testposition from the topboard of the candidates

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Thomas Lagershausen
Posts: 261
Joined: Mon Jun 11, 2007 4:59 pm

Re: testposition from the topboard of the candidates

Post by Thomas Lagershausen » Sat Mar 28, 2020 8:30 pm

@Alayan

I give moves, you give only words.

If you are not interested in the truth of the position leave us alone.
TL

jp
Posts: 1320
Joined: Mon Apr 23, 2018 5:54 am

Re: testposition from the topboard of the candidates

Post by jp » Sat Mar 28, 2020 8:32 pm

Thomas Lagershausen wrote:
Sat Mar 28, 2020 8:11 pm
Stockfish give +7 for white after 33.Rb7
FEN: 6r1/pR1nk3/4p2p/3pPp1N/q1pP1P1p/2P5/6P1/1Q4K1 b - - 1 33

Till now it looks like that 26...g6 is no improvement for the defence against 23.f4 !
Maybe you can impove the defence?
Check the black moves given by Lc0 in the posting of zullil.
If you didn´t find an improvement in this moves it is proven that 23.f4 !! is best.
You'd have to do the same for every other line before you can even begin to claim that.

e.g. If you checked, maybe Stockfish also gives more than +7 for White after the 33rd move of Lc0's other (not 23.f4) lines.

And then you'd need to look for improvement for Black for every move before that in every one of those lines. :shock:

zullil
Posts: 6073
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: testposition from the topboard of the candidates

Post by zullil » Sat Mar 28, 2020 9:01 pm

Thomas Lagershausen wrote:
Sat Mar 28, 2020 8:11 pm

If you didn´t find an improvement in this moves it is proven that 23.f4 !! is best.
As I said in my very first reply, White already had a won position before move 23. A variety of twenty-third moves by White maintain the win.

Previously, you suggested that "best" meant "the shortest way to a full point". I doubt that 23. f4 gives the fastest win.

Alayan
Posts: 252
Joined: Tue Nov 19, 2019 7:48 pm
Full name: Alayan Feh

Re: testposition from the topboard of the candidates

Post by Alayan » Sat Mar 28, 2020 9:10 pm

Thomas Lagershausen wrote:
Sat Mar 28, 2020 8:30 pm
@Alayan

I give moves, you give only words.

If you are not interested in the truth of the position leave us alone.
The evals/PVs of Leela + Stockfish carry much more weight than one man's opinion.

You can do nothing that would prove f4 superior to Rxb6 or the other winning moves, because when it comes to shortest mate in a position where it's far away, humans are powerless in front of the trillions of possibilities.

User avatar
Ovyron
Posts: 3950
Joined: Tue Jul 03, 2007 2:30 am

Re: testposition from the topboard of the candidates

Post by Ovyron » Sat Mar 28, 2020 11:30 pm

Thomas Lagershausen wrote:
Sat Mar 28, 2020 8:30 pm
If you are not interested in the truth of the position leave us alone.
If the truth is that most moves are winning then that means f4 is not better than the others. If the shortest mating line includes a position where it's very difficult for a human to find the winning move as white then that's NOT THE BEST LINE (the best line is the most efficient/easiest one to play.)

On the engine's side I propose the following test to rank winning moves, it's still "subjective" and "a matter of taste", but the concept here is that humans are still considerably stronger than Stockfish at Depth 1, so if the upcoming positions can be easily won for it they should also be easy for a human.

-Give Stockfish Depth 1 the winning side
-Can you find a defense for the losing side?

Finding a defense means throwing all your resources to the position and playing a line where D1 doesn't win. If D1 beats all you defenses with some move, that move is best. If D1 is capable of beating any defense with several moves, you're probably wasting your time trying to rank them.

If you can find a defense against all moves at D1:

-Give Stockfish Depth 2 the winning side
-Can you find a defense for the losing side?

And go on like this.

