Arasan 19.2

[jdart]

Arasan 19.2 is available from http://www.arasanchess.org.

Changes in Arasan 19.2:
1) In UCI mode, don't set tb type when a tb path is received if that path
is empty or the string "<empty>". Fixes issues under Shredder GUI.
2) Fixes to Syzygy tablebase handling. Don't correct TABLEBASE_WIN
scores for depth when storing/retrieving from the hash table. Adjust
search scores based on the tablebase score, when available.
3) Make UCI option handling insensitive to the case of the option name,
per the UCI spec. Add some error handling for invalid option names.
4) Remove threat extension and related code.
5) Change razoring margins.
6) Simplify king safety scoring and make most king safety parameters
tunable.
7) Tuner now uses the Texel method by default and supports the ADAM
gradient descent algorithm.
8) Changes to King cover calculation. Give bonus for King cover only
if King on rank <=2, and do not rate cover relative to King rank.

Self-play results, 1 core, Linux, time control 5+3:

Code: Select all

Rank Name          Elo    +    - games score oppo. draws
   1 Arasan 19.2  2423    8    8  1200   58%  2377   51%
   2 Arasan-19.1  2377    8    8  1200   42%  2423   51%

[styx]

Jon thank you for your wonderful engine. Arasan will always have a special place on my hard drive

[Dann Corbit]

For the current Arasan test suite, I have the following attribution for problem 19.084 (c0):
'WC25-SF20 ICCF Email 2001.08.01 Dahlgren, Jonas (SWE) vs Dolgitser, Konstantin'

The attribution is not found in the current test set (the only problem without a c0 listing).

Here is the PGN for that game:
[Event "W-ch25 sf20 email"]
[Site "ICCF email"]
[Date "2001.08.01"]
[Round "?"]
[White "Dahlgren, Jonas"]
[Black "Dolgitser, Konstantin"]
[Result "1-0"]
[BlackElo "2262"]
[ECO "B40r"]
[EventDate "2001.08.01"]
[Variation "Sicilian"]
[PlyCount "51"]
[Source "Opening Master"]
[SourceDate "2014.08.27"]
[BlackTeamCountry "WLS"]

1.e4 c5 2.Nf3 e6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 Qb6 6.Be3 Ng4 7.Qxg4 Qxb2 8.Kd2 Qxa1 9.Nxe6 Qxc3+ 10.Kxc3 dxe6 11.Qg3 a6 12.f4 Nc6 13.Kb2 g6 14.e5 Be7 15.h4 h5 16.Bd3 b5 17.Bxg6 fxg6 18.Qxg6+ Kd7 19.Bb6 Nb4 20.Rd1+ Nd5 21.Rxd5+ exd5 22.f5 Rb8 23.Qe6+ Ke8 24.Qc6+ Kf8 25.f6 Kf7 26.Ba7 1-0


I have 100% solution agreement. This is very unusual.

[jdart]

Thanks, will fix.

--Jon

[jdart]

In my experience popcnt makes a very small difference for Arasan. I'd expect it to be faster with popcnt enabled but it was never a big difference.

--Jon

[Dann Corbit]

I have seen with gcc 6 that the advanced CPU directives do not always help and sometimes even make it a bit slower.

[JVMerlino]

Unfortunately, the same problem as in version 19.1. The engine does not see the mat and evaluation is 7.24.

White: Kh1,Nh2,Bh3
Black: Kh8.

They begin to white.

[d]7k/8/8/8/8/7B/7N/7K w - - 0 1

If you are using tablebases, then there's definitely a problem.

If you aren't using TBs, then I don't see the issue. It's a mate in 29 and that's a pretty big search for even the best engines, depending on how long you want to wait. SF8 needs 66 seconds on my machine to announce mate, and it says M31.

[JVMerlino]

I used tablebase 6 men syzygy.

Well, that's a bug, then. Or an error in setup, I suppose.

[syzygy]

I used tablebase 6 men syzygy.

Well, that's a bug, then. Or an error in setup, I suppose.

Since the TBs do not give distance to mate, once it reaches a TB position Arasan probably switches to searching for mate without probing TBs. So Arasan probably just has difficulties finding the mate in 29 here.

The reason Arasan outputs 7.24 and not a "TB win" value seems to be that Arasan, when searching a root TB position like here, intentionally outputs the "raw" search score (returned by a probeless search).

SF takes much the same approach, but does not output the "raw" search score but a "TB win" value (unless the search returned a mate score). That is less confusing, as this thread shows

[Colin-G]

I am unable to run any of the Arasan 19.2 .exe files in Arena using linux with wine.
They immediately crash.
(Linux Mint 17 and Mint 18, wine 1.6, cpu AMD A4 6300, Arena 3.5)
I have no problems with Arasan 17.3 or earlier versions with the same conditions.
I tried to compile my own windows executables using i686-w64-mingw32-g++ instead of g++ but failed.
Although I have windows computers as well as linux, I do not have the Microsoft software to use the windows Makefile.
I have successfuly compiled 32 and 64 bit native linux versions though.