View more

View more

View more

### Image of the Day Submit

IOTD | Top Screenshots

### The latest, straight to your Inbox.

Subscribe to GameDev.net Direct to receive the latest updates and exclusive content.

# Glicko 2 rating deviaton during periods of inactivity

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

2 replies to this topic

### #1Boulougou  Members

Posted 18 April 2013 - 03:48 AM

Hello,

I am trying to use glicko 2 rating system in my game. I have implemented it using the formulas described in the document. However I am not sure I understand correctly how the rating deviation φ of a player is increased after periods of inactivity. The document mentions that when a player does not compete during a "rating period" the deviation φ is updated using this formula:

φ = sqrt(φ^2 + σ^2) (could not embed the formula, it seems that image extensions has been disabled)

where σ is rating volatility.

But using that formula it seems that φ is increased in a very slow rate. For example if a "rating period" is one month and a player has φ = 50 and σ = 0.06(the recommended value for volatility), after 10 years(120 "rating periods") of inactivity φ' = 50.004189. Is that the expected behavior or I am missing something? Is there any way to adjust how "quickly" φ is increased during player inactivity?

Let's hope that there is someone that has used this algorithm in the forums.

Posted 18 April 2013 - 04:08 AM

I had a quick look at the pdf. Are you sure phi = 50 is reasonable, the pdf divides the RD by 173.7178 (for some reason) to get phi.

If phi is standard deviations then 50 is enormous, 95% of random samples are usually within +/- 2 standard deviations of the mean. The tables in the data have reasonable values between 0 and 2 for phi.

"Most people think, great God will come from the sky, take away everything, and make everybody feel high" - Bob Marley

### #3Boulougou  Members

Posted 18 April 2013 - 05:49 AM

I had a quick look at the pdf. Are you sure phi = 50 is reasonable, the pdf divides the RD by 173.7178 (for some reason) to get phi.

Thanks for the reply, you are right. Silly mistake, I should have first scale the deviation before applying the formula and then rescale again as the document says. The correct value for the new deviation is 124.646889. That seems more reasonable.

However, this means that it takes 10 years(120 rating periods, assuming one rating period is one month) for deviation to increase from 50 to 124.646889. What if I would like to keep the same rating period(one month), but I would like deviation to increase more rapidly? I think Glicko1 had some constant that could adjust this. I cannot find a similar constant for Glicko2. It has to be some way, otherwise the algorithm would be quite constraining.

Edited by Boulougou, 18 April 2013 - 06:03 AM.

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.