ESL 1v1 leaderboard rating configuration
Posted: Thu 07 Jan, 2016 6:37 am
At the moment players on the leaderboard are ranked according to the 67% credible rating (raw Glicko rating minus 1 × ratings deviation), which is a "reasonable" filter, as it gives us a number the player is probably (67% probably) worth. However, this still makes it possible for some players with a high RD (i.e. few games and low rating reliability) to rise high on the leaderboard. In fact, to the #1 position at the moment.
Should we tighten the RD screw and use the 95% credible rating (raw Glicko minus 2 × ratings deviation) instead? This would reward more games and penalise less games more steeply, although it might be a bit unforgiving at first (you would start with a -700 modifier).
Besides adjusting the RD modifier, other options for affecting the leaderboard ranking would include setting a lower RD limit (currently players must have RD < 250 for appearing on the leaderboard) and setting a # of games limit (although that's basically what the RD limit does).
To give some perspective, I'd say you can get RD < 250 with 1–3 games and RD < 200 with 3–5 games, depending on who you play against (low RD opponents drop it a lot, high RD opponents little).
At the moment, the top 5 looks like this:
1. Mannoroth (5 games), 2036 ± 177 = 1859
2. BbBoS (14 games), 1895 ± 157 = 1738
3. Ser Topi (11 games), 1860 ± 139 = 1721
4. Ace of Swords (3 games), 1914 ± 215 = 1699
5. Tex (33 games), 1780 ± 83 = 1697
Using the 95% credible rating, it would look like:
1. Mannoroth (5 games), 2036 ± 177 = 1682
2. Tex (33 games), 1780 ± 83 = 1614
3. Forestradio (37 games), 1740 ± 77 = 1586
4. Ser Topi (11 games), 1860 ± 139 = 1582
5. BbBoS (14 games), 1895 ± 157 = 1581
6. BestN00b (13 games), 1765 ± 118 = 1529
7. Ace of Swords (3 games), 1914 ± 215 = 1484
Should we tighten the RD screw and use the 95% credible rating (raw Glicko minus 2 × ratings deviation) instead? This would reward more games and penalise less games more steeply, although it might be a bit unforgiving at first (you would start with a -700 modifier).
Besides adjusting the RD modifier, other options for affecting the leaderboard ranking would include setting a lower RD limit (currently players must have RD < 250 for appearing on the leaderboard) and setting a # of games limit (although that's basically what the RD limit does).
To give some perspective, I'd say you can get RD < 250 with 1–3 games and RD < 200 with 3–5 games, depending on who you play against (low RD opponents drop it a lot, high RD opponents little).
At the moment, the top 5 looks like this:
1. Mannoroth (5 games), 2036 ± 177 = 1859
2. BbBoS (14 games), 1895 ± 157 = 1738
3. Ser Topi (11 games), 1860 ± 139 = 1721
4. Ace of Swords (3 games), 1914 ± 215 = 1699
5. Tex (33 games), 1780 ± 83 = 1697
Using the 95% credible rating, it would look like:
1. Mannoroth (5 games), 2036 ± 177 = 1682
2. Tex (33 games), 1780 ± 83 = 1614
3. Forestradio (37 games), 1740 ± 77 = 1586
4. Ser Topi (11 games), 1860 ± 139 = 1582
5. BbBoS (14 games), 1895 ± 157 = 1581
6. BestN00b (13 games), 1765 ± 118 = 1529
7. Ace of Swords (3 games), 1914 ± 215 = 1484