Eventually you'll reach a Stockfish Depth that beats all your defenses with some move (if you didn't the position wasn't actually winning), this move would be the most efficient one, at least for Stockfish (perhaps Ethereal would have another, but it'd require more work to find it.) Would that move be f4?

Ron Langeveld
Posts: 135
Joined: Tue Jan 05, 2010 7:02 pm

Re: testposition from the topboard of the candidates

Post by Ron Langeveld » Sat Mar 28, 2020 11:45 pm

Thomas Lagershausen wrote:
Wed Mar 25, 2020 3:28 pm


23.f2-f4 !

Did a machine play this great move?

I do not see what is so special about f4. It is not a deciding move. Apart from not being deciding is is also the first move I would try in my local pub, or even blindfold.

Collingwood
Posts: 14
Joined: Sat Nov 09, 2019 2:24 pm
Full name: Colin Wood

Re: testposition from the topboard of the candidates

Post by Collingwood » Sat Mar 28, 2020 11:52 pm

Ron Langeveld wrote:
Sat Mar 28, 2020 11:45 pm
I do not see what is so special about f4. It is not a deciding move. Apart from not being deciding is is also the first move I would try in my local pub, or even blindfold.
Apart from that, all the evidence is that it's not even the best move by any rational definition of best.

User avatar
Ovyron
Posts: 3950
Joined: Tue Jul 03, 2007 2:30 am

Re: testposition from the topboard of the candidates

Post by Ovyron » Sun Mar 29, 2020 1:04 am

What about "best style"? If it's the most stylish move perhaps engine settings could be tweaked to make it play it, a "Lagrave" personality of sorts.

Might look visually better than swapping rooks, retreating a rook, or doubling them, even if those are stronger moves, so it could have some aesthetic merit (i.e. if it also wins you might as well play it because of rule of cool)

Zenmastur
Posts: 826
Joined: Sat May 31, 2014 6:28 am

Re: testposition from the topboard of the candidates

Post by Zenmastur » Sun Mar 29, 2020 1:35 am

I guess this means that no one bothered to check Zullil's or MikeB's analysis to see how many errors they had in them.

Both of them give similar lines for 23.f4:
MikeB wrote:
Sat Mar 28, 2020 7:12 pm
this is a ~1.2 trillion nodes searched

info depth 54 seldepth 77 multipv 1 score cp 261 nodes 1185774568113 nps 90646005 hashfull 1000 tbhits 0 time 13081377 pv b4b6
info depth 53 seldepth 89 multipv 2 score cp 222 nodes 1185774568113 nps 90646005 hashfull 1000 tbhits 0 time 13081377 pv f1b1 b6a6 b1a1 d7f8 b4b1 g7g5 d2c1 a6b6 c1a3 f8d7 f2f4 g5f4 e2f4 b6b3 b1b3 c4b3 c2b3 a5b6 b3b5 h8g8 a1b1 c6e7 g1f2 b6c7 b5b4 e7f5 b4b7 e8d8 f4e2 c7b7 b1b7 a7a6 b7a7 d8c8 a7a8 d7b8 a3c5 g8g5 c5a7 c8b7 a8b8 b7a7 b8f8 g5g7 g2g4 f5h4 f8e8 g7g4 e8e7 a7b6 e7f7 g4g2 f2f1 g2g5 e2f4 g5f5 f7f5 h4f5 f4e6 f5g3 f1g2 g3e4 e6g7 b6c7 g7f5 e4c3 f5h6 c3e2 h6f5 e2f4 g2g3
info depth 53 seldepth 92 multipv 3 score cp 196 nodes 1185774568113 nps 90646005 hashfull 1000 tbhits 0 time 13081377 pv b4b1 c6e7 b1b6 a7b6 d2c1 d7b8 c1a3 b8c6 f2f4 a5a7 c2b2 f7f5 e5f6 g7f6 b2b5 a7d7 a3e7 c6e7 b5b6 e8f7 f4f5 e7f5 a4a5 h8c8 e2f4 d7d6 b6b7 d6c7 b7b1 c8g8 a5a6 c7b8 b1b8 g8b8 f4d5 e6d5 f1f5 b8a8 f5d5 a8a6 d5c5 a6a1 g1h2 a1d1 c5c4 f7e6 c4c5 d1d3 d4d5 e6d6 c5c6 d6d5 c6f6 d3c3 f6h6 d5e5 h6g6 e5f5 g2g4 f5f4 h5h6 c3c2 h2h3
info depth 53 seldepth 103 multipv 4 score cp 173 nodes 1185774568113 nps 90646005 hashfull 1000 tbhits 0 time 13081377 pv f2f4 c6e7 b4b6 d7b6 f1a1 h8g8 d2c1 g7g6 g1f2 g6h5 c2h7 b6d7 h7h6 a5b6 c1a3 b6b3 h6h7 d7f8 h7b1 e7c6 b1c1 h5h4 a1b1 b3a4 a3d6 a4a6 b1b2 f7f5 e5f6 f8d7 d6e5 g8g6 f4f5 e6f5 c1f4 g6f6 e5f6 d7f6 f4f5 c6e7 b2b8 e8f7 f5f3 a6e6 b8a8 a7a6 f2g1 e6b6 e2f4 b6b1 g1h2 b1f5 f4e2 f6g4 h2g1 f5f3 g2f3 g4e3 a8a6 e3d1 g1g2 e7f5 a6c6 f7e7 c6b6 e7f7 b6a6 d1e3 g2h3 e3d1 h3g4
bestmove b4b6 ponder a7b6
Looks really impressive!

