Four wild handicaps for FM level player.

Laskos · Post by **Laskos** » Sun Sep 13, 2015 5:11 pm

whereagles wrote:huh... was that for real...? K forced Rybka 3 into playing half the game without its QUEEN?

Beautiful game, actually. It was the sole game I played between them, and such a win! I followed it a bit it, around the move 45 and later they had wildly diverging evals.

lkaufman · Post by **lkaufman** » Sun Sep 13, 2015 5:11 pm

Vinvin wrote:

lkaufman wrote:...
1. Seven moves. White gets to play seven moves without crossing the midline, then it is Black's turn. It has been shown here that eight moves is really an easy win, seven is the maximum playable.

An overnight analyze after e4 d4 c4 Nf3 Nc3 O-O Bc4 :

Code: Select all

FEN: rnbqkbnr/pppppppp/8/8/2BPPP2/2N2N2/PPP3PP/R1BQ1RK1 b kq d3 0 7

Stockfish_15090721_x64_modern:
 41/68	6:19:41	208.787.418k	9.165k	+4,20	7. ... d5 8.Bb3 Nf6 9.f5 c6 10.Ne5 e6 11.fxe6 Bxe6 12.exd5 cxd5 13.Ba4+ Nc6 14.Nxc6 bxc6 15.Bxc6+ Ke7 16.Bxa8 Qxa8 17.Bg5 h6 18.Qh5 Qb7 19.Rae1 hxg5 20.Qxh8 Qxb2 21.Ne2 Qxc2 22.Ng3 Ne4 23.Rb1 Qc7 24.Nf5+ Bxf5 25.Rxf5 Qc3 26.Rb7+ Ke6 27.Qh3 Qxh3 28.Re5+ Kd6 29.gxh3 f6 30.Re8 Kc6 31.Rxa7 Kb6 32.Rf7 Bd6 33.Rxg7 f5 34.Rg6 Kc7 35.Reg8 Kd7 36.R8g7+ Kd8
 41/68	6:19:41	208.787.418k	9.165k	+3,88	7. ... e6 8.f5 Nc6 9.a3 Be7 10.d5 Nf6 11.dxc6 dxc6 12.Qxd8+ Bxd8 13.fxe6 Bxe6 14.Bxe6 fxe6 15.e5 Nd7 16.Nd4 Nf8 17.Ne4 Be7 18.b4 Rg8 19.Rb1 0-0-0 20.Bb2 Kd7 21.Rbd1 Ng6 22.c4 h6 23.Rd2 Rde8 24.Rf7 Rd8 25.Nf5+ Ke8 26.Rxg7 Rxg7 27.Nxg7+ Kf7 28.Rxd8 Bxd8 29.Nc5 Nf4 30.Nxb7 Be7
 41/68	6:19:41	208.787.418k	9.165k	+3,81	7. ... Nc6 8.f5 e6 9.a3 Be7 10.d5 Nf6 11.dxc6 dxc6 12.Qxd8+ Bxd8 13.fxe6 Bxe6 14.Bxe6 fxe6 15.e5 Nd7 16.Nd4 Nf8 17.Ne4 Be7 18.b4 Rg8 19.Rb1 0-0-0 20.Bb2 Kd7 21.Rbd1 Ng6 22.c4 h6 23.Rd2 Rdf8 24.Rxf8 Nxf8 25.g3 Ng6 26.c5 b6 27.Bc3 Rd8 28.Nf5+ Ke8 29.Rxd8+ Bxd8 30.Nxg7+ Kf7 31.Nh5 Ne7
---------------------------------------------------------------------------
 42/68	8:05:30	268.162.787k	9.206k	+4,18	7. ... d5 8.Bb3 Nf6 9.f5 e6 10.Ng5 c6 11.e5 Be7 12.fxe6 Bxe6 13.exf6 Bxf6 14.Nxe6 fxe6 15.Qh5+ g6 16.Qg4 0-0 17.Bh6 Re8 18.Rae1 Kh8 19.Rxe6 Rxe6 20.Qxe6 Bxd4+ 21.Be3 Bxe3+ 22.Qxe3 Qb6 23.Qxb6 axb6 24.Rf7 Na6 25.Rxb7 Rb8 26.Ra7 Nc5 27.Rc7 d4 28.Na4 Nxa4 29.Bxa4 c5 30.Bc6 Rf8 31.Rb7 Rf6 32.Rxb6 c4 33.c3
 42/68	8:05:30	268.162.787k	9.206k	+3,93	7. ... e6 8.f5 Nc6 9.a3 Be7 10.d5 Nf6 11.dxc6 dxc6 12.Qxd8+ Bxd8 13.fxe6 Bxe6 14.Bxe6 fxe6 15.e5 Nd7 16.Nd4 Nf8 17.Ne4 Be7 18.b4 Rd8 19.Bb2 h5 20.Rf3 Rd5 21.Re1 h4 22.Nb3 Rh5 23.Nbc5 Bxc5+ 24.Nxc5 Rd2 25.Ref1 Rf5 26.Nxe6 Rxf3 27.Nxg7+ Ke7 28.Rxf3 Rxc2 29.Bd4 b6 30.Rf6 c5 31.bxc5 Rc1+ 32.Rf1 Rxf1+ 33.Kxf1 Ne6 34.cxb6 axb6
