Stockfish 162s dc

[Dann Corbit]

64 bit Windows binary only:
http://cap.connx.com/chess-engines/new- ... dc.exe.bz2

Source, project file, etc:
http://cap.connx.com/chess-engines/new- ... 162s-dc.7z

This version has three new UCI parameters:

Code: Select all

option name Linear Factor in Centipawns type spin default 18 min -500 max 500
option name Constant Factor in Centipawns type spin default 310 min -1500 max 1500
option name Log Factor in Centipawns type spin default 20 min -5000 max 5000

In the program, they are multiplied by .01 because they are scaled in centipawns and then used like this:

Code: Select all

#ifdef SMOOTH_REDUCTION
		double delta = approximateEval - beta;
		delta = max(delta, 1.0);
		double ddepth = double(depth);
		double r = Linear_Factor * ddepth + Constant_Factor + log(delta) * Log_Factor;
		r = r > ddepth ? ddepth : r;
		int R = int(r);
#else

If Linear factor is 18, Constant factor is 310, and Log factor is 20 then it will behave like the old smooth scaling stockfish.

The defaults are the above values.

[meschle]

Many thanks dann - happy new year

[Eelco de Groot]

64 bit Windows binary only:
http://cap.connx.com/chess-engines/new- ... dc.exe.bz2

Source, project file, etc:
http://cap.connx.com/chess-engines/new- ... 162s-dc.7z

This version has three new UCI parameters:

Code: Select all

option name Linear Factor in Centipawns type spin default 18 min -500 max 500
option name Constant Factor in Centipawns type spin default 310 min -1500 max 1500
option name Log Factor in Centipawns type spin default 20 min -5000 max 5000

In the program, they are multiplied by .01 because they are scaled in centipawns and then used like this:

Code: Select all

#ifdef SMOOTH_REDUCTION
		double delta = approximateEval - beta;
		delta = max(delta, 1.0);
		double ddepth = double(depth);
		double r = Linear_Factor * ddepth + Constant_Factor + log(delta) * Log_Factor;
		r = r > ddepth ? ddepth : r;
		int R = int(r);
#else

If Linear factor is 18, Constant factor is 310, and Log factor is 20 then it will behave like the old smooth scaling stockfish.

The defaults are the above values.

Thanks Dann!

There is now a 32-bit version as well, made from your sources, with some changes in the code. I added

Code: Select all

#define SMOOTH_REDUCTION 1

to the top of search.cpp, as you instructed Howard I believe. I am not sure I understand where the "1" is meant to go, have never used macros? I have not looked it up, or really tested the reductions yet but I assume the macro is working. My version is probably a bit weaker than your original 1.6.3s(dc) (that's my name for it) because of some other experiments again but I do like the raw branching speed of the new Stockfish! It reaches 27 whole plies in one hour on my PC It's the return of Michael Schumacher to Formula 1!! Racing through the plies! Is that what the one stands for? Or maybe it's like Luke Skywalker in his landracer, to be "One":




To be, what one can not be. Obe Wan Kenobe..

Did you know there actually is a little annual plant species in Illinois and the Midwest of the U.S. that is called Obe-Wan-Conobea? "Derivation of the common name is uncertain" it says in the description. That's a nice one




[d]rn3rk1/p3bppp/4p3/1Qp5/3q1B2/6P1/P3PPNP/1RR3K1 b - -

Engine: Stockfish 1.6.3s(dc) (Athlon 2009 MHz, 256 MB)
by Tord Romstad, Marco Costalba, Joona Kiiski. Modifications by Dann Corbit

1.00 0:00 +0.56 1...Nd7 (216) 0

2.00 0:00 +0.36 1...Nd7 2.Ne3 (265) 0

3.00 0:00 -3.67 1...Nd7 2.Rd1 a6 3.Qxd7 (1.043) 3

3.00 0:00 -1.45 1...c4 2.Rxc4 Qd7 (1.563) 4

3.00 0:00 -1.21 1...Qd7 2.Qb7 Qxb7 3.Rxb7 (2.334) 7

4.00 0:00 -0.08 1...Qd7 2.Qb7 Qxb7 3.Rxb7 Nc6 (4.070) 12

4.00 0:00 +0.24 1...a6 2.Qb7 Nd7 3.Ne3 (5.051) 15

5.00 0:00 -4.24 1...a6 2.Qb7 Nd7 3.Rd1 Qc3 4.Rxd7 (6.551) 19

5.00 0:00 -0.28 1...Qd7 2.Qb7 Qxb7 3.Rxb7 Nc6 4.Ne3 (7.724) 23

6.00 0:00 -0.36 1...Qd7 2.Qb7 Qxb7 3.Rxb7 Re8 4.Ne3 Nc6 (10.406) 30

6.00 0:00 +0.20 1...Rc8 2.Rd1 Qf6 3.Bxb8 Rcxb8
4.Qxb8+ Rxb8 5.Rxb8+ Bf8 (16.532) 45

