Greg,
I know it would be a little more work on your part, but it should be pretty easy to do. I suggest that the average rating of oppenents score be calculated with all of one's games every time a game is finished.
Or to put it another way, fractions should be carried over.
The way it is now, when you finish a new game, the previous average is calculated by multiplying the average by the previous number of games played (without fractions), and the new opponent rating is added to that total. The final total is then divided by the new number of games finished.
The end result being that if you play someone with a ranking that is not greater than your opponent's average rating by the number of games you have played, your average opponents rating does not increase. So if you have played 1,000 games, you would have to play someone 1,000 pts higher than your opponent average rating to get the average to move up one point.
The end result, particularly over time, is that the average opponent rating is nowhere near accurate. I think this would be an easy fix. Basically, just recalculate the average every time with all the ratings; do not use the already calculated average to determine the new average.
Or again, more simply, keep fractions when calculating the average.
What does everyone else think?
Changing how average rating of oppenents is calculated
Well, for me the opponent's average rating is not really meaningful.
It is calculated using ratings of all my opponents - including games played seven years ago, when ratings were not inflated.
I would like to propose other formula:
Fresh games should be more important.
So, if there are many games played, you can limit their "value".
NewAverage = (49*OldAverage+NewGame)/50
for any number of games above 50.
Some rating systems are based on this idea.
See also "Exponential moving average" in http://en.wikipedia.org/wiki/Exponential_moving_average
It is calculated using ratings of all my opponents - including games played seven years ago, when ratings were not inflated.
I would like to propose other formula:
Fresh games should be more important.
So, if there are many games played, you can limit their "value".
NewAverage = (49*OldAverage+NewGame)/50
for any number of games above 50.
Some rating systems are based on this idea.
See also "Exponential moving average" in http://en.wikipedia.org/wiki/Exponential_moving_average
-
- Site Admin
- Posts: 1388
- Joined: Sun Mar 14, 1999 11:13 am
- Location: Jeffersonville, IN
- Contact:
The primary reason for this field is to compute a player's provisional rating, but I always kept updating it mostly for fun. Now that you can go into a player's profile and see their actual rating history, including the ratings of the player's opponents, this field seems rather useless. Maybe it would be better to just not display it for established rated players, which would force people to go to the player's history page.
Greg Miller
True, recent games are more important than games finished hundreds of matches ago, but it is a quick and dirty rating that gives you a general impression of the quality of a player's opponents. It gives you an idea of the quality of the people you play as well.
If we are going to have it, it may as well be as accurate as possible, no? And I always look at a player's history page before accepting challenges, in addition to their average opponent's rating. Dropping one will not make me look at the other any more than I already do.
I actually think it is a pretty useful rating, but then again maybe i am weird.
If we are going to have it, it may as well be as accurate as possible, no? And I always look at a player's history page before accepting challenges, in addition to their average opponent's rating. Dropping one will not make me look at the other any more than I already do.
I actually think it is a pretty useful rating, but then again maybe i am weird.