EIDRaS Ratings

[bo_sox48]

I made a slight tweak to some of the weightings to give smaller swings for provisional players (i.e. 7 games played or less). I found that it was going a little wild and giving some very new players some very high ratings.

How do you account for this once they play their 8th game? Do those smaller swings get recalculated? Where did 7 games come from as opposed to any other number of games? There is one player at the very high end of the GR list right now that has soloed 8 games, for example, but if you look at his game history none of those 8 games really show what caliber of player he is, so obviously it isn't a magic number.

Does the caliber of those first seven games matter? I can see some Playdippers coming over or something of that sort and playing a number of high quality games, perhaps for a World Cup or something similar. Should those games still not be given the same weight?

[Yonni]

@Bo

7 was just chosen because that's what Judge did. It's in the litany of variables for us to discuss if we adopt the system.

Also, I misspoke in that quote. I adjusted the weighting of games that include provisional players. The reason for this is exactly as you pointed it out. We can get a new strong player (e.g. Village Idiot) that starts with a rating of 1000. That obviously isn't his "true rating" so we discount games played against him for a bit.

The formula that Judge used was:
A = 1 + p/7
;where p is the number of players with provisional ratings (the denominator obviously changes for map types)

I tweaked it to be:
A = 2*p/7

The maximum is the same but I further discounted games against provisional players. But, again, this can all be tweaked as we want.

@Ghug

I suppose you're correct but I'm not sure how to formalize it. In fact, I'm not entirely certain how they decided on the expected score for WTA (other than it giving the curve they like). Open to any ideas you might have.

Perhaps the value of "c" can be adjusted. As a reminder, the formula of expected score, X, is given by:

X = M * [e ^ (R *c)] / sum[e ^ (Ri *c]