It also looks like you missed something. The following position is after 29. … Qb6 (bold in the quote above) in the 23.f4 line. This position occurs in Zullil's analysis as well.



Instead of 30.Ba3 which occurs in both of your analysis of 23.f4 we have this:

Code: Select all

30.Qxh5 Qb3 31.Ba3 Qc2 32.Rh1 Nb6 33.Bxe7 Kxe7 34.f5 exf5 35.Qh4+ Kd7 36.g3 Rg6 37.Qh7 Ke7 38.Qh8 Nd7 39.a5 f4 40.Qh4+ f6 41.Qxf4 Rg7 42.exf6+ Nxf6 43.Qe5+ Kf7 44.Rf1 Rg6 45.Kg1 Kg8 46.Qe6+ Kg7 47.Rf5 Qb1+ 48.Kg2 Qe4+ 49.Qxe4 Nxe4 50.Rxd5 Rf6 51.Nf4 Rf7 52.a6 Kf8 53.Kf3 Nxc3 54.Rc5 Nb1 55.Ke3 Re7+ 56.Kf2 Ke8 57.Nd5 Rf7+ 58.Ke3 Na3 59.Rc8+ Kd7 60.Rc7+ Ke6 61.Nf4+ Kf6 62.Rc6+ Ke7 63.g4 Kd7 64.d5 Rg7 65.Nh5 Re7+ 66.Kf4 Nb5 67.Nf6+ Kd8 68.Rxc4 Re1 69.g5 Nd6 70.Rc6 Rf1+ 71.Ke3 Nf7 72.g6 Nh6 73.Rd6+ Kc8 74.Ke4 Rg1 75.Nh5 Rg4+ 76.Kf3 
I have this line at +4.09 which is much better than any of the lines you gave. It doesn't prove the OP is right, but it also doesn't support the idea that the OP is wrong either.

