Dedicated test....Nd4-f5

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
Eelco de Groot
Posts: 4186
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Dedicated test....Nd4-f5

Post by Eelco de Groot » Fri Aug 07, 2009 4:27 am

rightrook wrote:>>>Actually I am not so sure this is won for White Robert. So maybe it was not such a good analysis<<<

Maybe so....guess that is why some Novag machines take 1 or 2 hours to find it.... :o

Thanks for all the good replies!

Robert
I don't want to beat the position to death by analysis but I'm still sure how good white's winning chances are, even though Rybka predicts a draw. I have been analyzing the leaf positions of some Ancalagon's PVs with Shredder in the background, this is with an older version of the Shredder GUI so also with just an older version Shredder 9, but it should still be a lot better than just the static evaluations of the position 8-) Shredder is sometimes a bit over-optimistic but it does still see very slight winning chances it seems, in the early endgames that arise. No EGTB probes, those are still too far off, but the endgames are still complex enough to be interesting for the attacker not to give a draw and too complex to evaluate precisely even after an hour or two on my slow computer.

This is the result for the start position, build 211 is a bit slower but also more careful I hope:

[D]r1bq1r2/1p2Npkp/3p2p1/p7/3NP1nQ/1B2b3/PPP3PP/R4R1K w - -

Engine: Ancalagon 1.3 WS180BC5050 Build 211 (Athlon 2009 MHz, 256 MB)
by Romstad, Costalba, Kiiski, de Groot

2.00 0:00 +1.56 1.Rad1 Nf2+ 2.Rxf2 Bxf2 3.Qxf2 Qxe7 (14.045) 17

2.00 0:00 +5.50 1.c3 g5 2.Nef5+ Bxf5 3.Nxf5+ Kh8
4.Qxg4 (27.048) 33

3.01 0:01 +1.25 1.c3 Nf2+ (176.459) 161

3.02 0:01 +5.60 1.Rad1 g5 2.Nef5+ Bxf5 3.Nxf5+ Kh8
4.Qxg4 (179.298) 161

4.01 0:01 +5.58 1.Rad1 Nf2+ 2.Rxf2 Bxf2 3.Qxf2 Qxe7
4.Rf1 Qxe4 5.Bxf7 (257.122) 210

5.01 0:01 +1.25 1.Rad1 Nf2+ (441.277) 294

6.01 0:03 -0.39 1.Rad1 Nf2+ 2.Rxf2 Bxf2 3.Qxf2 Qxe7
4.Nb5 a4 5.Bc4 Be6 6.Be2 (1.423.490) 437

6.02 0:04 +1.50 1.Ndf5+ Bxf5 2.exf5 Nf6 3.fxg6 fxg6
4.Nxg6 hxg6 (1.852.277) 454

7.01 0:10 +0.49 1.Ndf5+ gxf5 2.exf5 Ne5 3.Rae1 f6
4.Rxe3 Qxe7 5.Rh3 Kh8 6.Kg1 (5.836.667) 535

8.01 0:18 +0.49 1.Ndf5+ gxf5 2.exf5 Ne5 3.Rae1 f6
4.Rxe3 Qxe7 5.Rh3 Kh8 6.Kg1 (11.164.067) 591

9.01 0:42 +0.66 1.Ndf5+ gxf5 2.exf5 Ne5 3.Rae1 f6
4.Nxc8 Bg5 5.Qg3 a4 6.Nxd6 axb3 (25.825.085) 607

10.01 1:22 +1.01 1.Ndf5+ (52.581.583) 640

11.01 2:11 +1.09 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Rad1 Qe8
8.Qxe8 Rfxe8 9.Bxd5 Nxd5 10.Rxd5 (80.472.197) 612

12.01 5:16 +1.09 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Rad1 Qe8
8.Qxe8 Rfxe8 9.Bxd5 Nxd5 10.Rxd5 (200.686.829) 634

13.01 14:51 +1.17 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 a3 7.bxa3 d5
8.Bb5 Kh8 (559.904.199) 628