7.00 0:00 +0.12 1...Rc8 2.Rd1 Qe4 3.f3 Qg6 4.Bxb8 Rcxb8
5.Qxb8+ Rxb8 6.Rxb8+ Bf8 (29.145) 77

8.00 0:00 -0.36 1...Rc8 2.Rd1 Qf6 3.Bxb8 a6 4.Qb7 Rcxb8
5.Qxb8+ Rxb8 6.Rxb8+ Bf8 7.Ne3 Qe5
8.Rb7 c4 9.Nxc4 Qxe2 (49.551) 121

9.00 0:00 -0.36 1...Rc8 2.Rd1 Qf6 3.Bxb8 a6 4.Qb7 Rcxb8
5.Qxb8+ Rxb8 6.Rxb8+ Bf8 7.Ne3 Qe5
8.Rb7 c4 9.Nxc4 Qxe2 (82.045) 174

10.00 0:00 -0.84-- 1...Rc8 2.Rd1 Qe4 3.Bxb8 Rcxb8
4.Qxb8+ Rxb8 5.Rxb8+ Bf8 6.Rdd8 (113.573) 220

10.00 0:00 -0.32 1...Qe4 2.f3 Qc6 3.Qb7 Re8 4.Be3 Qxb7
5.Rxb7 Na6 6.Nf4 e5 (136.190) 241

11.00 0:00 -0.52-- 1...Qe4 2.f3 Qd5 3.e4 Qc6 4.Qb7 Qxb7
5.Rxb7 Nc6 6.Rc7 (224.908) 319

11.00 0:00 -0.12++ 1...Rc8 2.Rd1 Qc3 3.Qb7 Nc6 4.Rd3 Qf6 (288.793) 355

12.00 0:01 0.00 1...Rc8 2.Rd1 Qc3 3.Qb7 Nc6 4.Rd7 Qf6
5.Ne3 Qg6 6.Nc4 Bf6 7.Rbd1 Qc2 8.Ne3 Qg6
9.Nc4 (740.595) 459

13.01 0:03 -0.56-- 1...Rc8 2.Rd1 Qc3 3.Bxb8 Rcxb8
4.Qxb8+ Rxb8 5.Rxb8+ Bf8 6.Ne3 Qa3
7.Re8 Qxa2 8.Rdd8 (1.828.394) 513

13.24 0:04 -0.20 1...Qd5 2.Ne3 Qc6 3.Qb7 Qxb7 4.Rxb7 Nc6
5.Ng4 Rfd8 6.Ne5 Rab8 7.Nxc6 Rxb7
8.Nxd8 Bxd8 9.Rxc5 Kf8 10.Kg2 Rb2 (2.343.615) 531

14.01 0:05 -0.12 1...Qd5 2.Ne3 Qc6 3.Qb7 Qxb7 4.Rxb7 Nc6
5.Ng4 Rfd8 6.Ne5 Rab8 7.Nxc6 Rxb7
8.Nxd8 Bxd8 9.Rxc5 Kf8 10.Rc2 Ke7
11.Kg2 Ba5 12.Kf3 (2.850.386) 541

15.01 0:08 -0.16 1...Qd5 2.Qb7 Qxb7 3.Rxb7 Nc6 4.Ne1 Rfd8
5.Be3 Rab8 6.Rc7 Rbc8 7.Rxc8 Rxc8
8.Nd3 Rd8 9.Bxc5 Bxc5 10.Rxc5 (4.620.540) 568

16.01 0:16 -0.08 1...Qd5 2.Ne3 Qc6 3.Nc4 Qxb5 4.Rxb5 Nc6
5.Rb7 Rfd8 6.Kg2 Kf8 7.Rcb1 Rd4
8.Ne5 Nxe5 9.Bxe5 Rb4 10.R1xb4 cxb4
11.Rd7 f6 (9.645.654) 599

17.01 0:22 -0.04 1...Qd5 2.Ne3 Qc6 3.Qb7 Qxb7 4.Rxb7 Nc6
5.Nc4 Rfd8 6.Kg2 Kf8 7.h3 f6 8.Be3 Rab8
9.Rc7 Rdc8 10.Rd7 Nd4 11.Bxd4 cxd4
12.Rxd4 (13.926.652) 611

