Discussion of anything and everything relating to chess playing software and machines.
Moderators: hgm , Rebel , chrisw
George Tsavdaris
Posts: 1627Joined: Thu Mar 09, 2006 12:35 pm
Post by George Tsavdaris Sun Nov 17, 2019 5:47 pm
NN engines seem to be able to solve the Gusev Qxe5!! incredible Queen sacrifice quite easily!
And with solve i mean not only to find the Qxe5 move, but to have a positive score and show correct continuation also!
Fat Fritz solves it in seconds, same as Leela's T40(20x256) nets, e.g T40B.4-160 but also many others.
New immature T60 Leela net e.g 61104 i tried, solves it also after 1.7 million nodes (around 1 minute).
Yet Leela's J13B.2-188 bigger(24x320) net as also the 136 one, cannot solve it after 5 million nodes searched.
But Stockfish dev(12 November 2019) cannot solve it even after 35 billion nodes searched and 67 ply. It cannot even find the Qxe5 move.
Any AB engine that can solve this relatively fast?
[d]4q1kr/p6p/1prQPppB/4n3/4P3/2P5/PP2B2P/R5K1 w - - 1 1
Fat Fritz 1.0:
1/2 00:00 3 53 +1,67 Qd6-a3 Rc6xe6 14/30 00:03 114.849 29.463 +0,77 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 a2-a4 Qc8-d8 Kg1-g2 g6-g5 Rf1-f7 Qd8-d2+ Kg2-g1 Qd2-e1+ Kg1-g2 Qe1xe4+ Kg2-g1 16/33 00:40 2.347.045 57.986 +1,30 Qd6xe5
Lc0 v22.0 JH J13B.2-188:
1/2 00:00 3 61 +1,55 Qd6-a3 Rc6xe619/41 01:51 5.161.143 46.115 +0,03 Qd6-a3
Lc0 v22.0 T40B.4-160:
1/2 00:00 3 62 +1,62 Qd6-d5 Rc6xe6 18/41 00:18 648.428 35.483 +0,88 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-d1 Rc7-g7 Bd1-b3 Qe8-e7 h2-h4 g6-g5 h4-h5 g5-g4 Bb3-d5 g4-g3 Kg1-g2 b6-b5 b2-b4 a7-a6 Rf1-f3 Rg7-g6 Rf3-f7 Qe7-e8 h5xg6 h7xg6 Rf7-g7+ 17/41 01:12 3.266.647 45.230 +1,07 Qd6xe5
Stockfish 121119 64 POPCNT:
3.Qd6-d4 Rc6xe6 4.Ra1-f1 Kg8-f7 5.Be2-d1 Qe8-b5 6.Bd1-b3 Rh8-e8 7.h2-h3 Qb5-d3 8.Qd4-f2 Ne5-c4 9.Bh6-g5 Kf7-g8 10.Bg5xf6 Re6xf6 11.Bb3xc4+ Qd3xc4 12.Qf2xf6 Qc4xa2 13.Qf6-d4 Qa2-e6 14.Qd4-d5 Kg8-g7 15.Qd5-d4+ 3.Qd6-a3 = (0.00) Depth: 67/11 00:11:46 34889MN, tb=75331776
Lc0 v22.0 61104:
1/2 00:00 2 117 +1,87 Qd6-d5 Rc6xe6 19/47 01:05 1.777.260 27.101 +0,75 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 Kg1-g2 Qc8-a8 h2-h3 Qa8-d8 h3-h4 Qd8-e8 b2-b4 Qe8-e7 c3-c4 Qe7xb4 Rf1-f7 g6-g5 Rf7-d7 Qb4-b2+ Kg2-h3 Qb2-c3+ Kh3-g4 Qc3-d4 20/47 01:59 3.578.764 30.028 +0,79 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 Kg1-g2 Qc8-d8 h2-h4 Qd8-a8 Kg2-g3 Qa8-d8 b2-b4 b6-b5 Rf1-f7 Qd8-d3+ Rf7-f3 Qd3-d8 a2-a3 Qd8-e8 Rf3-f6 Qe8-e7 Rf6-f1 g6-g5 h4xg5 Qe7-d6 Rf1-f6
Last edited by George Tsavdaris on Sun Nov 17, 2019 5:59 pm, edited 1 time in total.
After his son's birth they've asked him:
schack
Posts: 172Joined: Thu May 27, 2010 3:32 am
Post by schack Sun Nov 17, 2019 5:50 pm
I discuss this exact position as part of my review of Fat Fritz / F17 that comes out in two parts beginning today. (This specific position is discussed in Part II.)