14.01 38:29 +1.03 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bb5 Qd6
8.Rae1 a3 9.fxg6 axb2 (1.430.011.037) 619

15.01 131:15 +1.25 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bb5 a3
8.bxa3 Ra5 9.Rab1 Kg8 10.fxg6 fxg6
{Pos 1 +1.97 [D]3q1rk1/1p5p/5np1/rB1p4/8/P3Q3/P1P3PP/1R3R1K w - -}(4.611.591.443) 585

16.01 437:05 +1.11 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bd3 Kh8
8.Rab1 Re8 9.Qd4 g5 10.Rbe1 Kg8
11.Rxe8+ Nxe8
{Pos 2 +1.89 [D]r2qn1k1/1p3p1p/8/3p1Pp1/p2Q4/3B4/PPP3PP/5R1K w - -} (12.037.007.478) 458


Shredder evaluates the leafpositions still a bit higher than Ancalagon's static eval indicates so it is not sure the game would end with a bloodless draw...

Analysis of leaf-position 15 plies {Pos 1} by Shredder 9
3q1rk1/1p5p/5np1/rB1p4/8/P3Q3/P1P3PP/1R3R1K w - -

Engine: Shredder 9 UCI (64 MB)
by Stefan Meyer-Kahlen

17/39 6:46 +1.84 1.a4 Ra8 2.Qe5 Rc8 3.c4 d4 4.Rbd1 d3
5.c5 d2 (99.355.939) 244

17/41 8:21 +1.85++ 1.Qe5 (122.545.322) 244

17/44 10:03 +1.90 1.Qe5 Kg7 2.a4 Qb8 3.Qb2 Qd6 4.Rbd1 h6
5.c4 Kh7 6.Qd4 (146.977.035) 243

18/46 16:34 +2.11 1.Qe5 Kg7 2.a4 Ra8 3.Rbd1 Qb8 (241.279.578) 242

19/48 25:18 +2.12 1.Qe5 Kg7 2.a4 Ra8 3.Rbd1 Qb8 4.Rxd5 Qxe5
5.Rxe5 Rf7 6.Kg1 Rc8 7.Bd3 Nd7
8.Ree1 Ra8 (371.142.911) 244

20/52 45:48 +2.18 1.Qe5 Kg7 2.a4 Ra8 3.Rbd1 Qb8 4.Rxd5 Qxe5
5.Rxe5 Rf7 6.Kg1 Rc8 7.Bd3 Nd7 8.Rb5 Rxf1+
9.Kxf1 (675.733.042) 245

21/57 99:41 +1.95 1.Qe5 Kg7 2.a4 Qb8 3.Qe6 Ra8 4.Rb3 Qc8
5.Qe5 Qb8 6.Qc3 Qd8 (1.491.917.868) 249

21/57 148:32 +1.96++ 1.a4 (2.224.356.593) 249

21/57 162:25 +1.97 1.a4 Ra8 2.c4 d4 3.Qe5 Rc8 4.Rbd1 d3
5.Rf3 d2 6.Qe3 Rf7 7.Rxd2 Qe7 8.Kg1 (2.429.413.281) 249

best move: a3-a4 time: 171:09.828 min n/s: 249.543 CPU 49.8% n/s(1CPU): 501.090 nodes: 2.562.729.754


Analysis of leaf-position 16 plies {Pos 2} by Shredder 9

r2qn1k1/1p3p1p/8/3p1Pp1/p2Q4/3B4/PPP3PP/5R1K w - -

Engine: Shredder 9 UCI (64 MB)
by Stefan Meyer-Kahlen

18/41 7:48 +1.51 1.h4 Qf6 2.Qxf6 Nxf6 3.hxg5 Ne4
4.Re1 a3 5.Bxe4 dxe4 6.bxa3 Rxa3
7.Rxe4 Rxa2 8.c4 (100.830.353) 215

19/45 11:55 +1.55 1.h4 Qf6 2.Qxf6 Nxf6 3.hxg5 Ne4
4.Bxe4 dxe4 5.Re1 a3 6.bxa3 Ra4
7.Re3 Kg7 8.Rb3 Rc4 9.Rxb7 Rxc2
10.Rb3 (161.247.239) 225

