Zahak, a GoLang based chess engine

[amanjpro]

Since I edited my post several times (it is not easy to write in a foreign language), I deleted "slower than engines of similar strength". My impression was that it was because the net was bigger (your exe is bigger and achieves half the speed in number of nodes of Koivisto and Berserk, not to mention the depths). But to avoid further misunderstanding, just have in mind that I am not a programmer, and that I judge engines by using them for analysis and comparing their lines. While I am aware about the relativity of nodes, depths, etc., I have no technical knowledge to be precise about it.

I understand... I expect it to be slower than both engines. I am a bit less aggressive in pruning nodes when searching. But that is also a reason why they are both rated at least +200 over Zahak

Zahak 7 was faster than Zahak 8, and smaller (binary file wise), because the net was smaller and it didn't have EGBT support... but the speed was compensated by the evaluation quality

Usually to compare engine vs engine speed, one should look at the NPS not depth, because depth doesn't really mean much, and it is not comparable between two engines. some engines start pruning much later than others. (Zahak waits 6 plies to start pruning). The more aggressive you prune, the deeper you search, but the more chances you miss easy tactics. There is always the right balance that we all search for, but we all fail to find it

[Rebel]

Changes since 7.1
New network architecture
EGBT support (Syzygy)
Improved search
+300 elo hopefully

You are incredible

Games : 1000

Elo pool : 3180

http://rebel13.nl/b/grl.htm

[matejst]

In my view, too much pruning is good only for eng-eng competition and Elo hunting. Comparing different version of Komodo, I always thought that K12, although much stronger than K9 or K8, brought nothing useful for a chess player, the important "skill" option excepted. Similarly, Wasp 1.02 was an enjoyable engine reminiscent of Zarkov, with a very recognisable style (for me, quite "human"), but this style of play was lost in the subsequent versions -- although the last one, W4.50, seems to reconnect with its roots. In general, I do not understand why chess enthusiasts are focused on only two or three engines -- it makes no difference for OTB preparation, while many of the "middle tier" engines give sensible and often original lines one can use in a game.

So far, and in the perspective I tried to make a bit clearer in the previous paragraph, I am positively impressed by your engine. It seems both positionally and tactically sound (I am analysing a position of the Sicilian I already analysed with other engines, and I also can check the online CCD). But, like Dietrich when he made a remark about his Harmon net yesterday, I would rather use NN engines with nets trained to play, let's say, an aggressive and tactical brand of chess, or, quite the opposite, very positionally, even if they are several hundreds Elos weaker. I would also like to see engines with levels, multiPV, position learning, I think it is much more important than raw Elos.

[matejst]

Ed,

From my 20 games testing I think that your Elo pool is a bit too strong, and that you should try with 3120, something about the strength of Winter 0.9.

[Rebel]

Ed,

From my 20 games testing I think that your Elo pool is a bit too strong, and that you should try with 3120, something about the strength of Winter 0.9.

Only 99 games played so far, elo indication +202. Depending on the outcome I can always play another 1000 against a lower pool.

[amanjpro]

In my view, too much pruning is good only for eng-eng competition and Elo hunting. Comparing different version of Komodo, I always thought that K12, although much stronger than K9 or K8, brought nothing useful for a chess player, the important "skill" option excepted. Similarly, Wasp 1.02 was an enjoyable engine reminiscent of Zarkov, with a very recognisable style (for me, quite "human"), but this style of play was lost in the subsequent versions -- although the last one, W4.50, seems to reconnect with its roots. In general, I do not understand why chess enthusiasts are focused on only two or three engines -- it makes no difference for OTB preparation, while many of the "middle tier" engines give sensible and often original lines one can use in a game.

So far, and in the perspective I tried to make a bit clearer in the previous paragraph, I am positively impressed by your engine. It seems both positionally and tactically sound (I am analysing a position of the Sicilian I already analysed with other engines, and I also can check the online CCD). But, like Dietrich when he made a remark about his Harmon net yesterday, I would rather use NN engines with nets trained to play, let's say, an aggressive and tactical brand of chess, or, quite the opposite, very positionally, even if they are several hundreds Elos weaker. I would also like to see engines with levels, multiPV, position learning, I think it is much more important than raw Elos.

Scally is actually currently working on human-playable networks, it didn't make the cut for version 8.0 because I had changed the network format almost last minute... but hopefully the new nets will be shipped with the binary as skill_level_options.

I agree with you, a strong engine doesn't necessary mean a useful one, and I am interested in making Zahak useful

[amanjpro]

Ed,

From my 20 games testing I think that your Elo pool is a bit too strong, and that you should try with 3120, something about the strength of Winter 0.9.

Only 99 games played so far, elo indication +202. Depending on the outcome I can always play another 1000 against a lower pool.

Thanks a lot Ed

[Rebel]

Estimated elo gain for Zahak 8.0 : + 207

Code: Select all

Gambit Rating List
Running      : Gauntlet Zahak 8.0
Time Control : Time control : 40/120
Games        : 1000

Results from file gauntlet-zahak.pgn:

No. Name         Win Draw Loss Unf.  Score Games       %
--------------------------------------------------------
  1 Zahak 8.0   +252 =362 -386   *0  433.0  1000   43.3%
  2 Stockfish 5  +98  =56  -46   *0  126.0   200   63.0%
  3 Clover 2.4   +78  =82  -40   *0  119.0   200   59.5%
  4 Minic 3.14   +69  =79  -52   *0  108.5   200   54.2%
  5 Wasp 4.50    +77  =63  -60   *0  108.5   200   54.2%
  6 Halogen 10   +64  =82  -54   *0  105.0   200   52.5%

Total Games:    1000
White Wins:      285 (28.5%)
Black Wins:      353 (35.3%)
Draws:           362 (36.2%)
Unfinished:        0 (0.0%)

Estimated elo gain for Zahak_8.0
Elo pool : 3180
Zahak 7.0 : 2933.0
Zahak_8.0 : 3140.3
Difference : 207.3

[Rebel]

Another 1000.

Zahak 8.0 scored an estimated rating of 3140.

New Pool : 3142

http://rebel13.nl/b/grl.htm

[CMCanavessi]

Another 1000.

Zahak 8.0 scored an estimated rating of 3140.

New Pool : 3142

http://rebel13.nl/b/grl.htm

So far, +207 again