is...well... far from being 0.00 right now... very very far from this...
I looked at this position for a while and the PV went into a QPPqp endgame which may be won for white but was much too deep for stockfish (I would guess that the possible promotion was 80-100 plies away). A program like freezer should be able to solve it in seconds, however.JuLieN wrote:My deep Stockfish analysis of what would have happened if White had played 42.Qb1 is still going on. It's a Stockfish-Stockfish autoplay and takes times, because I use a fixed depth analysis set to depth 28, which result in thinking times varying from 15mn up to more than 4 hours, depending on the position (I have a Core i5 CPU).
Up to now, SF is thinking for its 59th move (so 32 half moves have been played yet). I won't tell u the result until the analysis is over, but I can yet tell you that it WILL be interesting.... and that the evaluation is...well... far from being 0.00 right now... very very far from this...
Nope, this is not at all what I've got here. The position white has at move 42 is actually fascinating because analyzing it with an engine at great depth shows that it's actually a constant zugzwang : there's only one possible move each time for each side, or playing another one leads to a faster disaster (for black) or to a draw (for white). It's funny, because it's not obvious to a human eye. Just give me some time to complete the analysis and you'll have the full package when it's done. The bad news is that SF is taking more and more time : at first the moves took about 15-20mn to get analyzed, then around an hour, two hours, and now, since 4-5 half moves, it's over 2h and closer to 4h. Yet I could stop the analysis, as the result from the position I have is obvious. But I like a job well done.jwes wrote:I looked at this position for a while and the PV went into a QPPqp endgame which may be won for white but was much too deep for stockfish (I would guess that the possible promotion was 80-100 plies away). A program like freezer should be able to solve it in seconds, however.JuLieN wrote:My deep Stockfish analysis of what would have happened if White had played 42.Qb1 is still going on. It's a Stockfish-Stockfish autoplay and takes times, because I use a fixed depth analysis set to depth 28, which result in thinking times varying from 15mn up to more than 4 hours, depending on the position (I have a Core i5 CPU).
Up to now, SF is thinking for its 59th move (so 32 half moves have been played yet). I won't tell u the result until the analysis is over, but I can yet tell you that it WILL be interesting.... and that the evaluation is...well... far from being 0.00 right now... very very far from this...
Stockfish branching factor is around 1.9 (which is actually the best existing branching factor), which means it would take almost twice long for each additional ply of depth.JuLieN wrote:The bad news is that SF is taking more and more time : at first the moves took about 15-20mn to get analyzed, then around an hour, two hours, and now, since 4-5 half moves, it's over 2h and closer to 4h. Yet I could stop the analysis, as the result from the position I have is obvious. But I like a job well done.
Okee, but I didn't meant any added ply was doubling the time, I meant that moves now take more time to analyze than they did just 30 half-moves before (all analyzed at depth 28), meaning that the positions reached make for less pruning for SF, hence more time needed to reach depth 28.Milos wrote:Stockfish branching factor is around 1.9 (which is actually the best existing branching factor), which means it would take almost twice long for each additional ply of depth.JuLieN wrote:The bad news is that SF is taking more and more time : at first the moves took about 15-20mn to get analyzed, then around an hour, two hours, and now, since 4-5 half moves, it's over 2h and closer to 4h. Yet I could stop the analysis, as the result from the position I have is obvious. But I like a job well done.
I think Rybka would win as in the endgames, you do not need to search deep, its just a matter of playing good looking moves before breaking through. However, we need to do tests. Perhaps a test on a standard endgame suite? What is a good suite these days of equal (by evaluation) endgame positions?AdminX wrote: Stockfish would win, Rybka needs it's tablebases in order to equalize the match.
If I understood correctly this is about analyzing the position after 42.Qb1 instead of the game move. Maybe I have missed a part of the discussion, so may I ask for the intention of that analysis? Is it to find out whether there is another way to win for white in addition to 42.Qa4 which has already been analyzed before, or are there substantial doubts about the outcome after 42.Qa4, or is it just to analyze an interesting position?JuLieN wrote:Nope, this is not at all what I've got here. The position white has at move 42 is actually fascinating because analyzing it with an engine at great depth shows that it's actually a constant zugzwang : there's only one possible move each time for each side, or playing another one leads to a faster disaster (for black) or to a draw (for white). It's funny, because it's not obvious to a human eye. Just give me some time to complete the analysis and you'll have the full package when it's done. The bad news is that SF is taking more and more time : at first the moves took about 15-20mn to get analyzed, then around an hour, two hours, and now, since 4-5 half moves, it's over 2h and closer to 4h. Yet I could stop the analysis, as the result from the position I have is obvious. But I like a job well done.jwes wrote:I looked at this position for a while and the PV went into a QPPqp endgame which may be won for white but was much too deep for stockfish (I would guess that the possible promotion was 80-100 plies away). A program like freezer should be able to solve it in seconds, however.JuLieN wrote:My deep Stockfish analysis of what would have happened if White had played 42.Qb1 is still going on. It's a Stockfish-Stockfish autoplay and takes times, because I use a fixed depth analysis set to depth 28, which result in thinking times varying from 15mn up to more than 4 hours, depending on the position (I have a Core i5 CPU).
Up to now, SF is thinking for its 59th move (so 32 half moves have been played yet). I won't tell u the result until the analysis is over, but I can yet tell you that it WILL be interesting.... and that the evaluation is...well... far from being 0.00 right now... very very far from this...
I did an end-game tournament relatively recently.shiv wrote:I think Rybka would win as in the endgames, you do not need to search deep, its just a matter of playing good looking moves before breaking through. However, we need to do tests. Perhaps a test on a standard endgame suite? What is a good suite these days of equal (by evaluation) endgame positions?AdminX wrote: Stockfish would win, Rybka needs it's tablebases in order to equalize the match.