20/44 18:41 +1.48 1.h4 Qf6 2.Qxf6 Nxf6 3.hxg5 Ne4
4.Bxe4 dxe4 5.Re1 a3 6.bxa3 Rxa3
7.Rxe4 Rxa2 8.Rc4 Kg7 9.Rc7 b5 (260.885.220) 232

20/47 21:31 +1.49++ 1.f6 (300.864.002) 232

20/47 29:20 +1.80 1.f6 a3 2.bxa3 Qd6 3.Rf5 h6 4.h4 Kf8
5.Rxd5 Qxf6 6.Qb4+ Qe7 7.Rd7 Qxb4
8.axb4 (415.004.587) 235

21/52 37:31 +1.73 1.f6 a3 2.bxa3 Qd6 3.Rf5 h6 4.h4 Kf8
5.Rxd5 Qxf6 6.Qb4+ Kg8 7.Rf5 Qc6
8.hxg5 Ra4 9.Qe7 (540.217.739) 239

22/50 51:10 +1.98++ 1.f6 a3 (747.019.445) 243

22/54 59:42 +1.98 1.f6 a3 2.bxa3 Qd6 3.Rf5 h6 4.h4 Kf8
5.Rxd5 Qxf6 (879.773.109) 245

23/58 160:14 +1.86 1.f6 Qd6 2.Rf5 h6 3.h4 Kf8 4.hxg5 hxg5
5.a3 Rd8 6.Bb5 Qg3 7.Bxa4 Nc7 8.Qe5 Qxe5
9.Rxe5 b5 10.Bb3 Ne8 11.Bxd5 (2.424.254.711) 252

24/57 216:37 +1.89 1.f6 Qd6 2.Rf5 h6 3.h4 Kf8 4.hxg5 hxg5
5.a3 Rd8 6.Bb5 Qe6 7.Re5 Qd6 8.Bxa4 Nxf6
9.Rxg5 (3.293.161.006) 253

Regards, Eelco

yanquis1972
Posts: 1762
Joined: Tue Jun 02, 2009 10:14 pm

Re: Dedicated test....Nd4-f5

Post by yanquis1972 » Fri Aug 07, 2009 6:16 am

the key move, again according to rybka is 16.01 437:05 +1.11 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bd3 Kh8
8.Rab1 Re8 9.Qd4 g5 10.Rbe1

[d]r2qr2k/1p3p1p/5n2/3p1Pp1/p2Q4/3B4/PPP3PP/4RR1K b - - 0 10[/d]
Kg8

instead Kg7 and now if Rxe8 simply Qxe8 is fine & i believe rybka is probably correct in her assessment that white has slim chances in the resulting position. the king at g7 protects the knight, enabling the queen to capture, so preventing white from occupying f6, which leads to mate threats & white mopping up on the king-side.

Steve B
Posts: 3697
Joined: Tue Jul 31, 2007 2:26 pm
Contact:

Re: Dedicated test....Nd4-f5

Post by Steve B » Fri Aug 07, 2009 10:05 am

rightrook wrote:[D]r1bq1r2/1p2Npkp/3p2p1/p7/3NP1nQ/1B2b3/PPP3PP/R4R1K w - - 0 1

Fidelity Mach 3.......7 ply ....37 minutes

TurboStar 432...about 30 minutes.

Polgar..........55 minutes.

Mondial......1 hour, and 21 min.

regards

Robert
Hi Rob

The Radio Shack Mega 2050X refuses to play Ndf5 and instead will play Rad1 after 2 hours and 9 ply
While the Radio Shack 2250XL wants to play c3 after 4 hours and 10 ply

Sigh Regards
Steve

rightrook
Posts: 1452
Joined: Wed Mar 08, 2006 7:45 pm

Re: Dedicated test....Nd4-f5

Post by rightrook » Fri Aug 07, 2009 4:21 pm

Hi...Steve....OK...thanks!

