Stockfish 1.7

[Frank Quisinsky]

Hi Marco, Tord und Joona,

thanks for the new version!

I think thousands of computer chess fans will follow in the next days / weeks the first results.

Again, interesting is the interview with the Stockfish team too:

English version:
http://www.schach-welt.de/spezial/compu ... -team.html

German version:
http://www.schach-welt.de/spezial/compu ... am-dt.html

Best
Frank

[mcostalba]

The updated search is a very nice read. Would you be willing to share some of your thoughts on the search changes and your plans for the future? I'm especially interested in the decision to do a full evaluation at every node, the new futility matrix, and the blocking moves pruning condition in qsearch. Open source is great, but I really love hearing the perspectives of authors and the ideas behind the code.

I can make an educated guess.

Having the full evaluation available at every node, if it can be done without a big performance impact (for example by restructuring the search) allows a multitude of extra pruning techniques to be tried.

I'm guessing they are currently trying to find the optimal parameters for those techniques, and I'd expect the next version to take a big leap.

Yes, I agree. Actually we don't have an auto-tuning system for search parameters as we have for evaluation ones, so it will be much more painful to find the good ones.

Aslo my thinking is that at this point we need some evaluation improvments to unleash the search potential. Now that we evaluate at every node a change in evaluation could benefit the search, especially the pruning accuracy. IMHO the pruning accuracy is the key to gain some ELO because we _already_ prune a lot (as you can guess from the average search depth that is much higher then in 1.6 series) and if we are able to improve accuracy even without pruning more we could achieve good results.

[lech]

Maybe you should know that the older version v1.6.3 JA works with Fancy,
a special software for solving chess problems. (Popeye would be another
option, too) I have got these informations from here:

http://free.of.pl/c/ccintorun/g/fancy.htm

Norbert

Author of Fancy is too busy and drinks too much beer to make a better interface (double pipe).
Is Stockfish 1.7 a good news for composers and fans of studies?

[Uri Blass]

In this release we have deeply rewritten the search, we don't expect a big ELO increase from 1.6.X series (if any

I wonder what is the definition of big ELO increase

So far I can see more than 60 elo based on the IPON list

I think that even if the real elo improvement is only 40 elo then it can be described as a big ELO increase.


STOCK17_1

Stockfish 1.7 JA - Rybka 3 mp (2907) 18.5 - 13.5 57.81% Perf=2961
Stockfish 1.7 JA - Naum 4.2 (2823) 19.0 - 13.0 59.38% Perf=2888
Stockfish 1.7 JA - Deep Shredder 12 (2801) 18.0 - 14.0 56.25% Perf=2844
Stockfish 1.7 JA - Komodo64 1.0 JA (2781) 21.5 - 10.5 67.19% Perf=2905
Stockfish 1.7 JA - Zappa Mexico II (2709) 24.5 - 7.5 76.56% Perf=2914
Stockfish 1.7 JA - Protector 1.3.2 JA (2698) 23.5 - 7.5 75.81% Perf=2896
Stockfish 1.7 JA - Onno-1-1-1 (2682) 22.0 - 9.0 70.97% Perf=2837
Stockfish 1.7 JA - Spark-0.3 VC(a) (2672) 25.5 - 5.5 82.26% Perf=2938
Stockfish 1.7 JA - Deep Sjeng WC2008 (2671) 25.5 - 5.5 82.26% Perf=2937
Stockfish 1.7 JA - Toga II 1.4 beta5c BB (2661) 25.5 - 5.5 82.26% Perf=2927
223.5 - 91.5 70.95% Perf=2896




315 out of 1000 games played

2 Stockfish 1.6.x JA : 2833 10 10 3200 65.4 % 2722 39.2 %

[nepossiver]

Was Dann Corbit smooth scaling introduced into the source? I remember the few tests done showed either no difference or a 20-30 increase in ELO (which is what I found in 1+1 games).

[James Constance]

The windows 32 exe isn't working for me (i'm using windows XP). I've not had problems with any previous version. It installs OK - comes up with the programmer names as per usual - but doesn't start infinite analysis or playing - just dies and freezes the program (shredder classic or fritz).

Is it just me?

[yanquis1972]

sorry for the dumb question, but how do i get it to use 8 threads? i have max and threads set to 8 but the task manager reports 50% CPU usage.

[Aaron Becker]

Was Dann Corbit smooth scaling introduced into the source? I remember the few tests done showed either no difference or a 20-30 increase in ELO (which is what I found in 1+1 games).

Here is the new nullmove depth reduction calculation:

Code: Select all

        // Null move dynamic reduction based on depth
        int R = 3 + (depth >= 5 * OnePly ? depth / 8 : 0);

        // Null move dynamic reduction based on value
        if (refinedValue - beta > PawnValueMidgame)
            R++;

I don't know how close this is to Dann's formulation.