George Tsavdaris
Posts: 1627Joined: Thu Mar 09, 2006 12:35 pm
Post by George Tsavdaris Sun Nov 17, 2019 5:53 pm
schack wrote: ↑ Sun Nov 17, 2019 5:50 pm
I discuss this exact position as part of my review of Fat Fritz / F17 that comes out in two parts beginning today. (This specific position is discussed in Part II.)
Where your review will be? Chessbase or some other place?
After his son's birth they've asked him:
schack
Posts: 172Joined: Thu May 27, 2010 3:32 am
Post by schack Sun Nov 17, 2019 5:56 pm
uschess.org - I'll post the link when it goes live.
Master Om
Posts: 449Joined: Wed Nov 24, 2010 10:57 amLocation: INDIA
Post by Master Om Sun Nov 17, 2019 7:45 pm
George Tsavdaris wrote: ↑ Sun Nov 17, 2019 5:47 pm
NN engines seem to be able to solve the Gusev Qxe5!! incredible Queen sacrifice quite easily!
And with solve i mean not only to find the Qxe5 move, but to have a positive score and show correct continuation also!
Fat Fritz solves it in seconds, same as Leela's T40(20x256) nets, e.g T40B.4-160 but also many others.
New immature T60 Leela net e.g 61104 i tried, solves it also after 1.7 million nodes (around 1 minute).
Yet Leela's J13B.2-188 bigger(24x320) net as also the 136 one, cannot solve it after 5 million nodes searched.
But Stockfish dev(12 November 2019) cannot solve it even after 35 billion nodes searched and 67 ply. It cannot even find the Qxe5 move.
Any AB engine that can solve this relatively fast?
[d]4q1kr/p6p/1prQPppB/4n3/4P3/2P5/PP2B2P/R5K1 w - - 1 1
Fat Fritz 1.0:
1/2 00:00 3 53 +1,67 Qd6-a3 Rc6xe6 14/30 00:03 114.849 29.463 +0,77 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 a2-a4 Qc8-d8 Kg1-g2 g6-g5 Rf1-f7 Qd8-d2+ Kg2-g1 Qd2-e1+ Kg1-g2 Qe1xe4+ Kg2-g1 16/33 00:40 2.347.045 57.986 +1,30 Qd6xe5
Lc0 v22.0 JH J13B.2-188:
1/2 00:00 3 61 +1,55 Qd6-a3 Rc6xe619/41 01:51 5.161.143 46.115 +0,03 Qd6-a3
Lc0 v22.0 T40B.4-160:
1/2 00:00 3 62 +1,62 Qd6-d5 Rc6xe6 18/41 00:18 648.428 35.483 +0,88 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-d1 Rc7-g7 Bd1-b3 Qe8-e7 h2-h4 g6-g5 h4-h5 g5-g4 Bb3-d5 g4-g3 Kg1-g2 b6-b5 b2-b4 a7-a6 Rf1-f3 Rg7-g6 Rf3-f7 Qe7-e8 h5xg6 h7xg6 Rf7-g7+ 17/41 01:12 3.266.647 45.230 +1,07 Qd6xe5
Stockfish 121119 64 POPCNT:
3.Qd6-d4 Rc6xe6 4.Ra1-f1 Kg8-f7 5.Be2-d1 Qe8-b5 6.Bd1-b3 Rh8-e8 7.h2-h3 Qb5-d3 8.Qd4-f2 Ne5-c4 9.Bh6-g5 Kf7-g8 10.Bg5xf6 Re6xf6 11.Bb3xc4+ Qd3xc4 12.Qf2xf6 Qc4xa2 13.Qf6-d4 Qa2-e6 14.Qd4-d5 Kg8-g7 15.Qd5-d4+ 3.Qd6-a3 = (0.00) Depth: 67/11 00:11:46 34889MN, tb=75331776
Lc0 v22.0 61104:
1/2 00:00 2 117 +1,87 Qd6-d5 Rc6xe6 19/47 01:05 1.777.260 27.101 +0,75 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 Kg1-g2 Qc8-a8 h2-h3 Qa8-d8 h3-h4 Qd8-e8 b2-b4 Qe8-e7 c3-c4 Qe7xb4 Rf1-f7 g6-g5 Rf7-d7 Qb4-b2+ Kg2-h3 Qb2-c3+ Kh3-g4 Qc3-d4 20/47 01:59 3.578.764 30.028 +0,79 Qd6xe5 f6xe5 Ra1-f1 Rc6-c7 Be2-b5 Qe8-d8 Bb5-d7 Rc7-c8 Bd7xc8 Qd8xc8 Kg1-g2 Qc8-d8 h2-h4 Qd8-a8 Kg2-g3 Qa8-d8 b2-b4 b6-b5 Rf1-f7 Qd8-d3+ Rf7-f3 Qd3-d8 a2-a3 Qd8-e8 Rf3-f6 Qe8-e7 Rf6-f1 g6-g5 h4xg5 Qe7-d6 Rf1-f6