Black can try 25. ...f5, 25. .. g6, or 25. ... Nbc8 but none seem to get black anywhere near the score of 1.73 (MikeB) or 1.83 (Zullil). I didn't search the lines too deep but the best I saw for black was 25. ... f5 @ +3.11 which is better than MikeB's best line (+2.61) and Zullil's best line (+2.04)

Code: Select all

( [Stockfish 250220 64 POPCNT] 53:+3.11 25...f5 26.exf6 gxf6 27.f5 e5 28.Qb1 Nd7 29.dxe5 Qc5+ 30.Kh1 fxe5 31.f6 Nxf6 32.Qb8+ Qc8 33.Qxe5 Rf8 34.Nf4 Qf5 35.Qb8+ Kf7 36.Qxa7 Qd7 37.Qxd7 Nxd7 38.a5 Ra8 39.Be3 Ra6 40.g3 d4 41.cxd4 Nf6 42.Kg2 Nc6 43.Ra4 Rxa5 44.Rxc4 Ne7 45.Rc5 Ra3 46.Bc1 Rb3 47.Ne2 Ng4 48.Ra5 Rb6 49.Kf3 Nh2+ 50.Ke4 Ng4 51.Nc3 Nf6+ 52.Kd3 Nc6 53.Rf5 Ne7 54.Rc5 Rc6 55.Rb5)
I didn't double check this line so it's probably not perfect. I also didn't bother looking at lines other than 23.f4. Their evaluations could very well be off by as much or more for better or for worse.
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

User avatar
Ovyron
Posts: 3950
Joined: Tue Jul 03, 2007 2:30 am

Re: testposition from the topboard of the candidates

Post by Ovyron » Sun Mar 29, 2020 6:30 am

Ovyron wrote:
Sun Mar 29, 2020 1:04 am
What about "best style"? If it's the most stylish move perhaps engine settings could be tweaked to make it play it, a "Lagrave" personality of sorts.

Might look visually better than swapping rooks, retreating a rook, or doubling them, even if those are stronger moves, so it could have some aesthetic merit (i.e. if it also wins you might as well play it because of rule of cool)
Unfortunately, if you try this the engine that likes to play f4 could switch to g4 with more depth:

*Modified engine personality*

24/39 0:11 +3.67++ 23.f4 (22.767.073) 2059
24/48 0:12 +4.09++ 23.f4 (25.208.811) 2056
24/51 0:13 +4.11 23.f4 Nxb4 24.cxb4 Rxb4 25.Qc3 Rxa4 26.Qxa5 Rxa5 27.Bxa5 Nb8 28.Rb1 Nc6 29.Kf2 Rf8 30.Bb4 Nxb4 31.Rxb4 f6 32.Rb8+ Kf7 33.Rb7+ Kg8 34.Rxa7 Rc8 35.exf6 gxf6 36.Re7 (26.789.920) 2055 TB:5
25/42 0:13 +3.91-- 23.f4 Rxb4 (27.542.439) 2055 TB:5
25/42 0:14 +3.85-- 23.f4 Rxb4 (29.367.635) 2053 TB:5
25/46 0:14 +3.68-- 23.f4 Rxb4 (30.138.206) 2054 TB:5
25/50 0:15 +3.83++ 23.f4 (31.223.675) 2055 TB:5
25/50 0:15 +3.78 23.f4 Rxb4 24.cxb4 Qb6 25.a5 Nxd4 26.axb6 Nxc2 27.bxa7 Ke7 28.Rc1 Ra8 29.Rxc2 Rxa7 30.Nd4 Ra8 31.Nc6+ Ke8 32.Bc3 g6 33.hxg6 fxg6 34.Kf2 g5 35.Nd4 Ke7 36.f5 (31.243.482) 2054 TB:5
26/41 0:16 +3.66-- 23.f4 Ne7 (34.756.165) 2047 TB:5
26/41 0:17 +3.53-- 23.f4 Ne7 (35.911.344) 2047 TB:5
26/41 0:17 +3.35-- 23.f4 Ne7 (36.216.796) 2047 TB:5
26/41 0:20 +3.11-- 23.f4 Ne7 (41.420.113) 2042 TB:5

