Uri Blass wrote: ↑Thu Jan 13, 2022 11:11 pmArrogant stockfish claims to get depth 104 but still cannot see the draw.
I wonder if some top program can see the draw(it is not the latest developement version but I doubt latest developement version is better)
[fen]2k5/2p5/1q1p4/pPpPp1pp/N1P1Pp2/P4PbP/KQ4P1/8 w - - 7 46[/fen]
Stockfish_22011019_x64_avx2:
NNUE evaluation using nn-ac07bd334b62.nnue enabled
...
It's quite stupid to attach a Human characteristic to a chess program, if it's output is wrong (for a 1/1000000 edge case).
Calling it 'arrogant' just leaves the (correct) impression you are biased.
You may be right but I do not like the big depth.
In the past when I analyzed the position I never remember getting this big depth.
Other programs also do not understand the position but at least show smaller depths and I do not like claiming to get very deep if you do not see a forced mate or a forced draw when usually stockfish cannot get depth 100 even after an hour of search(including positions with a big material advantage when it does not see a forced mate).
And what do you want to get changed so it doesn't show that high depths anymore? Divide depth by 3 if depth >= 50 so your high depth bias doesn't get triggered? That'd be a change worthy of Houdini's magical speed up.
BTW: Crystal is very close to 0 in evaluation. You should try it if you need to run engines on such positions.
A typical comment.
If you look at the last lines of the analysis given, you will notice the selective search depth to be significantly lower than the nominal one.
You will also see that almost NO time is used for the last depths.
hi Uri, i did a severe analysis with Crystal4 and thx for the analysis u provided above i was able to find out that after 1.Qb1 every move is losing for sure !!?....unfortunately this destroid my hashes and i cant proove it momentarely....hopefully later though...beyond from that im pretty much sure that every 1. King-move and nearly all 1. Queen-moves win as well maybe even 1.Nc3 idk...
so far... this was my mailine i backtracked from:
46. Qb1 h4 47. Qb2 Kb7 48. Kb3 Kb8 49. Kc3 Kb7 50. Kd3 Ka7 51. Ke2 Ka8 52.
Kf1 Kb8 53. Qc2 Kb7 54. Qb3 Bh2 55. g4 Bg3 56. Qc2 Ka8 57. Nb2 Qb8 58. Qb3
Bh2 59. Nd3 Qb6 60. Qc2 Bg3 61. a4 Kb7 62. Qb2 Kb8 63. Qc3 Kb7 64. Ke2 Bh2
65. Kf2 Bg3+ 66. Kg2 Kc8 67. Nc1 Qa7 68. Nb3 Kb7 69. Qxa5 +-
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.
I did not claim that Qb1 does not win but I did not understand the idea of that move.
Maybe it is to prevent not only Be1 but also the plan to Bh2 Bg1 when first white move the king to f1 to prevent counter play by the bishop and later white can attack the pawn a5 from the diagonal when the black bishop can do nothing.
By analyzing backward I find easily that after 54...Bh2 it is more than +8 for white.
Crystal can easily see with backward analysis that there is no good alternative to Bh2 but the score is only 4.92 for white at depth 39 with the line
54...Kc8 55.Nb2 Kb8 56.a4 Qa7 57.Qc3 Qa8 58.Nd3 Kb7 59.Kg1 Kb6 60.Nxe5 so maybe in some line white needs to sacrifice the knight for 2 pawns.
Uri Blass wrote: ↑Fri Jan 14, 2022 5:59 pm
I did not claim that Qb1 does not win but I did not understand the idea of that move.
Maybe it is to prevent not only Be1 but also the plan to Bh2 Bg1 when first white move the king to f1 to prevent counter play by the bishop and later white can attack the pawn a5 from the diagonal when the black bishop can do nothing.
By analyzing backward I find easily that after 54...Bh2 it is more than +8 for white.
Crystal can easily see with backward analysis that there is no good alternative to Bh2 but the score is only 4.92 for white at depth 39 with the line
54...Kc8 55.Nb2 Kb8 56.a4 Qa7 57.Qc3 Qa8 58.Nd3 Kb7 59.Kg1 Kb6 60.Nxe5 so maybe in some line white needs to sacrifice the knight for 2 pawns.
yes, acc to my analysis white can most probably sac the knight at e5 vs 2 pawns as well (instead of going for the a5 pawn !?)...so far i was able to "remember" Crystal4 a little bit again and got this output in MPV=4 after 1.Qb1...
Crystal4 after some forward/backward-analysis in MPV=4 after 1.Qb1:
maybe im also able to manage a MPV-search from the root pos with similar outputs for the other moves but currently i have no time to try it, hopefully later...
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.
I got the following analysis with crystal at move 54(still not +8 but I think I will go backward because the score is more than 6.5
[fen]8/1kp5/1q1p4/pPpPp1p1/N1P1Pp1p/PQ3PbP/6P1/5K2 b - - 15 54[/fen]
Crystal.4.0.x64.avx2:
NNUE evaluation using nn-ac07bd334b62.nnue enabled
Spliffjiffer wrote: ↑Fri Jan 14, 2022 6:33 pm
yes, acc to my analysis white can most probably sac the knight at e5 vs 2 pawns as well (instead of going for the a5 pawn !?)...so far i was able to "remember" Crystal4 a little bit again and got this output in MPV=4 after 1.Qb1...
Crystal4 after some forward/backward-analysis in MPV=4 after 1.Qb1:
maybe im also able to manage a MPV-search from the root pos with similar outputs for the other moves but currently i have no time to try it, hopefully later...
According to my database in the game A.B.Petrosjan-L.Hazay, Schilde 1970, at 46th move White has some ways of winning by force manoeuvring the Knight to b3 to threaten Black's a-pawn, the other one pawn at e5 is threatened in some lines too. Most important point is to keep Black from pushing a-pawn to a4, to deny that Black has to have triangles with Queen and King before removing Knight from a4 at a moment, when a-line isn't guarded by black Queeen. As soon as Black gets into zugzwang, such a position is to be reached and to come into that (zugzwang), its Bishop must not be able to shuffle to 1st rank anymore. Thus White has to keep g1- field guarded too with Queen or King.
In the original game win was missed by 46.Nxb6(?).
The previous move when I could not get even +6 when I go back ward is move 48 of black after 46.Qb1 h4 47.Qb2 Kb7 48.Kb3
When I only get the following line
I will stop the analysis and it seems convincing crystal by backward analysis that 46.Qb1 is winning is not so easy(I am not convinced that +4 is winning)
[fen]8/1kp5/1q1p4/pPpPp1p1/N1P1Pp1p/PK3PbP/1Q4P1/8 b - - 3 48[/fen]
Crystal.4.0.x64.avx2:
NNUE evaluation using nn-ac07bd334b62.nnue enabled
Crystal's eval is generally somewhat more conservative than SF's.
Additionally, the progress (or lack thereof) detection is both more rigorous and more severe than probably any other engine.
A score of +4 essentially translates into minor piece + pawn which is almost always winning.
+5 is a Rook which is solo mating material, so gg.
If there is any doubt, it can be good to look at the score trend vs depth. If decreasing, cause for concern. If flat, it doesn't know (too many rule50 resets available).