[snino64]

I'm trying stockfish 1.7 and encountered a problem with the following position: 8/B7/1P6/1B6/8/8/3K1ppp/6kq w - - 0 1
with multipv=1
2 +0.00 1.b7 (0.00)
3 +0.00 1.b7 (0.00)
4 +0.00 1.b7 (0.00)
5 +0.00 1.b7 (0.00)
6 +0.00 1.b7 (0.00)
7 +0.00 1.b7 (0.00)
8 +0.00 1.b7 (0.00)
9 +0.00 1.b7 (0.00)
10 +0.00 1.b7 (0.00)
11 +0.00 1.b7 (0.00)
12 +0.00 1.b7 (0.00)
13 +0.00 1.b7 (0.03)
14 +0.00 1.b7 (0.03)
15 +0.00 1.b7 (0.03)
16 +0.00 1.b7 (0.03)
17 +0.00 1.b7 (0.03)
18 +0.00 1.b7 (0.03)
19 +0.00 1.b7 (0.03)
20 +0.00 1.b7 (0.04)
21 +0.00 1.b7 (0.04)
22 +0.00 1.b7 (0.05)
23 +0.00 1.b7 (0.07)
24 +0.00 1.b7 (0.09)
25 +0.00 1.b7 (0.12)
26 +0.00 1.b7 (0.15)
27 +0.00 1.b7 (0.18)
28 +0.00 1.b7 (0.22)
29 +0.00 1.b7 (0.32)
30 +0.00 1.b7 (0.41)
31 +0.00 1.b7 (0.50)
32 +0.00 1.b7 (0.96)
33 +0.00 1.b7 (1.13)
34 +0.00 1.b7 (1.70)
.... Stockfish depth has reached 34 and have not found the mate.
with Multipv=5,
1 20 M7 1.Be2 f1=N+ 2.Kd3 Kf2 3.b7+ Ne3 4.Bxe3+ Kg3 5.b8=Q+ Kh3 6.Qc8+ Kg3 7.Qg4#
2 20 M9 1.Kc3 f1=R 2.b7+ Rf2 3.Bc6 Kf1 4.b8=Q g1=Q 5.Qb1+ Ke2 6.Qc2+ Ke1 7.Qc1+ Ke2 8.Qd2+ Kf1 9.Qd1#
3 20 +0.00 1.b7
4 19 -10.78 1.Kc2 f1=Q 2.b7+ Qf2+ 3.Bxf2+ Kxf2 4.b8=Q Qd1+ 5.Kxd1 g1=Q+ 6.Kc2 Qg6+ 7.Kb2 Qf6+ 8.Kc2 h1=Q 9.Qa7+ Kg3 10.Qc7+ Qf4 11.Qc3+ Qhf3 12.Bd3 Qg2+ 13.Kb3 Qb7+ 14.Kc2 Qf2+ 15.Be2+ Kg2 16.Qc4 Qf6 17.Qb5 Qc7+ 18.Bc4 Qf2+ 19.Kb3 Qf3+ 20.Kc2 Qe2+ 21.Kb3 Qd1+ 22.Kc3 Qc1+ 23.Kd3 Qg3+ 24.Ke4 Qf3+ 25.Ke5 Qcf4+ 26.Ke6
5 19 -19.03 1.Bd3 f1=Q 2.b7+ Qf2+ 3.Bxf2+ Kxf2 4.b8=Q Qe1+ 5.Kc2 h1=Q 6.Qb6+ Qe3 7.Qf6+ Qf3 8.Qb6+ Kg3 9.Qg6+ Qg4 10.Qd6+ Qf4 11.Qg6+ Kh3 12.Qh7+ Qh4 13.Bf5+ Kg3 14.Qg6+ Kf4 15.Qd6+ Kxf5 16.Qd5+ Kg4 17.Qe6+ Kg3 18.Qb3+ Kh2 19.Qb8+ Kh3 20.Qc8+ Qg4 21.Qc3+ Kh4 22.Qh8+ Kg3 23.Qc3+ Qf3 24.Qg7+ Kf4 25.Qf6+ Ke3 26.Qc3+ Kf2 27.Qc5+ Qe3 28.Qf5+ Kg3
If multipv = 1 stockfish-1.7 can not find the combination of mate!

[Tord Romstad]

I'm trying stockfish 1.7 and encountered a problem with the following position: 8/B7/1P6/1B6/8/8/3K1ppp/6kq w - - 0 1

Hello Benigno,

That bug is already discussed in the thread named "Stockfish 1.7 - nullmove blunder".