Andscacs - New version 0.83

[cdani]

Thank You!! Keep Up the good work my friend...testing shows your very close to the top....some loses hard to explain...depth seems to not be the issue.....loses games in late middlegame....fades late....cause unknown
in my opinion.....Thx AR

Thanks! If you have an example of this strange loses, will be very helpful

[tttony]

Excelent!!

Do you have a table showing the elo gain by each added/improved feature?

[carldaman]

Some detailed changes that maybe someone will like to implement in his engine:
* Optimized a little the prefetches.
* Added idea of re-search at intermediate depth if reduction is very high from Stockfish.
* In move ordering, a little penalization for move king outside rank 0 when queens.
* Do not consider pawn shelter of king when in center.
* Grouped half of the hash entry in a single uint64_t.
* Added the idea "If both sides have only pawns, score for potential unstoppable pawns" of Stockfish.
* Increased bonus for bishop pair that cannot be exchanged for other bishop or knights of rival because it does not have them.
* Diminish bonus for bishop pair if the rival knights are good or the center is very blocked.
* Diminish attack value if only one attacker.
* If rival has more pawns in front of king, increase safety penalization. Idea from Hakkapelitta.
* Increased penalization for underdevelopment on some piece configurations.

Nice progress, Daniel. I've always liked the way Andscacs plays. It seems to like the initiative, going for robust but flexible structures, and is not shy when it comes to flank attacks on the kingside.


Regarding the latter aspect, will the "fewer pawns in front of the King" penalty not deter Andscacs too much from undertaking kingside pawn storms, especially when its own King has also castled that way?

There are many situations, when the center is locked/closed, where such pawn storms are correct and one's king shelter is of secondary importance.

Regards,
CL

[cdani]

* If rival has more pawns in front of king, increase safety penalization. Idea from Hakkapelitta.

Regarding the latter aspect, will the "fewer pawns in front of the King" penalty not deter Andscacs too much from undertaking kingside pawn storms, especially when its own King has also castled that way?

There are many situations, when the center is locked/closed, where such pawn storms are correct and one's king shelter is of secondary importance.

If the king has less pawns in front, will have this penalization, but in the other side will have open lines, that if can be occupied by rooks can favor a possible attack. The penalization is interesting I will say basically to discourage the king to castle in this side.

[cdani]

* If rival has more pawns in front of king, increase safety penalization. Idea from Hakkapelitta.

Regarding the latter aspect, will the "fewer pawns in front of the King" penalty not deter Andscacs too much from undertaking kingside pawn storms, especially when its own King has also castled that way?

There are many situations, when the center is locked/closed, where such pawn storms are correct and one's king shelter is of secondary importance.

If the king has less pawns in front, will have this penalization, but in the other side will have open lines, that if can be occupied by rooks can favor a possible attack. The penalization is interesting I will say basically to discourage the king to castle in this side.

Also is interesting to say that Andscacs never had a pawn storm bonus, so I suppose it does not attack a lot with pawns.

[Jorge Garcia]

Congratulations Daniel!
Good work, keep on although it's hard to evolve when being on the top...
Best regards

[Sylwy]

http://talkchess.com/forum/viewtopic.ph ... 76&t=50060

[cdani]

Excelent!!

Do you have a table showing the elo gain by each added/improved feature?

Is really difficult to tell, because many of them were evaluated in selfplay. I'm sorry.

[MikeB]

Thanks all!
To give a little idea of the evolution:

Code: Select all

100' + 0.03 0.82
Rank Name                Elo    +    - games score oppo. draws 
 1 Komodo 5.1r2 64-bit   107   40   39   198   69%   -33   24%
 2 Elektro 1.2            43   37   37   200   62%   -33   34%
 3 BlackMamba 2.0 x64     21   38   38   198   58%   -33   26%
 4 Protector 1.8.0         8   37   36   198   56%   -33   36%
 5 Texel 1.05 64-bit      -1   37   37   198   55%   -33   31%
 6 Nirvanachess 2.1c      -7   36   36   200   54%   -33   37%
 7 Naum 4.6              -16   38   37   200   53%   -33   27%
 8 Hannibal 1.5 x64      -26   36   36   198   51%   -33   41%
 9 Andscacs 0.82         -33   12   12  1986   45%     3   32%
10 Senpai 1.0            -45   37   37   198   48%   -33   32%
11 Gull 1.2 x64          -51   37   37   198   47%   -33   30%

100' + 0.03 0.83
Rank Name                Elo    +    - games score oppo. draws 
 1 Komodo 5.1r2 64-bit   144   28   28   380   72%    -4   32%
 2 Elektro 1.2            61   28   27   384   59%    -4   27%
 3 BlackMamba 2.0 x64     52   27   27   384   58%    -4   32%
 4 Protector 1.8.0        20   26   26   378   54%    -4   42%
 5 Texel 1.05 64-bit       0   27   27   384   51%    -4   32%
 6 Andscacs 0.82206       -4    9    9  3819   49%     0   33%
 7 Hannibal 1.5 x64      -35   27   27   380   45%    -4   33%
 8 Nirvanachess 2.1c     -40   26   26   384   44%    -4   35%
 9 Naum 4.6              -45   26   27   384   44%    -4   34%
10 Senpai 1.0            -61   27   27   378   42%    -4   30%
11 Gull 1.2 x64          -92   27   28   383   37%    -4   30%

Excellent Daniel! Do we have Mac OS version or can I help get you a Mac OS version? To be rated higher than Hannibal, Naum, Senpai and Gull is saying something ! ( sorry - no nothing about Niravanachess - but it must be strong too!)

[cdani]

Excellent Daniel! Do we have Mac OS version or can I help get you a Mac OS version? To be rated higher than Hannibal, Naum, Senpai and Gull is saying something ! ( sorry - no nothing about Niravanachess - but it must be strong too!)

Those Gull and Nirvana versions are old ones

The problem is I only know Windows stuff like Visual Studio, and also my C++ is not necessarily compatible with other operating systems because I never took time to know what is and what is not portable because I never learnt C seriously, only by example, so I suppose I have things that are incompatible. Anyway It will be nice to be capable to port the engine, sure! Maybe in some time I will try. Thanks for your offer