Yes sorry for this stupid question. It was due to a manipultion erreur of my own; I disabled the 7 endgame database for test purpose!Ed Gilbert wrote:Hi Gerard,
I am surprised that you ask about this position, because the end of the game can be seen fairly quickly without an endgame database.
-- Ed
No problem at all Gerard. It was actually a nice test position for engines with smaller databases or no databases. I tried it on flits, truus, and dam2.2 also.Yes sorry for this stupid question. It was due to a manipultion erreur of my own; I disabled the 7 endgame database for test purpose!
Position after 38...32-37Ed Gilbert wrote:Hi Bert,
We were discussing the Georgiev game earlier today. Kingsrow finds that Ndjofang gave up winning positions at several places in the game. I have inserted some comments in the game pdn below.
[Event "WC 2011 round 14"]
[Date "2011.05.21"]
[White "Ndjofang"]
[Black "Georgiev"]
[Result "1/2-1/2"]
1. 32-28 17-22 2. 28x17 12x21 3. 33-28 7-12 4. 39-33 11-17 5. 44-39 6-11 6. 50-44 1-6 7. 37-32 19-23 8. 28x19 14x23 9. 42-37 9-14 10. 32-28 23x32 11. 37x28 14-19 12. 47-42 10-14 13. 41-37 5-10 14. 37-32 21-26 15. 34-29 26x37 16. 42x31 19-23 17. 28x19 14x34 18. 39x30 10-14 19. 44-39 16-21 20. 31-26 21-27 21. 32x21 11-16 22. 46-41 16x27 23. 41-37 6-11 24. 37-31 4-9 25. 31x22 18x27 26. 33-28 11-16 27. 30-25 20-24 28. 40-34 12-18 29. 39-33 2-7 30. 43-39 18-22 31. 49-43 7-12 32. 45-40 12-18 33. 34-29 13-19 {Database draw}
34. 29x20 15x24 35. 40-34 18-23 36. 34-29 23x32 37. 29x20 19-24 {the db draw changes to an advantage for white, but not enough to win}
38. 20x29 32-37 39. 39-34 9-13 40. 34-30 {Database draw now.}
8-12 41. 33-28 22x42 42. 43-38 42x24 43. 30x10 27-31 44. 36x27 37-41 45. 10-4 3-9 46. 4-15 41-46 47. 15-33 46-23 48. 33x6 23-45 {This is a database loss for black. 23-18 draws.}
49. 6-50 {This only draws. 6-33 wins.}
12-18 {Loses. 45-18 draws.}
50. 25-20 9-14 51. 20x9 13x4 52. 35-30 18-23 53. 27-22 {Draws. 30-25 wins.}
45-40 54. 22-17 23-28 55. 50x22 16-21 56. 30-25 21x12 57. 22-50 40-23 58. 50-44 4-10
-- Ed
I did not find a win. I looked at this path:Isn't it possible to win by 29-23 ? It is not very clear for me.
Bert, 41-47 might be a loss, but that makes it an inferior move, since 14-19 seems to draw. You have to let black defend with its best moves when trying to find a white win.Ed I was also looking at 41 .... 41-47, which seem to be a white win, but as the 41. .... 14-19 is a draw , I assume this position is a draw.
Seeing your path I totally agree with you but I am not able to give a conclusion.Ed Gilbert wrote:Hi Gerard,I did not find a win. I looked at this path:Isn't it possible to win by 29-23 ? It is not very clear for me.
39. 29-23 27-31 40. 36x18 37-41 41. 39-34 14-19 42. 23x14 9x20 43. 25x14 41-46 44. 14-9 3x14 45. 33-29 14-20 and now kingsrow sees a database draw. Did you have some different play for white?
-- Ed
Oops I forget to say that my feeling is that this position is really a draw. My point is that a "db draw" in this kind of position is not sufficient for a conclusion. I am experimenting some other approaches to deal with such situations.TAILLE wrote:Seeing your path I totally agree with you but I am not able to give a conclusion.Ed Gilbert wrote:Hi Gerard,I did not find a win. I looked at this path:Isn't it possible to win by 29-23 ? It is not very clear for me.
39. 29-23 27-31 40. 36x18 37-41 41. 39-34 14-19 42. 23x14 9x20 43. 25x14 41-46 44. 14-9 3x14 45. 33-29 14-20 and now kingsrow sees a database draw. Did you have some different play for white?
-- Ed
White to play
My point is the following: in the above position I consider that white has a material advantage (4 men against a king). If black is able to reach a position where the material balance is 3 men against a king then, in the majority of cases, the evaluation function will give black a small advantage. That means that you can easily conclude to a db draw though the position is very very unclear.
In the above position how do you continue after 34-30 ? I am still unable to say if it is a win or a draw.