[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[Analysis "Sting"]
[AnalysisTime "285"]
[FEN "4q1kr/p6p/1prQPppB/4n3/4P3/2P5/PP2B2P/R5K1 w - - 0 1"]
[SetUp "1"]
1. Qxe5 fxe5 2. Rf1 Rc8 3. Bd1 Rc4 4. Bb3 b5 5. Kg2 a6 6. Bxc4 bxc4 7. Rf2 a5
8. b3 cxb3 9. axb3 Qe7 10. h3 Qe8 11. b4 a4 12. b5 a3 13. b6 a2 14. Rf1 a1=R
15. Rxa1 Qxe6 16. Rf1 Qf7 17. c4 Qxf1+ 18. Kxf1 Kf7 19. c5 Rb8 20. Be3 Ke6 21.
c6 Kd6 22. c7 Rc8 {(0:04:45) 3649045kN}{[%eval 276,36,
Sting SF 18 x64 ][%meval 285s]} *
Always Expect the Unexpected
jorose
Posts: 358Joined: Thu Jan 22, 2015 3:21 pmLocation: Zurich, SwitzerlandFull name: Jonathan Rosenthal
Post by jorose Sun Nov 17, 2019 10:59 pm
By the way, do we know for sure that the Gusev game is not in the Fat Fritz training dataset? Its from a high level human game, so if CB uses human games in its training it is not impossible?
-Jonathan
Spliffjiffer
Posts: 416Joined: Thu Aug 02, 2012 7:48 pmLocation: Germany
Post by Spliffjiffer Mon Nov 18, 2019 12:56 am
i have no access to a relevant analysis by any NN because i dont use a seperate grafics-card...imo its pretty easy to make (queen)-sacs without having "-9" in your head
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.
Spliffjiffer
Posts: 416Joined: Thu Aug 02, 2012 7:48 pmLocation: Germany
Post by Spliffjiffer Mon Nov 18, 2019 1:15 am
Spliffjiffer wrote: ↑ Mon Nov 18, 2019 12:56 am
i have no access to a relevant analysis by any NN because i dont use a seperate grafics-card...imo its pretty easy to make (queen)-sacs without having "-9" in your head
apart from that though!..i have to admit that an EV by the NN's in favor for white does impress me definately and i dont want to scale down the possibilities of NN's
about 2 years back i predicted a superiority of NN's in approximately 5 years or less and i still dont see why this will be not the case...oc for analysis AB-engines will remain important much longer afaics...the end is coming for (pure) AB-engines, thats for sure imo but i might well err in my 5 years-prediction
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.
Ovyron
Posts: 4556Joined: Tue Jul 03, 2007 4:30 am
Post by Ovyron Mon Nov 18, 2019 10:37 am
Note Sting finds the solution in 22 seconds (credits to Dann Corbit)30 +1.90 00:22 195440K Qxe5 fxe5 Rf1 Rc7 Bb5 Qe7 Ba4 b5 Bb3 Rc4 Kg2 Qe8 Bxc4 bxc4 b3 cxb3 axb3 a5 Kg3 Qe7 Rf2 Qe8 c4 Qe7 Kg2 Qa3 Rf3 Qe7 Rf1 Qa3
George Tsavdaris
Posts: 1627Joined: Thu Mar 09, 2006 12:35 pm
Post by George Tsavdaris Mon Nov 18, 2019 10:55 am
Spliffjiffer wrote: ↑ Mon Nov 18, 2019 12:56 am
i have no access to a relevant analysis by any NN because i dont use a seperate grafics-card...imo its pretty easy to make (queen)-sacs without having "-9" in your head
If you don't have the Q=+9 in your head and go for a sac, then you will end up making in more than 99.99% of cases a blunder that throws the game.
After his son's birth they've asked him:
