|
Sorry I'm late today. I had some students to torture.
Contents (thanks wo1fwood, didn't know BBcode could do this) Shameless plug first: interview on ESFI World.
What is this?
Aligulac.com is an ongoing statistical project and website in development started in December 2012. It offers a comprehensive database of games from the pro and semipro SC2 scene, as well as a unique rating system aimed at rating players and teams, and predicting games. Every two weeks, when a rating list is finalized, I write a post like this.
The FAQ might be able to answer your questions. If not, I'll be keeping an eye on this thread so you can ask away here.
Also, before we start, I want to quote Heartland from one of my previous threads who had this to share, better put into words than I ever could.
On February 03 2013 02:50 Heartland wrote: I think what's cool and great about this work is that it does what statistics are good for. They give you the ability to create data and then to look at it critically. But maybe people in this thread confuse statistics with the Truth with a capital p (sic). That's not the way you should read statistics, whether in the morning paper or on TL. Rather statistics can make us think about deeper connections that we haven't seen before, twist and turn around concepts and play with them through statistical models. They're not meant to say "Scarlett should be in Code S." Obviously there are flaws or issues with these stats, but it's common for statistics everywhere. What you can do with that is to add or change some modifier, let it meet other forms of reasoning or to extrapolate on what we take for granted.
So yeah, tl;dr, lies, damn lies and statistics are the case with all stats but it's not the point of stats. So, let's get on with the news.
New features since last time
The player match histories have changed a bit. They don't sort by event, only by date (so it loads much faster for people with tons of games). The list is also filterable by opponent race, nationality (KR/non-KR at least) and series length. For other filtering you can still use this feature.
URLs are now generally human-readable, i.e. http://aligulac.com/players/10-Stephano/ instead of http://aligulac.com/players/10/ (of course the latter will still work).
The ratings are now updated every six hours instead of every two weeks, so you can see them change «live». The periods are still two weeks long. All you're seeing is a preview of the rating list as it would be if it were calculated right now.
The «most specialised» stats on a rating list are now weighted heavier towards the top of the list, to provide more relevant and interesting results.
The rating system now uses the logistic distribution instead of the normal distribution, which should help accuracy in cases where the skill disparity is large.
There have also been some optimizations in the backend that you are unlikely to notice unless I get several hundred concurrent viewers.
Old features and things you should still know about
Everyone can submit results. Just go here: http://aligulac.com/add/ – some pro players have been scouring their histories and submitting the missing bits (mostly the wins cough, cough), which is great!
Help is still needed! The database is probably the biggest in the world of its kind and it's being maintained by a handful of people. There are still 13,000 matches that have not been catalogued and I dare not imagine how many games we have yet to add. Almost all the Playhem cups are still missing, for example.
People volunteer fairly regularly, but the workload is too damn high for the current staff. For all the rejected TLPD volunteers (I know you are out there, I was one myself) this would be the perfect project.
Anyway, let's get to the juicy bit.
Top 10 players
- Life 1997
- Leenock 1992
- DongRaeGu 1971
- PartinG 1966
- Bomber 1960
- TaeJa 1958
- Rain 1948
- HyuN 1942
- viOLet 1926
- ForGG 1911
In spite of a disappointing GSL Ro16, Life gains 6 points after winning some IPL6 tournament. He passes Leenock, who lost two points mostly on account of a loss to Parting in the FXO Invitational. DongRaeGu, like Life, loses in the GSL, but gains 12 points after manhandling Liquid in IPTL. He passes Taeja, Bomber and Parting, who lose 18, 14 and 12 points respectively. Rain has gone 9–2 these weeks and is awarded with 11 points but no ranks. Hyun and ForGG are the new faces in the top 10, ousting Roro and Hero. Hyun in particular gained 88 points after a nice 41–10 performance.
Top 10 foreigners
- VortiX 1885
- Scarlett 1867
- LucifroN 1866
- Stephano 1854
- Snute 1743
- Sen 1735
- Kas 1730
- TitaN 1709
- Fraer 1687
- Nerchio 1683
Mostly the usual suspects. The Spanish inquisition unexpectedly (of course) gained points after the aAa, IPL D.I.C.E. and VasaCast. Both lost exactly one series these weeks, to each other. Stephano also gains points in spite of his GSL fiasco. The most significant losers are Scarlett (IPL D.I.C.E.) and Snute (too many to mention).
Top 10 teams, all-kill edition
- MVP 91.26%
- StarTale 90.85%
- AZUBU 90.68%
- SK Telecom T1 90.07%
- Incredible Miracle 88.77%
- Team Liquid 88.55%
- FXOpen e-Sports Korea 85.78%
- Evil Geniuses 85.66%
- Prime 85.62%
- STX SouL 84.38%
Top 10 teams, proleague-style edition
- MVP 78.35%
- Incredible Miracle 75.40%
- StarTale 75.28%
- AZUBU 75.07%
- SK Telecom T1 72.89%
- Evil Geniuses 67.24%
- Team Liquid 67.24%
- FXOpen e-Sports Korea 64.95%
- STX SouL 62.34%
- Prime 61.05%
There is a nice correlation between allkill rank and proleage rank. Some teams do change though. It appears that IM and EG have deeper rosters than most. The rank is very close and do note that differences in the 2-3% range are probably insignificant.
