Redqueen v0.9 release

[bhlangonijr]

RedQueen/Release 0.9

- Clean up code;
- Fixed bugs;
- Implemented Non-capturing checks at quiescence search;
- Improved evaluation of passed pawns, rooks and knights:
- Added candidate passer knowledge;
- Bonus to rooks in open/half open files;
- Added knight outposts;
- PST's values refined;
- Improved LMR code/parameters;
- The PV is now retrieved from the transposition table (I was using before Bruce Moreland's method);
- Implemented singular extensions (+20 Elo worth perhaps);
- Storing static evaluation information in hash table;
- Improved move ordering;

New features:

- Static evaluation UCI command. Type "test eval<enter>" in UCI prompt and it will display the internal evaluation values for the current position:

Code: Select all

RedQueen. Copyright (C) 2008-2011 Ben-Hur Carlos Vieira Langoni Junior. 
position fen 1Q6/5kpp/2p4r/b4p2/P2R4/1N4P1/5PKP/4q3 b - - 0 1
test eval
Material[WHITE]:          2735
Material[BLACK]:          2740
Pieces(PST&Other)[WHITE]: -21
Pieces(PST&Other)[BLACK]: -39
Pawns[WHITE]:             11
Pawns[BLACK]:             8
Mobility&Space[WHITE]:    47
Mobility&Space[BLACK]:    35
Pieces threats[WHITE]:    7
Pieces threats[BLACK]:    0
King threats[WHITE]:      1
King threats[BLACK]:      23
Endgame score(WHITE):    89
Endgame score(BLACK):    29
Middlegame score(WHITE): 24
Middlegame score(BLACK): 31
Main eval:                +13

- Benchmark: type "redqueen bench<enter>" and it will search a set of 12 FEN positions at fixed depth.
On an I7 2.8Ghz, Linux 64 bits environment:

Code: Select all

RedQueen. Copyright (C) 2008-2011 Ben-Hur Carlos Vieira Langoni Junior. 
test bench
Executing benchmark...
Position[1]: 8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - 0 1
Position[2]: 1q2r1k1/5pb1/2bppnp1/p1p5/Nr2PPP1/1BBP4/1PP4Q/2KR3R w - - 0 1
Position[3]: 1k1r3q/1ppn3p/p4b2/4p3/8/P2N2P1/1PP1R1BP/2K1Q3 w - - 0 1
Position[4]: rn3rk1/pbppq1pp/1p2pb2/4N2Q/3PN3/3B4/PPP2PPP/R3K2R w KQ - 6 11
Position[5]: 1Q6/5kpp/2p4r/b4p2/P2R4/1N4P1/5PKP/4q3 b - - 0 1
Position[6]: r1r2bk1/pp1n1p1p/2pqb1p1/3p4/1P1P4/1QN1PN2/P3BPPP/2RR2K1 w - - 0 1
Position[7]: 6k1/8/8/8/8/1p6/6K1/8 w - - 0 1
Position[8]: 8/1B3k2/4Rbp1/3Pp1p1/5p2/5P1P/3r2PK/8 b - - 0 1
Position[9]: 1r6/r7/2bkp3/1p1p1p1p/p1pP1PpP/P1P1P1B1/1PK4P/1N1B4 b - - 0 150
Position[10]: 8/8/8/8/3k4/1r6/6K1/8 w - - 0 1
Position[11]: 2n5/2P5/1Pk5/2B5/8/p7/8/1K6 b - - 2 86
Position[12]: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Finished benchmark:  
Total time(seconds): 61
Total nodes:         147337027
Nodes/Seconds:       2415361

On a Core2Duo 2.4Ghz, Windows 32 bits environment:

Code: Select all

RedQueen. Copyright (C) 2008-2011 Ben-Hur Carlos Vieira Langoni Junior. 
test bench
Executing benchmark...
Position[1]: 8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - 0 1
Position[2]: 1q2r1k1/5pb1/2bppnp1/p1p5/Nr2PPP1/1BBP4/1PP4Q/2KR3R w - - 0 1
Position[3]: 1k1r3q/1ppn3p/p4b2/4p3/8/P2N2P1/1PP1R1BP/2K1Q3 w - - 0 1
Position[4]: rn3rk1/pbppq1pp/1p2pb2/4N2Q/3PN3/3B4/PPP2PPP/R3K2R w KQ - 6 11
Position[5]: 1Q6/5kpp/2p4r/b4p2/P2R4/1N4P1/5PKP/4q3 b - - 0 1
Position[6]: r1r2bk1/pp1n1p1p/2pqb1p1/3p4/1P1P4/1QN1PN2/P3BPPP/2RR2K1 w - - 0 1
Position[7]: 6k1/8/8/8/8/1p6/6K1/8 w - - 0 1
Position[8]: 8/1B3k2/4Rbp1/3Pp1p1/5p2/5P1P/3r2PK/8 b - - 0 1
Position[9]: 1r6/r7/2bkp3/1p1p1p1p/p1pP1PpP/P1P1P1B1/1PK4P/1N1B4 b - - 0 150
Position[10]: 8/8/8/8/3k4/1r6/6K1/8 w - - 0 1
Position[11]: 2n5/2P5/1Pk5/2B5/8/p7/8/1K6 b - - 2 86
Position[12]: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Finished benchmark:  
Total time(seconds): 124
Total nodes:         146813142
Nodes/Seconds:       1183976

Source & linux64/win32 binaries: https://sourceforge.net/projects/redqueenchess/
v0.9 svn tag: https://redqueenchess.svn.sourceforge.n ... elease-0.9

I think v0.9 is about 100 Elo stronger than v0.8.

Regards,

[Graham Banks]

RedQueen/Release 0.9..............I think v0.9 is about 100 Elo stronger than v0.8.

Thanks Ben-Hur. You're on the up and up!

[Milos]

Source & linux64/win32 binaries: https://sourceforge.net/projects/redqueenchess/
v0.9 svn tag: https://redqueenchess.svn.sourceforge.n ... elease-0.9

Man, what kind of internet connection do you have? It took you 8 minutes since you've made this post to upload 1.5MB. That's a staggering 25kb/s upload speed .

[bhlangonijr]

Source & linux64/win32 binaries: https://sourceforge.net/projects/redqueenchess/
v0.9 svn tag: https://redqueenchess.svn.sourceforge.n ... elease-0.9

Man, what kind of internet connection do you have? It took you 8 minutes since you've made this post to upload 1.5MB. That's a staggering 25kb/s upload speed .

25 mb/s.They provide a great and cheap internet service here in Netherlands. From where I come from I have to pay a lot for that.

[Milos]

RedQueen/Release 0.9
- PST's values refined;

Somehow, I cannot avoid conclusion that your PST values are absolutely equal to Ippo opening PST values + 2...
I guess your "automatic tuner" has a systematic error of 2cp maybe you should check it out .

[bhlangonijr]

RedQueen/Release 0.9
- PST's values refined;

Somehow, I cannot avoid conclusion that your PST values are absolutely equal to Ippo opening PST values + 2...
I guess your "automatic tuner" has a systematic error of 2cp maybe you should check it out .

No error. Those values are what work best for RQ.

[Milos]

RedQueen/Release 0.9
- Improved LMR code/parameters;

Non-PV reduction is almost identical to SF's but for PV you decided to use 1/3 coefficient instead of 1/4.5. Piece of advice, 1/3 is too high.

[kinderchocolate]

Am I the only one that has problems in unzipping the download from Sourceforge?

[Milos]

RedQueen/Release 0.9
- PST's values refined;

Somehow, I cannot avoid conclusion that your PST values are absolutely equal to Ippo opening PST values + 2...
I guess your "automatic tuner" has a systematic error of 2cp maybe you should check it out .

No error. Those values are what work best for RQ.

Btw. I'm surprised that Ippo PST + SF LMR are worth less than 100 elo for you (assuming eval changes are also worth something), I would expect it to be much more...

[bhlangonijr]

RedQueen/Release 0.9
- PST's values refined;

Somehow, I cannot avoid conclusion that your PST values are absolutely equal to Ippo opening PST values + 2...
I guess your "automatic tuner" has a systematic error of 2cp maybe you should check it out .

No error. Those values are what work best for RQ.

Btw. I'm surprised that Ippo PST + SF LMR are worth less than 100 elo for you (assuming eval changes are also worth something), I would expect it to be much more...

Both altogether are worth about +-30 Elo. The majority of the Elo gain comes from the other changes.