Drofa 2.0.0

[Frank Quisinsky]

Hi,

at the moment I do nothing.
The calm before the storm!

I will do different things, new events end of the year.
Will start in November!



But let us not discuss about it in Drofa thread.

Drofa is interesting!!
Engine runs without any problems, I will look in the games for statistics end of the still running tourney.
If Shredder 13 on one core = 3125 Elo ... Drofa should have around 2900 Elo. I think at the moment place ~ 50 in the World.
I like the engine, looking in the evening in the still running tournament games.

Best
Frank

[AlexChess]

Hi,

at the moment I do nothing.
The calm before the storm!

I will do different things, new events end of the year.
Will start in November!



But let us not discuss about it in Drofa thread.

Drofa is interesting!!
Engine runs without any problems, I will look in the games for statistics end of the still running tourney.
If Shredder 13 on one core = 3125 Elo ... Drofa should have around 2900 Elo. I think at the moment place ~ 50 in the World.
I like the engine, looking in the evening in the still running tournament games.

Best
Frank

Yes, Sasha has a great potential, Drofa is resilient also against very strong engines!

Regards, Alex

[No4b]

Hello all!
Drofa 3.3.0 released
https://github.com/justNo4b/Drofa/releases/tag/3.3.0
As usual, it shows bout +100 elo improvement in the self-play.

STC 8'' + 0.08''

Code: Select all

ELO   | 109.76 +- 3.63 (95%)
CONF  | 8.0+0.08s Threads=1 Hash=8MB
GAMES | N: 20000 W: 8648 L: 2532 D: 8820

LTC 40'' + 0.4''

Code: Select all

ELO   | 115.92 +- 3.40 (95%)
CONF  | 40.0+0.40s Threads=1 Hash=64MB
GAMES | N: 20000 W: 8235 L: 1799 D: 9966

Overall there is about 40 elo-gaining patches since 3.2.0.
Complete list of the changes with elo-gains can be found in Drofa_changelog_6
Most important changes:
Evaluation (39+-5 elo in total)

• Improved definition of the doubled and isolated pawns

• Evaluate own pieces blocking pawns

• Evaluate distance of enemy knights to our passers

• Evaluate checks in king safety

• Count pawns with squares in front attacked by 2 enemy pawns as rammed

• Use PSQT for connected/supported pawns

• Evaluate candidate passing pawns

• Evaluate ability of the minors to jump in outposts

• Evaluate Rooks on Rammed line and Rook on line where our pawn is attacking another pawn

• Evaluate ability of the Queen to do `in-face` check

• Apply penalty if there is no pawns at all in 2 neighbour lines from our king

• Evaluate backward pawns

• Evaluate pawn push threats

Search

• Tweak history bonuses depending on position

• Use countermove history to adjust LMR

• Allow LMR to extend moves

• Less LMP in pvNodes

• Behind NMP & Cutnode logic

• Implement probcut

Other

• Implemented prefetch

• BugFix: Start pre-loop pruning from the first move and not the second

• Timemanager now can stop search early/later depending on %nodes spent on the best move

• BugFix: Do not copy Movelist in search

• BugFix: Use very conservative time allocation for non-increment TC to prevent reporting no move

Hopefully with this release Drofa would surpass 3100 CCRL mark.

I`d like to thank Kim (Koivisto developer) for various discussions on search and evaluation.
Also as you can notice, several patches here are based on existing knowledge found in Koivisto, Ethereal and Stockfish;
ideas implemented in this engines contributed a lot to 3.3.0 developement.

[AlexChess]

Hello all!
Drofa 3.3.0 released
https://github.com/justNo4b/Drofa/releases/tag/3.3.0
As usual, it shows bout +100 elo improvement in the self-play.

STC 8'' + 0.08''

Code: Select all

ELO   | 109.76 +- 3.63 (95%)
CONF  | 8.0+0.08s Threads=1 Hash=8MB
GAMES | N: 20000 W: 8648 L: 2532 D: 8820

LTC 40'' + 0.4''

Code: Select all

ELO   | 115.92 +- 3.40 (95%)
CONF  | 40.0+0.40s Threads=1 Hash=64MB
GAMES | N: 20000 W: 8235 L: 1799 D: 9966

Overall there is about 40 elo-gaining patches since 3.2.0.
Complete list of the changes with elo-gains can be found in Drofa_changelog_6
Most important changes:
Evaluation (39+-5 elo in total)

• Improved definition of the doubled and isolated pawns

• Evaluate own pieces blocking pawns

• Evaluate distance of enemy knights to our passers

• Evaluate checks in king safety

• Count pawns with squares in front attacked by 2 enemy pawns as rammed

• Use PSQT for connected/supported pawns

• Evaluate candidate passing pawns

• Evaluate ability of the minors to jump in outposts

• Evaluate Rooks on Rammed line and Rook on line where our pawn is attacking another pawn

• Evaluate ability of the Queen to do `in-face` check

• Apply penalty if there is no pawns at all in 2 neighbour lines from our king

• Evaluate backward pawns

• Evaluate pawn push threats

Search

• Tweak history bonuses depending on position

• Use countermove history to adjust LMR

• Allow LMR to extend moves

• Less LMP in pvNodes

• Behind NMP & Cutnode logic

• Implement probcut

Other

• Implemented prefetch

• BugFix: Start pre-loop pruning from the first move and not the second

• Timemanager now can stop search early/later depending on %nodes spent on the best move

• BugFix: Do not copy Movelist in search

• BugFix: Use very conservative time allocation for non-increment TC to prevent reporting no move

Hopefully with this release Drofa would surpass 3100 CCRL mark.

I`d like to thank Kim (Koivisto developer) for various discussions on search and evaluation.
Also as you can notice, several patches here are based on existing knowledge found in Koivisto, Ethereal and Stockfish;
ideas implemented in this engines contributed a lot to 3.3.0 developement.

Thank you dear, I start testing it right now