We have a FAQ here. FAQ means Frequently Asked Questions. Odds are your Question is Frequently Asked.
For those of us who are interested in this sort of thing, the last few weeks have been tense, as ForGG keeps threatening to overtake Life on the Aligulac rating list, unleashing what we have dubbed the ForGGeddon. He came within three points of managing, too, but thanks to TLO and MarineKing, it was not to be. Polt now rears his ugly (not really) head. Will we see the Poltocalypse instead?
For now, anyway.
- Life 1883 (-30: WCS KR, GSTL)
- ForGG 1865 (-8: WCS EU)
- Polt 1859 (+30: ZOTAC, WCS AM, Gigabyte Quals)
- Leenock 1832 (-5: WCS KR)
- LucifroN 1815 (+39: LVP, gNations, EIZO)
- Flash 1793 (+19: WCS KR, Proleague)
- PartinG 1792 (-14: WCS KR, Proleague)
- Bomber 1787 (-12: GSTL)
- viOLet 1763 (+20: WCS AM, GSTL)
- Symbol 1745 (+18: WCS KR, GSTL)
- LucifroN 1815 (+39: LVP, gNations, EIZO)
- Sen 1684 (-27: TeSL, WCS AM)
- Happy 1654 (+60: EnerJ, Russia vs. Ukraine, Yegalisk MC, EIZO, WCS EU)
- VortiX 1653 (-36: LVP, WCS EU)
- Kas 1644 (+26: WCS EU, Roccat, Russia vs. Ukraine, Yegalisk MC, Gigabyte Quals, EIZO)
- Welmu 1630 (+75: EMS, WCS EU Quals, SC2.fi, M-House Quals)
- Snute 1594 (-17: ATC, Proleague)
- Scarlett 1584 (+28: ATC, WCS AM, gNations)
- Jim 1577 (+47: WCS AM)
- Stephano 1575 (-31: WCS EU, EIZO, Gigabyte Quals)
Much has been written about whether the heterogeneous SCII scene ruins the ratings. I have some things to show you relating this.
First of all, here is a chart showing how many points Koreans "steal" from foreigners per game they play. As you can see graph tends strongly to the positive side, which indicates (as expected) that Koreans as a whole are underrated. When this graph stabilizes somewhere around zero (over time, of course it will always fluctuate), we can say that the ratings have caught up.
At least it seems that Koreans steal fewer points now than they used to, indicating that the rating error isn't as large any more as it was before. If we accept that the red line is a good fit (which it isn't, but whatever), it predicts that we will achieve balance sometime in the second half of 2014.
So yeah, patience is a virtue y'all.
This is another chart showing some of the difficulty I'm trying to bridge by tweaking parameters. This is the familiar predictive power plot, showing different scenes (note that this is from my experiments, and doesn't show the true situation using the actual parameters in production on aligulac.com). The tendency is always that predictions in Korea are weaker than elsewhere. Why is this? Beats me! I've tried weighing KR vs. KR games heavier, but this paradoxically has the opposite effect, and makes the problem worse.
Many of you have probably noticed that international predictions tend to be much more reliable than Korean ones. Well, here you see why (well not really, but at least you have it confirmed). There are vastly more international games in the database, and so the parameters have mostly been optimized with those games in mind. This is something I'm trying to fix, but it's not easy.
Thirdly, just because a player is rated ahead of someone else, doesn't necessarily mean he is better (or that you should think he is). I've written a small new feature for this which can calculate p-values comparing two players directly: http://aligulac.com/compare/
If the p-values are larger than, say, 0.05, then you're not really statistically justified in claiming one to be better than the other.
As a rule of thumb, there should be a rating difference of 250 points (overall) or 450 (matchup ratings) before you can go blowing your horns about.
So for example, just because the ratings have Lucifron > Innovation, this is with p=0.35. This is not significant.
So yeah... let the discussion commence. And don't forget to PM me if you're interested in actually doing something.
Shout outs this week for Kaelaris and Nicolai Czempin for their show during the WCS EU Premier Ro16 Group A broadcast!