<- Lagrave's spirit is lost
26/45 0:23 +3.30++ 23.g4 (47.981.024) 2039 TB:5
26/45 0:24 +3.50 23.g4 Rxb4 24.cxb4 Qb6 25.a5 Nxd4 26.axb6 Nxc2 27.bxa7 Ke7 28.Rc1 Ra8 29.Rxc2 Rxa7 30.f4 f6 31.Bc3 Ra3 32.Nd4 fxe5 33.Nc6+ Kd6 34.Nxe5 Rxc3 35.Rxc3 Nxe5 36.fxe5+ (49.700.768) 2037 TB:5
27/44 0:28 +3.62++ 23.g4 (57.303.004) 2022 TB:5
27/44 0:29 +3.52 23.g4 Rxb4 24.cxb4 Qb6 25.a5 Nxd4 26.axb6 Nxc2 27.bxa7 Ke7 28.Rc1 Ra8 29.Rxc2 Rxa7 30.f4 f6 31.Bc3 fxe5 32.fxe5 Ra3 33.Kf2 Kf7 34.Bd4 g6 35.hxg6+ Kxg6 36.Nf4+ (59.263.706) 2021 TB:7
28/36 0:30 +3.64++ 23.g4 (62.755.825) 2025 TB:28
28/44 0:32 +3.76++ 23.g4 (66.303.182) 2030 TB:32
28/46 0:35 +3.39-- 23.g4 Nxb4 (71.295.175) 2033 TB:33
28/46 0:37 +3.49 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Rb3 31.Rc1 Nd7 32.Kf2 Nb8 33.Nf4 Nc6 34.Ng6 Ke8 35.Ne5 Nb4 36.Bxb4 (76.296.366) 2035 TB:35
29/49 0:39 +3.61++ 23.g4 (79.518.665) 2030 TB:50
29/51 0:41 +3.37-- 23.g4 Ke7 (84.950.195) 2026 TB:60
29/51 0:46 +3.19-- 23.g4 Ke7 (94.780.728) 2022 TB:60
29/51 0:49 +3.37++ 23.g4 (99.558.021) 2020 TB:60
29/51 0:50 +3.69++ 23.g4 (101.937.544) 2019 TB:61
29/51 0:51 +3.69 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Rb3 31.Kf2 g6 32.hxg6+ Kxg6 33.Rc1 Kg7 34.Nf4 Kf7 35.Ke3 Nh7 36.Ne2 (103.273.934) 2019 TB:61
30/44 0:53 +3.60 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Rb3 31.Kf2 g6 32.hxg6+ Kxg6 33.Rc1 Kg7 34.Nf4 Kf7 35.Ke3 Nh7 36.Ne2 (107.313.797) 2021 TB:68
31/53 0:58 +3.73++ 23.g4 (118.692.759) 2021 TB:69
31/53 1:03 +3.85++ 23.g4 (128.089.971) 2020 TB:91
31/53 1:03 +3.72 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Rb3 31.Kf2 Nd7 32.Ke3 Nb8 33.Kd2 Nc6 34.a5 Nb4 35.Ra1 Nd3 36.Nc1 (129.191.660) 2021 TB:92
32/51 1:15 +3.60-- 23.g4 Nxb4 (153.104.946) 2026 TB:107
32/51 1:26 +3.72++ 23.g4 (175.226.840) 2028 TB:159
32/51 1:28 +3.78 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Nh7 31.Kf2 Ng5 32.Nf4 Rb3 33.Rc1 a5 34.Ng6 Ra3 35.Bxa5 Rxa4 36.Bc3 (179.235.128) 2027 TB:159
33/51 1:32 +3.78 23.g4 Nxb4 24.cxb4 Rxb4 25.Qc3 f6 26.exf6 Nxf6 27.f3 Kf7 28.Qxb4 Qxb4 29.Bxb4 Rb8 30.Bc3 Nh7 31.Kf2 Ng5 32.Nf4 Rb3 33.Rc1 a5 34.Ng6 Ra3 35.Bxa5 Rxa4 36.Bc3 (187.529.341) 2025 TB:263


I have no idea about the quality of 23.g4 (those scores look like gibberish), but it certainly looks more stylish than 23.f4. If g4 also wins then I don't think f4 has a chance to be best in the style category.

Post Reply