Winboard Forum

Hi Andrew,

welcome in the Chess960 world!

How do you estimate FireFly's actual playing strength?

Here is a first game with Smirf:

Code: Select all

[Event "Chess960 Welcome 20 min + 5 sec"]
[Site "CHESSBOX-MAIN-X"]
[Date "2005.06.11"]
[Time "06:12:27"]
[Round "1"]
[White "Smirf-Beta BC-062"]
[Black "FireFly v1.3.1"]
[Result "1-0"]
[Annotator "R. Scharnagl"]
[SetUp "1"]
[FEN "bbnnqrkr/pppppppp/8/8/8/8/PPPPPPPP/BBNNQRKR w KQkq - 0 1"]

1.Ne3 {(12.02) +0.121} Nb6 2.c4 {(12.07) +0.133} e5 3.b4 {(12.00) +0.351} Nc6
4.f4 {(12.00) +0.504} d6 5.fxe5 {(12.01) +0.479} Nxe5 6.Qg3 {(12.02+) +0.479}
f6 7.Nb3 {(09.32) +0.519} Nbxc4 8.Nxc4 {(12.01) +0.737} Nxc4 9.Rxf6 {(12.25)
-0.722} Rxf6 10.Bxf6 {(13.39) +1.074} Qf7 11.O-O {(13.47) +0.928} O-O 12.Rf4
{(11.53) +0.861} g6 13.Qh3 {(12.01) +2.201} Nb6 14.Bxg6 {(14.29) +2.214} Qxg6
15.Qe6+ {(14.50) +2.193} Rf7 16.Rg4 {(15.13) +2.267} Qxg4 17.Qxg4+ {(16.41)
+6.371} Kf8 18.Bg5 {(16.02) +6.900} c5 19.Qe6 {(15.16) +9.353} Kg8 20.Bh6
{(15.14) +9.397} Nd7 21.Qe8+ {(15.37) +10.08} Nf8 22.Qxb8 {(16.36) +10.57} Rf6
23.Be3 {(16.31) +10.83} b6 24.Qxa8 {(17.01+) +11.49} Rf7 25.Bh6 {(16.01)
+11.76} cxb4 26.Qd5 {(14.01=) +M~006} Ng6 27.Nd4 {(02.01=) +M~007} b3 28.axb3
{(02.01=) +M~005} b5 29.Nxb5 {(02.02=) +M~004} Nf8 30.Nxd6 {(02.01=) +M~003}
Ne6 31.Qxe6 {(02.01=) +M~003} a5 32.Nxf7 {(02.01=) +M~002} a4 33.Qe8# {(02.00?)
+M~001} 1-0

Regards, Reinhard.

Hi Andrew,

I guess that FRC strength is a bit weaker than normal game strength since no book is available.

that sounds strange to me, because your program's opponent neither would have an opening book. ;-)


Reinhard.

Hi Andrew,

I noticed through the beginning of following game, that FireFly's evaluation has been up to nearly two pawns better than Smirf's estimation. Knowing that Smirf's evaluation function still is very basic, I wonder, what would cause those obvious differences.

Code: Select all

[Event "Chess960 Welcome 20 min + 5 sec"]
[Site "CHESSBOX-MAIN-X"]
[Date "2005.06.11"]
[Time "21:12:08"]
[Round "2"]
[White "FireFly v1.3.1"]
[Black "Smirf-Beta BC-062"]
[Result "0-1"]
[Annotator "R. Scharnagl"]
[SetUp "1"]
[FEN "bnnrkrqb/pppppppp/8/8/8/8/PPPPPPPP/BNNRKRQB w KQkq - 0 1"]

