Names by Color and Elo Values by Color
Posted: 29 Jul 2008, 16:33I noticed on the FIDE ratings site that they now compute result percentages by color.
http://ratings.fide.com/
It recalled something I was working on about a year ago. I went back and finished it up. Here is what it does:
My new utility splits each players name into a "black" and a "white" name. For example, "Engine A" is split into "Engine A (bl)" and "Engine A (wh)".
Combined with other "40H" utility programs, I can compute elo ratings for the original player, and for each of the color names. It seems that it is typical for black to white percentages to be around 44% to 56%. This translates into a 43 point elo differential in favor of having the white pieces.
http://chess.about.com/od/chessratings/ss/aa07b17_3.htm
By comparing the black/white elo differential for an engine, the developer can see if the differential is close to the expected 43 elo points. Here is some sample data obtained with the help of additional "40H" utility programs:
In the following data, none of the engines played against themselves, and all games were contained in the same pgn database.
The new utility is called "nameByColor" and it will be included in the next release of "40H", which will be ready by 2008.08.01 at
http://www.hoflink.com/~npollock/chess.html
http://ratings.fide.com/
It recalled something I was working on about a year ago. I went back and finished it up. Here is what it does:
My new utility splits each players name into a "black" and a "white" name. For example, "Engine A" is split into "Engine A (bl)" and "Engine A (wh)".
Combined with other "40H" utility programs, I can compute elo ratings for the original player, and for each of the color names. It seems that it is typical for black to white percentages to be around 44% to 56%. This translates into a 43 point elo differential in favor of having the white pieces.
http://chess.about.com/od/chessratings/ss/aa07b17_3.htm
By comparing the black/white elo differential for an engine, the developer can see if the differential is close to the expected 43 elo points. Here is some sample data obtained with the help of additional "40H" utility programs:
In the following data, none of the engines played against themselves, and all games were contained in the same pgn database.
- Code: Select all
Original
Name Games Elo W=wins : D=draws : L=losses
Bright 0.2c 4CPU 1030 2667 165 W : 390 D : 475 L
Bright 0.3a 4CPU 653 2711 124 W : 280 D : 249 L
Bright 0.3b 2CPU 150 2677 32 W : 50 D : 68 L
Bright 0.3d 4CPU 457 2732 87 W : 200 D : 170 L
Deep Fritz 10 2CPU 799 2713 222 W : 329 D : 248 L
Deep Fritz 10 4CPU 537 2708 119 W : 214 D : 204 L
Deep Fritz 10.1 4CPU 1260 2753 335 W : 528 D : 397 L
------------------------------------------------------------------------
Names By Color
Name Games Elo W=wins : D=draws : L=losses
Bright 0.2c 4CPU (bl) 513 2637 56 W : 185 D : 272 L
Bright 0.2c 4CPU (wh) 517 2690 109 W : 205 D : 203 L
Bright 0.3a 4CPU (bl) 325 2686 48 W : 127 D : 150 L
Bright 0.3a 4CPU (wh) 328 2730 76 W : 153 D : 99 L
Bright 0.3b 2CPU (bl) 76 2695 15 W : 25 D : 36 L
Bright 0.3b 2CPU (wh) 74 2663 17 W : 25 D : 32 L
Bright 0.3d 4CPU (bl) 231 2695 31 W : 90 D : 110 L
Bright 0.3d 4CPU (wh) 226 2764 56 W : 110 D : 60 L
Deep Fritz 10 2CPU (bl) 400 2711 92 W : 175 D : 133 L
Deep Fritz 10 2CPU (wh) 399 2719 130 W : 154 D : 115 L
Deep Fritz 10 4CPU (bl) 261 2689 41 W : 111 D : 109 L
Deep Fritz 10 4CPU (wh) 276 2721 78 W : 103 D : 95 L
Deep Fritz 10.1 4CPU (bl) 630 2735 128 W : 275 D : 227 L
Deep Fritz 10.1 4CPU (wh) 630 2767 207 W : 253 D : 170 L
The new utility is called "nameByColor" and it will be included in the next release of "40H", which will be ready by 2008.08.01 at
http://www.hoflink.com/~npollock/chess.html