Some Novag machines as I pointed out will take about 2 hours to find it....and that's too long to wait....:-)

But they do find it...eventually.

Bye....with regards.

:D

rightrook
Posts: 1452
Joined: Wed Mar 08, 2006 7:45 pm

Re: Dedicated test....Nd4-f5

Post by rightrook » Fri Aug 07, 2009 4:24 pm

the key move, again according to rybka is 16.01 437:05 +1.11 1.Ndf5+

Thanks John...that looks and sounds like a good result to me...

regards

Robert

User avatar
Eelco de Groot
Posts: 4186
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Dedicated test....Nd4-f5

Post by Eelco de Groot » Fri Aug 07, 2009 4:58 pm

yanquis1972 wrote:the key move, again according to rybka is 16.01 437:05 +1.11 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bd3 Kh8
8.Rab1 Re8 9.Qd4 g5 10.Rbe1

[d]r2qr2k/1p3p1p/5n2/3p1Pp1/p2Q4/3B4/PPP3PP/4RR1K b - - 0 10[/d]
Kg8

instead Kg7 and now if Rxe8 simply Qxe8 is fine & i believe rybka is probably correct in her assessment that white has slim chances in the resulting position. the king at g7 protects the knight, enabling the queen to capture, so preventing white from occupying f6, which leads to mate threats & white mopping up on the king-side.
That John! I think that is correct. White can deviate earlier of course, but so far I don't see that is helping the winning chances. And 10...Kg7! was totally missed! That is the problem if there is a critical move deep or high up in the tree and you simply need enough searchdepth to see that, you have to find a way to see the position is not quiescent yet, that is the ever-present task of search 8-) And searching other moves at this critical point is not really saying much about the postion.

Shredder 9's opinion:

[D]r2qr2k/1p3p1p/5n2/3p1Pp1/p2Q4/3B4/PPP3PP/4RR1K b - -

Engine: Shredder 9 UCI (64 MB)
by Stefan Meyer-Kahlen

21/48 37:37 -0.75 1...Kg7 2.a3 Rxe1 3.Rxe1 Qd7 4.Qe3 h6
5.Qd4 Re8 6.Rf1 Re7 7.Qb6 Qc6 8.Qb4 (467.915.311) 207

best move: Kh8-g7 time: 42:02.187 min n/s: 210.391 CPU 45.4% n/s(1CPU): 463.416 nodes: 530.506.060

Doing a 21 ply search is more or less out of the question for the next few years anyway and I don't think Rybka is up to such a feat either :? ...

Regards, Eelco
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

User avatar
Eelco de Groot
Posts: 4186
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Dedicated test....Nd4-f5

Post by Eelco de Groot » Fri Aug 07, 2009 5:19 pm

rightrook wrote:the key move, again according to rybka is 16.01 437:05 +1.11 1.Ndf5+

Thanks John...that looks and sounds like a good result to me...

regards

Robert
Hi Robert,

I'm afraid that there is a slight misunderstanding, because John found an improvement in Acalagon's analysis, namely 10...Kg7! and showed that that would lead to a draw. So Rybka's analysis is still better and +1.11 is a bit too high (Ancalagon's line) But I don't mind, if people start passing off or mistaking Ancalagon's analysis as Rybka's, what else could you aim for :) But we are not quite there yet.

Following output is with the latest version, build 212, complete overhaul of the approximateEval almost and it compiled without fatal errors straight away :o

This version maybe prunes a little less with the new changes, but the Futilty pruning margins are not tuned yet and could maybe be set lower if I would want to increase the Futilty pruning rate. IIDMargin could also maybe get another value now.