 42/68	8:05:30	268.162.787k	9.206k	+3,93	7. ... Nc6 8.f5 e6 9.a3 Be7 10.d5 Nf6 11.dxc6 dxc6 12.Qxd8+ Bxd8 13.fxe6 Bxe6 14.Bxe6 fxe6 15.e5 Nd7 16.Nd4 Nf8 17.Ne4 Be7 18.b4 Rg8 19.Rb1 0-0-0 20.Bb2 Kd7 21.Rbd1 Ng6 22.c4 h5 23.g3 Rdf8 24.Rxf8 Nxf8 25.b5 cxb5 26.Nxb5+ Kc8 27.Nxa7+ Kb8 28.Nb5 Kc8 29.Nd4 h4 30.g4 Kd7 31.a4 Ng6 32.Kg2 Nf4+ 33.Kf3 Ng6 34.Nf5+ Kc6 35.Nxe7+ Nxe7 36.Kf4 Rf8+ 37.Kg5
---------------------------------------------------------------------------
 42/68	9:54:26	329.432.638k	9.237k	+4,18	7. ... d5 8.Bb3 Nf6 9.f5 e6 10.Ng5 c6 11.e5 Be7 12.fxe6 Bxe6 13.exf6 Bxf6 14.Nxe6 fxe6 15.Qh5+ g6 16.Qg4 0-0 17.Bh6 Re8 18.Rae1 Kh8 19.Rxe6 Rxe6 20.Qxe6 Bxd4+ 21.Be3 Bxe3+ 22.Qxe3 Qb6 23.Qxb6 axb6 24.Rf7 Na6 25.Rxb7 Rb8 26.Ra7 Nc5 27.Rc7 d4 28.Na4 Nxa4 29.Bxa4 c5 30.Bc6 Rf8 31.Rb7 Rf6 32.Rxb6 c4 33.c3
 43/68	9:54:26	329.432.638k	9.237k	+3,84	7. ... e6 8.f5 Nc6 9.a3 Nf6 10.e5 d5 11.exf6 dxc4 12.d5 gxf6 13.fxe6 Bxe6 14.Bf4 Bc5+ 15.Kh1 Qd7 16.dxc6 bxc6 17.Ne4 Be7 18.Qe1 0-0-0 19.Qa5 c5 20.Bg5 Qc6 21.Nxf6 h6 22.Bh4 Qb7 23.Rae1 Kb8 24.Qc3 Ka8 25.Ne4 Bxh4 26.Nxh4 Qb6 27.Qe5 Rb8 28.Rb1 Rhd8 29.Nf5 Bd5 30.Ne3 Qe6 31.Qxe6 Bxe6 32.Nxc5
 43/68	9:54:26	329.432.638k	9.237k	+3,80	7. ... Nc6 8.f5 e6 9.a3 Nf6 10.e5 d5 11.exf6 dxc4 12.d5 gxf6 13.fxe6 Bxe6 14.Bf4 Bc5+ 15.Kh1 Qd7 16.dxc6 bxc6 17.Ne4 Be7 18.Qe1 0-0-0 19.Qa5 c5 20.Bg5 Qc6 21.Nxf6 h6 22.Bh4 Qb7 23.Rae1 Kb8 24.Qc3 Ka8 25.Ne4 Bxh4 26.Nxh4 Qb6 27.Nf3 Rb8 28.Rb1 Rhd8 29.Ne5 Rd4 30.Qf3 Bd5 31.Nd7[/code
Despite the fact the score is about the same as a knight odd, I thing it's harder for humans to win because white have not simply to exchange a lot of pieces to have a simple winning position.[/quote]

     That's the same setup I analyzed (suggested by IM Eugene Meyer when I proposed the handicap), with similar conclusions. White wins a piece quickly; whether it's then easier or harder to win than knight odds I can't say, but in any case it's not very interesting if White can just remember a few moves and win a piece. So I'm thinking that in the upcoming GIlden match I'll give him just six moves in that game, but since this and the king in the center handicaps are probably not quite enough I'll also give hime triple time (45- + 15" to 15' + 5"), which will shorten the session considerably to minimize the fatigue problem and make it more enjoyable for the spectators. In serious matches with GMs I think we should stick to equal time to see what's a fair handicap, but it's not so important to do that in these "fun" matches with unusual handicaps.

lkaufman · Post by **lkaufman** » Sun Sep 13, 2015 5:15 pm

Laskos wrote:
whereagles wrote:huh... was that for real...? K forced Rybka 3 into playing half the game without its QUEEN?