18.01 0:40 -0.24-- 1...Qd5 2.Bxb8 Raxb8 3.Qxb8 Rxb8
4.Rxb8+ Bf8 5.Ne3 Qxa2 6.Rxc5 Qxe2
7.Rcc8 Qe1+ 8.Kg2 (25.369.130) 620

19.01 0:49 -1.09-- 1...Qd5 2.Bxb8 Raxb8 3.Qxb8 Rxb8
4.Rxb8+ Bf8 5.Ne3 Qxa2 6.Rxc5 Qxe2
7.Rcc8 Qe1+ 8.Kg2 (31.064.949) 630

19.04 1:20 -0.64 1...e5 2.Be3 Qe4 3.f3 Qg6 4.Bxc5 a6
5.Qb7 Bxc5+ 6.Rxc5 Nd7 7.Rc6 Nf6
8.Rd1 Rab8 9.Qxa6 Rb1 10.Rcc1 Rb2
11.Ne3 Re8 (50.994.952) 634

19.17 1:35 -0.24 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 g5 6.Bc7 Rdc8 7.Kf1 f6 8.Rc3 c4
9.Rxc4 Nb6 10.Rd4 Nd5 11.Ba5 Kf7
12.e4 (60.464.078) 635

20.01 2:08 -0.28 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Bc7 Re8
5.Nf4 Rec8 6.Nd3 Kf8 7.Kg2 f6 8.f4 Ke8
9.Kf2 Kf8 10.Kf3 c4 11.Rxc4 Kf7
12.Rc1 Ba3 13.Rc3 (81.881.594) 637

21.01 3:06 -0.16 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Bc7 Rdc8
5.Ne3 Nf6 6.Nc4 Nd5 7.Be5 Bf8 8.e4 Nb6
9.Bc7 Nxc4 10.Rxc4 g5 11.Rc1 Kg7
12.f4 Kg6 13.fxg5 Kxg5 14.Bf4+ Kg6 (119.719.930) 643

22.01 5:09 0.00 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Bc7 Rdc8
5.Ne3 Nf6 6.Nc4 Nd5 7.Be5 Bf8 8.Bd6 Bxd6
9.Nxd6 Rc6 10.Nxf7 Rb6 11.Rd7 Nf6
12.Rc7 Nd5 13.Rd7 (199.298.023) 644

23.01 8:17 0.00 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Bc7 Rdc8
5.Ne3 Nf6 6.Nc4 Nd5 7.Be5 Bf8 8.Bd6 Bxd6
9.Nxd6 Rc6 10.Nxf7 Rb6 11.Rd7 Nf6
12.Rc7 Nd5 13.Rd7 (320.242.846) 643

24.01 12:00 -0.12 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 Ke8 6.Bc7 Rdc8 7.f4 Bd8 8.Bxd8 Kxd8
9.Kf2 Ke8 10.Rc4 Rcb8 11.Nxc5 Nxc5
12.Rxb8+ Rxb8 13.Rxc5 Kd7 14.Ke3 h6 (460.870.902) 640

25.01 16:36 -0.04 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 Ke8 6.Bc7 Rdc8 7.f4 Bd8 8.Bxd8 Kxd8
9.Kf2 Ke8 10.Rc4 Rcb8 11.Nxc5 Nxc5
12.Rxb8+ Rxb8 13.Rxc5 Kd7 14.Ke3 Rb7 (635.626.942) 637

26.01 26:30 -0.04 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 Ke8 6.Bc7 Rdc8 7.f4 Bd8 8.Bxd8 Kxd8
9.Kf2 Ke8 10.Rc4 Rcb8 11.Nxc5 Nxc5
12.Rxb8+ Rxb8 13.Rxc5 Kd7 14.Ke3 Rb7 (1.019.838.409) 641

27.01 58:15 0.00 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 Ke8 6.Bc7 Rdc8 7.f4 Bd8 8.Bxd8 Kxd8
9.Kf2 Ke8 10.Ne5 Nxe5 11.fxe5 a6
12.Rc3 c4 13.Rf3 c3 14.Rfxf7 c2 (2.243.288.948) 641