Contact us
For comments, feedback, feature requests and especially volunteers, you can reach us by:
- This thread, duh.
- PM. Me.
- IRC: #aligulac on quakenet.
- E-mail to evfonn(at)gmail(dot)com.
- Pilgrimage to Zürich. I will find you.
Thanks to
New guys: Otolia and Susurrus.
Usual suspects: Conti, kiekaboe, Grovbolle, Inflicted, PhoenixVoid and scisyhp.
Others: Day[9] for using his post for this year in my thread. Hendralisk for great feedback.
I have probably forgotten many.
See you in two weeks!
|
Edit: Forget what I said.
|
This is so great.
The TLPD game records are so incomplete, it would be great if TL could work with you to integrate Aligulac into TL in some way.
|
|
Well, yeah, I kinda fucked up. You're right, I completely overlooked that, missed it as I said.
Keep it up, y'all. Maybe I'll try and submit something again some time and then don't fuck it up. D:
|
On February 22 2013 02:15 StarVe wrote: Well, yeah, I kinda fucked up. You're right, I completely overlooked that, missed it as I said.
No problem man Just checking
|
On February 22 2013 02:15 StarVe wrote: Well, yeah, I kinda fucked up. You're right, I completely overlooked that, missed it as I said.
Keep it up, y'all. Maybe I'll try and submit something again some time and then don't fuck it up. D: Wow, so apologetic. It's fine. Sheesh.
|
United Kingdom14103 Posts
HYUN! Yeahhh! Thank ya for the update
|
Fun fact: Life's elo is his birth year. Neat.
|
opterown
Australia54649 Posts
|
You are flattering me TheBB. I hope I'll can figure in there with significant contribution next time !
|
|
Good job, as usual, keep going !
|
If I may, could you describe what is going on with the world's 11th greatest Protoss Titan?
So, he is the top protoss foreigner. And he is currently 9-22 in his most recent matches, and his wins are against "low-level" opponents.
But he is still the best foreigner, and I find it strange. Could you provide a detailed description of how his points went up and down?
He is at 1633, then loses 3-8 to Hyun (roughly 1900) and gets projected at 1800? Is this intended behavior?
|
On February 22 2013 08:12 fezvez wrote: He is at 1633, then loses 3-8 to Hyun (roughly 1900) and gets projected at 1800? Is this intended behavior? Where does he get «projected at 1800»? (Not sure exactly what that means.)
There are details for each rating adjustment made here: http://aligulac.com/players/60-TitaN/period/79/
Browse back and forth using the arrows.
If you're thinking about the numbers here: http://aligulac.com/players/60-TitaN/results/ then they are matchup-specific ratings. He has 1633 vZ but the 1797 is vT. You can see Titan has a high PvT rating and a low PvZ rating. All his ratings have mostly been decreasing for the last three months.
On February 22 2013 08:12 fezvez wrote: If I may, could you describe what is going on with the world's 11th greatest Protoss Titan? That's your interpretation, not mine.
|
23 games winning streak for ForGG, not bad.
|
god I love these posts. You color my world theBB
|
I feel like this would be a low priority, but would it be possible to eventually implement by-event statistics? By that I mean... say you click on an event, like http://aligulac.com/results/events/449-Proleague/ , then at the top there is a collapsible stats panel which has things like TvP/PvZ/ZvT scores, number of games/players total, etc. Much like TLPD.
|
Just curious...how are you going to handle HOTS? WoL data won't be too relevant...will it be a clean slate? will there be a formula that gives notion to the WoL stats? Will it just be added to the WoL data?
|
|
This site is one of my favourites to peruse when I'm procrastinating.
Thanks to TheBB and everyone involved in keeping it up!
|
On February 22 2013 10:14 MCXD wrote:I feel like this would be a low priority, but would it be possible to eventually implement by-event statistics? By that I mean... say you click on an event, like http://aligulac.com/results/events/449-Proleague/ , then at the top there is a collapsible stats panel which has things like TvP/PvZ/ZvT scores, number of games/players total, etc. Much like TLPD. Your wish is my command!
Example: http://aligulac.com/results/events/36-GSL-2013-Season-1/
I couldn't get you the player count because the query was waaaay to slow. Need to do some database optimization first. TT
Also, I opened a ticket list: https://github.com/TheBB/aligulac/issues
If you want you can just dump your feature requests and bugs there, too.
|
On February 22 2013 10:36 Prplppleatr wrote: Just curious...how are you going to handle HOTS? WoL data won't be too relevant...will it be a clean slate? will there be a formula that gives notion to the WoL stats? Will it just be added to the WoL data?
