When I beat someone at my rating I used to get 11 rating points (with a normal glicko RD). Now it is only 9 or 8. What has happened? Did anyhting change? Waiting to hear from the developers @thibault @Toadofsky.
u have rd of 50 in blitz because you played quite a few games recently.
The RD floor was reduced from 60 to 50, to stabilize ratings of frequent players.
Thanks. Why though was it reduced? Now it takes much more time to recover from rating losses (like chess.com). Any hopes of returning back to 60?
The RD floor was reduced from 60 to 50, to stabilize ratings of frequent players.
You are right but not only 9 or 8...two players with the same rating sometimes +9 sometimes +8 ..it is not known how is calculated..There is a problem to be solved.
#4 Regarding "Now it takes much more time to recover from rating losses"...
Lichess rating system has many similarities to Glicko: http://www.glicko.net/glicko/glicko.pdf . Before github.com/ornicar/lila/pull/4034 merged (June 30) I couldn't make that claim. In some cases I may have claimed that (just because we're using a Glicko-2 library) Lichess ratings are similar to Glicko-2; the more I look, the more I think Lichess ratings are similar to Glicko-1.
So yes, now if you lose 100 points (which is a more difficult endeavor than earlier, because ratings are more stable) then gaining 100 points is more difficult (because ratings are more stable).
What's next?
1. Glickman's paper doesn't mention a floor for RD. Gradually reduce the RD floor to 0.
2. Glickman's paper mentions the concept of a "rating period". I don't see a performant way to fit that concept into Lichess code. Maybe if players' RD values are too low, tune system parameters (such as the RD decay rate), holding the rating floor constant.
3. If there's a performant way to do so, consider implementing what OGS implemented forums.online-go.com/t/ogs-has-a-new-glicko-2-based-rating-system/13058
Lichess rating system has many similarities to Glicko: http://www.glicko.net/glicko/glicko.pdf . Before github.com/ornicar/lila/pull/4034 merged (June 30) I couldn't make that claim. In some cases I may have claimed that (just because we're using a Glicko-2 library) Lichess ratings are similar to Glicko-2; the more I look, the more I think Lichess ratings are similar to Glicko-1.
So yes, now if you lose 100 points (which is a more difficult endeavor than earlier, because ratings are more stable) then gaining 100 points is more difficult (because ratings are more stable).
What's next?
1. Glickman's paper doesn't mention a floor for RD. Gradually reduce the RD floor to 0.
2. Glickman's paper mentions the concept of a "rating period". I don't see a performant way to fit that concept into Lichess code. Maybe if players' RD values are too low, tune system parameters (such as the RD decay rate), holding the rating floor constant.
3. If there's a performant way to do so, consider implementing what OGS implemented forums.online-go.com/t/ogs-has-a-new-glicko-2-based-rating-system/13058
"1. Glickman's paper doesn't mention a floor for RD. Gradually reduce the RD floor to 0."
FICS has no RD floor. In its early days with the Glicko system in the mid 90s, very active blitz and bullet players got their RDs so low that their ratings were only changing by 1 point each game. While this probably resulted in more accurate ratings, it was also unpopular, although I personally liked it.
After a couple of years, a minimum K factor of 16 was instituted.
FICS has no RD floor. In its early days with the Glicko system in the mid 90s, very active blitz and bullet players got their RDs so low that their ratings were only changing by 1 point each game. While this probably resulted in more accurate ratings, it was also unpopular, although I personally liked it.
After a couple of years, a minimum K factor of 16 was instituted.
@Toadofsky Do you know when the RD floor got reduced from 60 to 50?
This topic has been archived and can no longer be replied to.