Beautiful game, actually. It was the sole game I played between them, and such a win! I followed it a bit it, around the move 45 and later they had wildly diverging evals.

That's pretty funny, since I'm mostly responsible for the evaluation functions of both engines! Probably king safety is sufficiently different to explain this.

Jesse Gersenson · Post by **Jesse Gersenson** » Sun Sep 13, 2015 5:55 pm

Giving time handicap could be offered as something the player can take if they want. It won't change the games to shorten komodo's time by 15 minuets, for example.

thekingman · Post by **thekingman** » Mon Sep 14, 2015 7:20 am

Laskos wrote:
whereagles wrote:huh... was that for real...? K forced Rybka 3 into playing half the game without its QUEEN?

Beautiful game, actually. It was the sole game I played between them, and such a win! I followed it a bit it, around the move 45 and later they had wildly diverging evals.

I have noticed that in most of the test games I've run at odd handicaps, such as this one, the engines often have wildly diverging evals. I think these "strange" positions accentuate the strength differences between engines.

For example, suppose when Komodo and Rybka disagree, there's a 90% chance Komodo is right. But, both engines are very well tuned for "normal" positions, and so they might agree 80% of the time. But in these odd positions, maybe they only agree 50% of the time. Even if Komodo is still right for only 90% of the disagreements, by increasing their frequency, Komodo is now effectively playing much stronger.

(This also suggests that engine development would benefit from starting tests in positions where different engines have drastically divergent evals, which intuitively makes a lot of sense.)

I don't know how you might test this, except by comparing the results of Ke5 matches to "normal" +2 eval position matches. If it is true, I would suspect it to be even more true when playing against a human. If almost all humans think the advantage of Ke5 is much larger than it actually is, humans will probably also misevaluate many of the "downstream" positions. This differs greatly from f7 handicap-type positions, where the human evaluation tends to be fairly accurate.

Uri Blass · Post by **Uri Blass** » Mon Sep 14, 2015 11:08 am

thekingman wrote:
Laskos wrote:
whereagles wrote:huh... was that for real...? K forced Rybka 3 into playing half the game without its QUEEN?

Beautiful game, actually. It was the sole game I played between them, and such a win! I followed it a bit it, around the move 45 and later they had wildly diverging evals.
I have noticed that in most of the test games I've run at odd handicaps, such as this one, the engines often have wildly diverging evals. I think these "strange" positions accentuate the strength differences between engines.

For example, suppose when Komodo and Rybka disagree, there's a 90% chance Komodo is right. But, both engines are very well tuned for "normal" positions, and so they might agree 80% of the time. But in these odd positions, maybe they only agree 50% of the time. Even if Komodo is still right for only 90% of the disagreements, by increasing their frequency, Komodo is now effectively playing much stronger.

(This also suggests that engine development would benefit from starting tests in positions where different engines have drastically divergent evals, which intuitively makes a lot of sense.)

I don't know how you might test this, except by comparing the results of Ke5 matches to "normal" +2 eval position matches. If it is true, I would suspect it to be even more true when playing against a human. If almost all humans think the advantage of Ke5 is much larger than it actually is, humans will probably also misevaluate many of the "downstream" positions. This differs greatly from f7 handicap-type positions, where the human evaluation tends to be fairly accurate.

Humans have little experience with this type of advantage(king at e5)

Maybe if we make a tournament from the following position between humans then humans are going to learn something

[D]rnbq1bnr/pppppppp/8/4k3/8/8/PPPPP2P/RNBQKBNR w KQ - 0 1

Laskos · Post by **Laskos** » Mon Sep 14, 2015 1:41 pm

Uri Blass wrote:
Humans have little experience with this type of advantage(king at e5)

Maybe if we make a tournament from the following position between humans then humans are going to learn something

[D]rnbq1bnr/pppppppp/8/4k3/8/8/PPPPP2P/RNBQKBNR w KQ - 0 1

Although long search with both Komodo and Stockfish seems to give balanced position, it's not clear whether it's truly balanced at long time control. At shorter 1'+0.6'' games, the position seems to heavily favor White:

Komodo 9.2 self-games

Code: Select all

Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;     30 &#40;60.0 %)
Black Wins   &#58;      4 ( 8.0 %)
Draws        &#58;     16 &#40;32.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 76.0 %
Black Perf.  &#58; 24.0 %

Komodo 9.2 - SF 30 Aug. 2015

Code: Select all

Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;     32 &#40;64.0 %)
Black Wins   &#58;      5 &#40;10.0 %)
Draws        &#58;     13 &#40;26.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 77.0 %
Black Perf.  &#58; 23.0 %

lkaufman · Post by **lkaufman** » Mon Sep 14, 2015 3:15 pm

Laskos wrote:
Uri Blass wrote:
Humans have little experience with this type of advantage(king at e5)

Maybe if we make a tournament from the following position between humans then humans are going to learn something

[D]rnbq1bnr/pppppppp/8/4k3/8/8/PPPPP2P/RNBQKBNR w KQ - 0 1
Although long search with both Komodo and Stockfish seems to give balanced position, it's not clear whether it's truly balanced at long time control. At shorter 1'+0.6'' games, the position seems to heavily favor White:

Komodo 9.2 self-games
Code: Select all
Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;     30 &#40;60.0 %)
Black Wins   &#58;      4 ( 8.0 %)
Draws        &#58;     16 &#40;32.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 76.0 %
Black Perf.  &#58; 24.0 %
Komodo 9.2 - SF 30 Aug. 2015
Code: Select all
Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;     32 &#40;64.0 %)
Black Wins   &#58;      5 &#40;10.0 %)
Draws        &#58;     13 &#40;26.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 77.0 %
Black Perf.  &#58; 23.0 %

Instead of removing the two White pawns, if we remove the b1 knight we get a position that Komodo evaluates as winning or nearly winning for Black. However I think that most grandmasters would choose to play White given the choice of color. We might learn something from playing this position out.

Laskos · Post by **Laskos** » Mon Sep 14, 2015 8:26 pm

lkaufman wrote:
Instead of removing the two White pawns, if we remove the b1 knight we get a position that Komodo evaluates as winning or nearly winning for Black. However I think that most grandmasters would choose to play White given the choice of color. We might learn something from playing this position out.

Uri's position (two pawns off) at five times longer time control (5'+3'') is favoring white even more, although engines think it's pretty balanced:

Komodo self-games:

Code: Select all

Games        &#58;     30 &#40;finished&#41;

White Wins   &#58;     23 &#40;76.7 %)
Black Wins   &#58;      3 &#40;10.0 %)
Draws        &#58;      4 &#40;13.3 %)
Unfinished   &#58;      0

White Perf.  &#58; 83.3 %
Black Perf.  &#58; 16.7 %

With knight b1 off, engines (Komodo and SF) evaluate it correctly as almost won for black, in 50 games of Komodo at 1'+0.6'':

Code: Select all

Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;      3 ( 6.0 %)
Black Wins   &#58;     42 &#40;84.0 %)
Draws        &#58;      5 &#40;10.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 11.0 %
Black Perf.  &#58; 89.0 %

Humans, on the other hand, favor white.

Uri Blass · Post by **Uri Blass** » Mon Sep 14, 2015 10:07 pm

Laskos wrote:
lkaufman wrote:
Instead of removing the two White pawns, if we remove the b1 knight we get a position that Komodo evaluates as winning or nearly winning for Black. However I think that most grandmasters would choose to play White given the choice of color. We might learn something from playing this position out.
Uri's position (two pawns off) at five times longer time control (5'+3'') is favoring white even more, although engines think it's pretty balanced:

Komodo self-games:
Code: Select all
Games        &#58;     30 &#40;finished&#41;

White Wins   &#58;     23 &#40;76.7 %)
Black Wins   &#58;      3 &#40;10.0 %)
Draws        &#58;      4 &#40;13.3 %)
Unfinished   &#58;      0

White Perf.  &#58; 83.3 %
Black Perf.  &#58; 16.7 %
With knight b1 off, engines (Komodo and SF) evaluate it correctly as almost won for black, in 50 games of Komodo at 1'+0.6'':
Code: Select all
Games        &#58;     50 &#40;finished&#41;

White Wins   &#58;      3 ( 6.0 %)
Black Wins   &#58;     42 &#40;84.0 %)
Draws        &#58;      5 &#40;10.0 %)
Unfinished   &#58;      0

White Perf.  &#58; 11.0 %
Black Perf.  &#58; 89.0 %
Humans, on the other hand, favor white.

I am not sure about the opinion of humans.

Without computers I will probably evaluate it as unclear because I have no experience in this type of position.

Even when humans sacrifice for attack it usually happens
after developing some pieces and here the fact that white still did not develop pieces is against white.

Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.

Re: Four wild handicaps for FM level player.