28.01 81:49 0.00 1...Qd7 2.Qxd7 Nxd7 3.Rb7 Rfd8 4.Ne1 Kf8
5.Nd3 Ke8 6.Bc7 Rdc8 7.f4 Bd8 8.Bxd8 Kxd8
9.Kf2 Ke8 10.Ne5 Nxe5 11.fxe5 a6
12.Rc3 c4 13.Rf3 c3 14.Rfxf7 c2 (3.149.524.868) 641
best move: Qd4-d7 time: 93:50.031 min n/s: 643.808 nodes: 3.624.630.307

Eelco

[Jim Walker]

Hello Dann,
I tried to test this version of Stockfish but I'm having trouble on my dual machine. Previous versions of Stockfish I tried worked ok. This version runs fine on my Quad laptop with Win 7. The dual has Windows XP pro x64. I get an error the program failed to install correctly when I try to creat a UCI engine in the Fritz GUI. Any ideas why this would work on the Quad but not on the dual?
Jim

[swami]

Where can I download the 32 bit file?

[Volker Pittlik]

I get some absurd evaluation when I test this version with "./stockfish
bench 128 1".

For example for testposition 2

Code: Select all

Searching: r4rk1/1b2qppp/p1n1p3/1p6/1b1PN3/3BRN2/PP3PPP/R2Q2K1 b - -
infinite: 1 ponder: 0 time: 0 increment: 0 moves to go: 0
 2     +0.48   00:00      229 Rad8 a3 
 3     +0.16   00:00      950 Rad8 Neg5 Nxd4 Bxh7+ Kh8 
 4     +0.16   00:00     1558 Rad8 Neg5 Nxd4 Bxh7+ Kh8 
 5     +0.16   00:00     3490 Rad8 a3 Ba5 Neg5 Nxd4 Bxh7+ Kh8 
 6     +0.20   00:00     8864 Rad8 a3 Ba5 Nc5 Bb6 Qc2 
 7  <  +0.00   00:00    11766 Rad8 a3 Ba5 Nc5 Bb6 Qb1 
 8   21691754.00   00:00    49154 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Qxc6 Rc1 
 9  >  +0.20   00:00   114304 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 f4 
10     +0.24   00:00   193970 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Be4 Rd2 
11   21691754.00   00:00   361898 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Rc8 f3 Rfd8 Qd2 Qb6 Re1 
12   21691754.00   00:01   927403 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Rc8 f3 Rfd8 Qd2 Qb6 Re1 
13   21691754.00   00:02    2232k Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Be4 
                              Qb3 Bd5 Qb4 Qxb4 axb4 Bb7 Rd3 Be4 Rdd1 Rd6 
13  >  +0.24   00:04    3113k Rfd8 a3 Bd6 d5 exd5 Nxd6 Qxd6 Ng5 g6 Qf3 Qf8 Qf6 
                              h6 Rh3 Qg7 
14     +0.24   00:09    6939k Rfd8 a3 Bd6 Nxd6 Rxd6 a4 g6 axb5 Nxd4 bxa6 Nxf3+ 
                              Rxf3 Bxf3 Qxf3 Rad8 a7 Rxd3 a8=Q Rxf3 Qxf3 h6 Qc3 
                              e5 
15     +0.32   00:19   14865k Rfd8 a3 Bd6 d5 exd5 Nxd6 Qxd6 Ng5 g6 Qf3 Rf8 Qh3 
                              h5 Rae1 d4 R3e2 Qf4 Qd7 Qxg5 Qxb7 Rfc8 
16  <21691754.00   00:25   19606k Rfd8 a3 Bd6 d5 exd5 Neg5 Qf6 Bxh7+ Kf8 Nxf7 Qxf7 
                              Ng5 Bxh2+ Kh1 
16   21691754.00   00:44   32969k f5 Nc3 Bd6 d5 Nb4 Rxe6 Qd7 Nd4 Nxd5 Nxd5 Bxd5 
                              Bxf5 Bxh2+ Kxh2 Rxf5 Nxf5 Bxe6 Qxd7 Bxd7 g4 
Nodes: 44790182

For some position these "221691754.00" values are there all the time for other positions they never occur.

In terms of nps your version seems to be faster. So far I cannot see differences in playing strength between yours and the standard version.

However, these strange values seems to be suspect.

vp

BTW: Again I found that a gcc compilation (linux 64-bits) is faster than a pgo compilation with the intel compiler.

[Eelco de Groot]

I get some absurd evaluation when I test this version with "./stockfish
bench 128 1".