Currently we are adding matches from both games, yet we have an indicator on each match whether it is WoL or HotS.
I believe currently the rankings contain both, and the plan is to just let it transition with the launch of HotS.
|
Ah I didn't realize the source code was on github, cool!
About the fake games, could you add a fraction instead of a full win and full lose? Just adding 0.1 Loss and 0.1 Win or something like to prevent infinity problems. Otherwise it seems like it would slow down rating movement more than it should, especially of players that go 1-0 many periods in a row. Or wouldn't it be better to just maximize the new rating all at once -- maximize log likelyhood given period game results *and* prior myr/mys value? Instead it looks like you first maximize this period performance, and then as a second step combine with prior myr/mys to come up with a new rating.
Another problem with adding fake games is doesn't that cause the confidence in the rating to go up 3X as fast in the case of players going 1-0 or 0-1 every period? This makes the problem of those players' rating getting stuck worse too.
Also, where can I find these values? from aligulac.settings import RATINGS_MIN_DEV, RATINGS_INIT_DEV
ETA: Also maybe you could just keep the fake games even when players do not go 0-N/N-0. Otherwise it seems to me that there are weird cases where going 3-1 against a stronger opponent could make your rating move more than going 3-0.
|
Hey KillerDucky, you seem to know what you're talking about.
«Fake» games: Yeah, it's possible that a different value than 1-1 could help. I know that it can slow down rating change, but I didn't consider that it can decrease the RD too. I'll think about it. Of course I could also maximize the whole thing in one go. I did it this way since I was following Glickman's algorithm and I never thought to change it. I'm not exactly sure it it would have any strange effects.
The settings file is kept from the repo because it contains some passwords. (I didn't bother filtering those out.) The constants you need are:
RATINGS_INIT_DEV = 0.12 RATINGS_MIN_DEV = 0.02 RATINGS_DEV_DECAY = 0.008
|
Is there a way with Aligulac to plot the rating of say the 50 best players over the last two years all at once ?
For a while it was possible with the site mengsk.com (unfortunately the site is now dead). You would have all the graph of all the players superimposed and selecting a particular player would highlight his curve. It was quite fascinating to see the progression of a player ELO along the years.
For an illustration, the graphs from mengsk.com have been used in this TL article (comparing broodwar and chess)
|
On February 28 2013 00:16 Arzar wrote:Is there a way with Aligulac to plot the rating of say the 50 best players over the last two years all at once ? For a while it was possible with the site mengsk.com (unfortunately the site is now dead). You would have all the graph of all the players superimposed and selecting a particular player would highlight his curve. It was quite fascinating to see the progression of a player ELO along the years. For an illustration, the graphs from mengsk.com have been used in this TL article (comparing broodwar and chess)
Not currently possible. But in theory yes that should be doable, since the system already supports graphs, just not implemented.
Edit: This is not me saying that this is something being considered, just that it should be possible.
|
If you want players to be highlightable I guess I have to use Javascript. Does anyone know of a graph widget like that?
|
On February 28 2013 03:16 TheBB wrote: If you want players to be highlightable I guess I have to use Javascript. Does anyone know of a graph widget like that?
Highcharts is a very powerful Javascript graphing library. It has highlighting/hover ability. I use it extensively at GGTracker and I'd be happy to answer questions about it if anyone gets stuck with a Highcharts problem.
|
|
Thank you, I fixed it now.
|
On February 28 2013 06:00 dsjoerg wrote:Show nested quote +On February 28 2013 03:16 TheBB wrote: If you want players to be highlightable I guess I have to use Javascript. Does anyone know of a graph widget like that? Highcharts is a very powerful Javascript graphing library. It has highlighting/hover ability. I use it extensively at GGTracker and I'd be happy to answer questions about it if anyone gets stuck with a Highcharts problem. Free for non-commercial use written by Norwegians? Can't say no to this!
|
|
Maybe aligulac thinks Zenio is so bad that Rain would need to beat him 2-0 in a Bo1 in order to maintain his rating, I don't know.
|
Go Switzerland! I would love to visit that country one more time, but my parents are wanted criminals there, so... :D
|
The ratings don't change independently. His other performances are dragging him down. The interpretation should be that yeah, he beat Zenio, but he has performed so poorly otherwise that it's reasonable to believe that his strength versus Zerg actually isn't any higher.
|
On March 05 2013 06:08 TheBB wrote:The ratings don't change independently. His other performances are dragging him down. The interpretation should be that yeah, he beat Zenio, but he has performed so poorly otherwise that it's reasonable to believe that his strength versus Zerg actually isn't any higher.
I see thanks. I hope I can get some time to play with the code myself, just reading it isn't cutting it.
|
|
|
|