A small fragment of the new code in search.cpp:

Code: Select all

   // Go with internal iterative deepening if we don't have a TT move
   if &#40;UseIIDAtNonPVNodes && ttMove == MOVE_NONE && depth >= 8*OnePly&#41;
	&#123;
		ei.futilityMargin = Value&#40;0&#41;; // Manually initialize futilityMargin
      IID_Eval = evaluate&#40;pos, ei, threadID&#41;;
		IID_Evalmargin = ei.futilityMargin;

		if &#40;IID_Eval >= beta - &#40;IIDMargin + IID_Evalmargin&#41;)
        &#123;
            IID_Eval = search&#40;pos, ss, beta, Min&#40;depth/2, depth-2*OnePly&#41;, ply, false, threadID&#41;;
            ttMove = ss&#91;ply&#93;.pv&#91;ply&#93;;
        &#125;
		approximateEval = IID_Eval; // Should be okay because ttMove == MOVE_NONE
	&#125;
Replacing this part:

Code: Select all

    // Go with internal iterative deepening if we don't have a TT move
    if &#40;UseIIDAtNonPVNodes && ttMove == MOVE_NONE && depth >= 8*OnePly &&
        evaluate&#40;pos, ei, threadID&#41; >= beta - IIDMargin&#41;
    &#123;
        search&#40;pos, ss, beta, Min&#40;depth/2, depth-2*OnePly&#41;, ply, false, threadID&#41;;
        ttMove = ss&#91;ply&#93;.pv&#91;ply&#93;;
    &#125;
[D]r1bq1r2/1p2Npkp/3p2p1/p7/3NP1nQ/1B2b3/PPP3PP/R4R1K w - -

Engine: Ancalagon 1.3 Weak Squares 180 Board Control middlegame 50 endgame 50 Build 212 (Athlon 2009 MHz, 256 MB)
by Romstad, Costalba, Kiiski, de Groot

2.00 0:00 +5.60 1.Rad1 g5 2.Nef5+ Bxf5 3.Nxf5+ Kh8
4.Qxg4 (23.145) 70

3.00 0:00 +4.96 1.Rad1 Kh8 2.Bxf7 Bh6 (178.896) 286

4.00 0:01 +5.31 1.Rad1 Nf2+ 2.Rxf2 Bxf2 3.Qxf2 Bg4
4.Rf1 Qxe7 (410.657) 410

5.01 0:01 +1.25 1.Rad1 Nf2+ (546.175) 454

6.01 0:02 -0.43 1.Rad1 Nf2+ 2.Rxf2 Bxf2 3.Qxf2 Qxe7
4.Nb5 Bg4 5.Rf1 Kh8 6.Bxf7 Qxe4
7.Nxd6 (1.408.448) 509

6.03 0:04 +1.50 1.Ndf5+ Bxf5 2.exf5 Nf6 3.fxg6 fxg6
4.Nxg6 hxg6 (2.224.898) 512

7.01 0:08 +0.56 1.Ndf5+ gxf5 2.exf5 Ne5 3.Rae1 f6
4.Nxc8 Bg5 5.Qh5 Rxc8 6.Kg1 Qb6+
7.Kh1 (5.112.140) 568

8.01 0:16 +0.94 1.Ndf5+ gxf5 2.exf5 Ne5 3.Rae1 f6
4.Rxe3 Qxe7 5.Bd5 Kh8 6.Rg3 (10.328.140) 620

9.01 0:54 +1.07 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Kh8 5.Bxd5 Qg5 6.Rad1 gxf5
7.Qxf5 Qxf5 8.Rxf5 (34.057.887) 622

10.01 1:19 +1.45 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Kh8 5.fxg6 fxg6 6.Rxf8+ Qxf8
7.Bxd5 Bh6 8.Rd1 Qg7 (51.751.368) 654

11.01 3:18 +1.05 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Rad1 a3
8.fxg6 axb2 9.gxf7 Rxf7 (130.421.954) 656

12.01 6:14 +1.09 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Rad1 Qe8
8.Qxe8 Rfxe8 9.Bxd5 Nxd5 10.Rxd5 (248.794.525) 664

13.01 18:55 +1.13 1.Ndf5+ Bxf5 2.exf5 Nf6 3.Nd5 Nxd5
4.Qh3 Nf6 5.Qxe3 a4 6.Bc4 d5 7.Bb5 Kg8
8.Qa3 Qa5 9.fxg6 Qxb5 10.Rxf6 hxg6 (710.548.039) 625


Regards, Eelco
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Post Reply