Ratings accuracy.
-
- Site Admin
- Posts: 1388
- Joined: Sun Mar 14, 1999 11:13 am
- Location: Jeffersonville, IN
- Contact:
Ratings accuracy.
I was inspired by this article:
http://chessbase.com/newsdetail.asp?newsid=7114
I wanted to see how well the net-chess ratings stacked up to what Elo's system said the should be. I actually expected it to be way off, but was pretty surprised to see the curve appears to match even better than the FIDE ratings do.
The numbers across the top is the difference in rating points, each label goes to the line just to the left of itself. For example, the label of -1500 is on the very left edge, so any difference of larger than -1500 doesn't appear on the graph.
The numbers along the left edge is the expected score, and the labels go with the line just above them.
The blue dots are the average score of all of the games with that rating difference. The red line is the ideal Elo predicted score.
To orient yourself with the graph, note that the red line goes right through the point 0,.5. So with a rating difference of 0, you're expected to have a score of .5 (a draw).
http://chessbase.com/newsdetail.asp?newsid=7114
I wanted to see how well the net-chess ratings stacked up to what Elo's system said the should be. I actually expected it to be way off, but was pretty surprised to see the curve appears to match even better than the FIDE ratings do.
The numbers across the top is the difference in rating points, each label goes to the line just to the left of itself. For example, the label of -1500 is on the very left edge, so any difference of larger than -1500 doesn't appear on the graph.
The numbers along the left edge is the expected score, and the labels go with the line just above them.
The blue dots are the average score of all of the games with that rating difference. The red line is the ideal Elo predicted score.
To orient yourself with the graph, note that the red line goes right through the point 0,.5. So with a rating difference of 0, you're expected to have a score of .5 (a draw).
Greg Miller
-
- Site Admin
- Posts: 1388
- Joined: Sun Mar 14, 1999 11:13 am
- Location: Jeffersonville, IN
- Contact:
The raw data is here:
http://www.net-chess.com/ratings2.txt
It contains a column for the number of games played at each point difference too.
Compute the ideal Elo expected score by (C# code):
http://www.net-chess.com/ratings2.txt
It contains a column for the number of games played at each point difference too.
Compute the ideal Elo expected score by (C# code):
Code: Select all
class Elo
{
public static double getExpectedScore(int ratingDifference)
{
return 1 / (1 + Math.Pow(10, ((-ratingDifference) / 400f)));
}
}
Greg Miller
-
- Posts: 880
- Joined: Mon May 14, 2001 10:56 pm
-
- Posts: 380
- Joined: Sat Sep 06, 2008 1:35 pm
-
- Posts: 880
- Joined: Mon May 14, 2001 10:56 pm
-
- Posts: 380
- Joined: Sat Sep 06, 2008 1:35 pm