Gerard, it's true that unless you do something extra in your search to propagate additional information so you can tell whether any part of the search score depends on heuristics, or if it is all from database results, then a db draw search score does not mean a guarantee of draw as it is with a db win or db loss.My point is the following: in the above position I consider that white has a material advantage (4 men against a king). If black is able to reach a position where the material balance is 3 men against a king then, in the majority of cases, the evaluation function will give black a small advantage. That means that you can easily conclude to a db draw though the position is very very unclear.
In the above position how do you continue after 34-30 ? I am still unable to say if it is a win or a draw.
Yes Ed. I totally agree with you. In positions with a material balance of 1 king against 3 or 4 men it is very difficult to be sure of what is a "db draw". We need to continue at greater depth in order to have a better feeling.Ed Gilbert wrote:Gerard, it's true that unless you do something extra in your search to propagate additional information so you can tell whether any part of the search score depends on heuristics, or if it is all from database results, then a db draw search score does not mean a guarantee of draw as it is with a db win or db loss.My point is the following: in the above position I consider that white has a material advantage (4 men against a king). If black is able to reach a position where the material balance is 3 men against a king then, in the majority of cases, the evaluation function will give black a small advantage. That means that you can easily conclude to a db draw though the position is very very unclear.
In the above position how do you continue after 34-30 ? I am still unable to say if it is a win or a draw.
In this position, at move 46 it looks like white's only hope for a win is 34-30, as all others show a db draw score by depth 15, and I am searching the positions to a depth of at least 25 plies. Black continues 20-25, and white again has only one viable move, 30-24. Black plays 25-30, and then white's only good move is 38-33 30x19. White only has 18-13 (what else?), then 8-12 13x24. Kingsrow search is showing db draw by depth 20, and it continues db draw up to at least depth 29 where I stopped it. If white has a win, then I would expect in 29 plies that kingsrow would at least see some hint of it in a better evaluation for white. White now has a material advantage and a significant tempo advantage, but the search returns only draw. I conclude that this is very likely a draw, but you are right that it not a certainty.
-- Ed
The fact that 38-32 gave a draw score on one search does not at all suggest that the move is a highly probable draw. With 18 pieces on the board, it is likely that a large percentage of leaf nodes are scored by heuristics, not db results, so a draw score here means mosty a heuristic draw instead of a database result. In an extreme case you might even get a draw score with 30 pieces on the board, and the end of that PV might be the only leaf node in the whole search that was the result of a db lookup! You always have to consider how many pieces are on the board.No doubt that 26-21 and 39-34 are good moves but what about the evaluation of move 38-32 which looks like a db draw?
Maybe I am wrong but I suspect Kingrow has too quickly eliminated the sequence really played in the game namely 38-32 28-33 39x28 22x33 49-43 31-36 43-38 36-41? 38x29 41-46 which leads to very good material balance for black though it is a losing position!
Our nigthmare is always to see our program pruning too agressively the sequence really played in the game. Is it the case here?
What is your analysis Ed?
I agree with you that 38-32 leads to a draw but you understood clearly that it was not my point.Ed Gilbert wrote:The fact that 38-32 gave a draw score on one search does not at all suggest that the move is a highly probable draw. With 18 pieces on the board, it is likely that a large percentage of leaf nodes are scored by heuristics, not db results, so a draw score here means mosty a heuristic draw instead of a database result. In an extreme case you might even get a draw score with 30 pieces on the board, and the end of that PV might be the only leaf node in the whole search that was the result of a db lookup! You always have to consider how many pieces are on the board.No doubt that 26-21 and 39-34 are good moves but what about the evaluation of move 38-32 which looks like a db draw?
Maybe I am wrong but I suspect Kingrow has too quickly eliminated the sequence really played in the game namely 38-32 28-33 39x28 22x33 49-43 31-36 43-38 36-41? 38x29 41-46 which leads to very good material balance for black though it is a losing position!
Our nigthmare is always to see our program pruning too agressively the sequence really played in the game. Is it the case here?
What is your analysis Ed?
However, the evidence you presented does not indicate that 38-32 is a loss, and after more analysis I think it probably is a draw. I play it forward 38-32 28-33 39x28 27x33, and now I think black can probably draw playing either 9-13 or 4-10. I say that because there are now 16 pieces on the board, and I get the value db draw for at least 9 or 10 successively deeper searches. In the game black played 31-36, which is not as clear, but in any case after 31-36 43-38, black made a mistake playing 36-41 which might have lead to the loss. So 38-32 is not an example of an errant draw score.
I have seen a few examples of draw scores that looked credible but turned out to be incorrect. In each case they were the result of very deep combinations that were as you suggested somewhat hidden by search reductions. This is the tradeoff that we have to make. Without extensions and reductions our engines can only go to shallow search depths and will on average play much weaker. You can always find exceptions where this doesn't work well, but you have to play the percentages.
-- Ed