1.g4 f5 {(11.01) -0.020} 2.gxf5 Rxf5 {(10.27) +0.130} 3.Nc3 Qe6 {(10.35)
+0.047} 4.Qg3 Nd6 {(10.01) +0.117} 5.Bf3 Nc6 {(10.23) +0.480} 6.Bg4 Nb4
{(11.48) -0.043} 7.O-O Nxc2 {(11.23) -0.062} 8.Nb3 g6 {(11.00) -0.028} 9.Bxf5
Qxf5 {(11.02+) +0.148} 10.d3 b5 {(11.01) +0.417} 11.e4 Qh5 {(12.01+) +0.520}
12.Rc1 Nxa1 {(12.01+) +0.621} 13.Rxa1 c5 {(11.41) +0.882} 14.Rfe1 b4 {(12.18)
+0.940} 15.Nd1 c4 {(11.35) +0.882} 16.Nd2 c3 {(12.44) +2.206} 17.bxc3 bxc3
{(12.43) +2.364} 18.Nf3 c2 {(12.39) +2.413} 19.Rc1 Nf5 {(12.23) +2.838} 20.Qf4
cxd1=B {(13.01+) +2.890} 21.Rcxd1 Bc3 {(13.28) +3.043} 22.Rf1 Nd4 {(13.21)
+3.027} 23.Kg2 d5 {(13.41) +3.470} 24.Ng5 h6 {(13.15) +3.812} 25.Qf7+ Kd7
{(02.00?) +1.766} 26.Nh7 Ne2 {(13.12+) +4.289} 27.Qf3 Rh8 {(13.44) +5.362}
28.Qxh5 gxh5 {(15.14) +5.456} 29.Kf3 Nd4+ {(15.29) +5.676} 30.Ke3 Rxh7 {(14.00)
+5.707} 31.Rc1 Nb5 {(14.31) +5.891} 32.Rb1 a6 {(13.33) +6.063} 33.a4 Bd4+
{(14.16) +6.077} 34.Kd2 Nc3 {(14.33) +6.297} 35.Rb4 e5 {(13.32) +6.523} 36.Rb8
Bc6 {(14.01+) +6.732} 37.exd5 Nxd5 {(13.27) +6.988} 38.a5 Bc3+ {(13.02+)
+7.144} 39.Kc2 Rf7 {(14.01+) +7.663} 40.Kb3 Bxa5 {(14.04+) +7.918} 41.Rh8 Bd2
{(13.01+) +8.044} 42.f3 Bb5 {(13.00) +8.229} 43.Rb8 Ba5 {(11.21+) +8.831}
44.Rh8 Bxd3 {(13.05) +10.09} 45.Rf2 Rf4 {(13.10) +10.12} 46.Rxh6 Rb4+ {(14.01=)
+M~007} 47.Ka3 Ne3 {(02.01=) +M~006} 48.Ka2 Nc4 {(02.01=) +M~006} 49.Rc2 Bxc2
{(02.01=) +M~004} 50.Rh7+ Bxh7 {(02.01=) +M~003} 51.h4 Bc2 {(02.01=) +M~002}
52.f4 Ra4# {(02.00?) +M~001} 0-1

Regards, Reinhard.

Hi Andrew,

Smirf's evaluation does not distinguish game phases like opening, middle- or endgame. It does neither use piece-coordinate driven bonus or malus values from tables, nor support different values depending on game phases for passed pawns. Smirf moreover still has not yet any mobility component in its evaluation.

You see, there is room for improvements in Smirf.

Reinhard.

No mobility evaluation in smirf?

I think that mobility is one of the obvious things to add.

You can use the number of possible moves of the pieces of both sides to have some evaluation of mobility that is better than nothing.

Uri

Hi Uri,

Smirf indeed has still a very basic evaluation function. But I cannot work on a lot of places simultaneously.

There are subjects like my leading move pair iterative deepening, which I write completely on my own and have to fix method details by testing. That also holds for concurrent / parallel pruning methods and multi level evaluations of nodes.

To stabilize Smirf's search actually has priority. Increasing performance already could be noticed, though the convergent definitions of methods in my view are the bigger progress.

Reinhard.