For example for testposition 2

Code: Select all

Searching: r4rk1/1b2qppp/p1n1p3/1p6/1b1PN3/3BRN2/PP3PPP/R2Q2K1 b - -
infinite: 1 ponder: 0 time: 0 increment: 0 moves to go: 0
 2     +0.48   00:00      229 Rad8 a3 
 3     +0.16   00:00      950 Rad8 Neg5 Nxd4 Bxh7+ Kh8 
 4     +0.16   00:00     1558 Rad8 Neg5 Nxd4 Bxh7+ Kh8 
 5     +0.16   00:00     3490 Rad8 a3 Ba5 Neg5 Nxd4 Bxh7+ Kh8 
 6     +0.20   00:00     8864 Rad8 a3 Ba5 Nc5 Bb6 Qc2 
 7  <  +0.00   00:00    11766 Rad8 a3 Ba5 Nc5 Bb6 Qb1 
 8   21691754.00   00:00    49154 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Qxc6 Rc1 
 9  >  +0.20   00:00   114304 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 f4 
10     +0.24   00:00   193970 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Be4 Rd2 
11   21691754.00   00:00   361898 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Rc8 f3 Rfd8 Qd2 Qb6 Re1 
12   21691754.00   00:01   927403 Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Bb7 
                              Rd3 Rc8 f3 Rfd8 Qd2 Qb6 Re1 
13   21691754.00   00:02    2232k Rad8 a3 Bd6 Nxd6 Qxd6 Be4 f5 Bxc6 Bxc6 Ne5 Be4 
                              Qb3 Bd5 Qb4 Qxb4 axb4 Bb7 Rd3 Be4 Rdd1 Rd6 
13  >  +0.24   00:04    3113k Rfd8 a3 Bd6 d5 exd5 Nxd6 Qxd6 Ng5 g6 Qf3 Qf8 Qf6 
                              h6 Rh3 Qg7 
14     +0.24   00:09    6939k Rfd8 a3 Bd6 Nxd6 Rxd6 a4 g6 axb5 Nxd4 bxa6 Nxf3+ 
                              Rxf3 Bxf3 Qxf3 Rad8 a7 Rxd3 a8=Q Rxf3 Qxf3 h6 Qc3 
                              e5 
15     +0.32   00:19   14865k Rfd8 a3 Bd6 d5 exd5 Nxd6 Qxd6 Ng5 g6 Qf3 Rf8 Qh3 
                              h5 Rae1 d4 R3e2 Qf4 Qd7 Qxg5 Qxb7 Rfc8 
16  <21691754.00   00:25   19606k Rfd8 a3 Bd6 d5 exd5 Neg5 Qf6 Bxh7+ Kf8 Nxf7 Qxf7 
                              Ng5 Bxh2+ Kh1 
16   21691754.00   00:44   32969k f5 Nc3 Bd6 d5 Nb4 Rxe6 Qd7 Nd4 Nxd5 Nxd5 Bxd5 
                              Bxf5 Bxh2+ Kxh2 Rxf5 Nxf5 Bxe6 Qxd7 Bxd7 g4 
Nodes: 44790182

For some position these "221691754.00" values are there all the time for other positions they never occur.

In terms of nps your version seems to be faster. So far I cannot see differences in playing strength between yours and the standard version.

However, these strange values seems to be suspect.

vp

BTW: Again I found that a gcc compilation (linux 64-bits) is faster than a pgo compilation with the intel compiler.

Hello Volker,

On behalf of Dann here if he maybe is somewhere on a party or New Year's visit: Thanks for your bugreport Could there possibly be some sort of conflict when using <cmath> in the gcc compiler? That is the only difference I can think of that could have caused something like this, from what I have seen in Dann's version of the code, but all the changed indentations make it difficult to be sure -There must be a switch in WinMerge that I can use to compare the codechanges better but I only know you can ignore white space. -

At the top of search.cpp there is

Code: Select all

#include <cmath>

that Dann needs for the log function. It is a C++ library that should work also with gcc, but maybe it changes the sizes of some datatypes or something? It almost seems as if the Score datatype breaks down when you see such evals. Does this also occur in an Intel compile? Could a compiler switch to include more precise floating point calculations make a difference? Such a switch exists for the MSVC compiler but I don't know about gcc or Intel.

I will try to reproduce it with the MSVC 2005 build I made, have not seen any such evals as you describe.

Regards, Eelco

[Jim Walker]

I cannot get this version to run on my dual/XP pro x64 computer but on my Quad I got 104 games last night vs Rybka 3 (1cpu) @ time of 3'+1". Stockfish scored +37 -27 =40 for a plus Elo of 31 over Rybka 3 (1 cpu). That's slightly better than Robbo scored so I will play it against Robbo today.
Jim