Season 1&2 ST rankings

Regional Matchmaking Northwest
81

No. The reason why I posted that thread was to hype up games for the UFGL. The main UFGL thread (where I plan on physically going to the major tournaments and hosting them) is pretty dead, so I want to hype up the games, and the only way to do that is through online (Look at how popular XvSF got recently, compared to when it came out). Online isn’t a good way to tell the best players (playing head-to-head, or side-to-side always will be), but it is a great way to get some attention.

Beasley: I understand, that ranking doesn’t look right to me either (I’m sure other people would agree, too.) I hope this becomes very successful… we’ll finally see who the actual best player is, rather than favorites. If you want an easier test, I can give you some SNK and UFGL results (the number of people in each tournament isn’t too high to number crunch, and there isn’t a long history.)

I have a few more questions:

-Will the ratings decrease when someone becomes inactive, or does the mu increase?

-Is there some sort of bias in the system? Does it favor something over another based on a certain result?

-Will this be able to calculate who the absolute best fighting game players are? Guys that are dominant at multiple games I’m sure would get a huge boost. How will this be done, so that it shows their greatness?

-Let’s say a new game comes out, someone dominates the game, but doesn’t play after a year, while the games depth increases (making the previous dominator’s skills somewhat outdated); will the rating for each player that placed high in tournaments in the present be significantly higher than the previous dominator? or will it be similar, or does the previous dominator’s rating level off, and to what factors?

-One example that I’m very interested in, will we be able to examine a past meets present match-up with this? Someone that played in 2010 vs someone that played in 2030, will we be able to determine which player will get the better? If you don’t understand what I’m saying, the ELO system has a rating for how much one person will beat down on another person, so if someones rating is 2800, and another is 2600, then that person would think 200 moves ahead (something like that, it’s a little hard to explain right now).

-Will the rating take into account areas where the competition is higher than other places?

I’ve got more questions, but my minds not in it right now. Good to see you posting again.

82

Yeah, that worldwide list is DESPERATELY in need of data, cus DxP is a muuuuch rawer player than me and he’s 4th from last… (I assume dxp=Darksydephil?)

83

Exactly, and I am a much better player than you simply because I enjoy Kidd Valley burgers and you do not.

J/K. This makes me want to practice ST for once, GET HYPED.

84

I bet you there is some fat dude who won the lottery in the mid-80’s so every day he goes into Kidd Valley and buys all 4 of his daily meals there (breakfast, lunch, merienda and dinner), and is on a first name basis with all the minimum-wage warriors who take his order there.

He’d instantly rule the Seattle ST rankings if we taught him how to walk forward and do a fireball. WATCH YOUR BACK FROM KIDD VALLEY FATGUY.

85

I’ve now coded a page so that you can see how a players rankings progresses over time. Here’s a sample:

please let me know what you think. I’ll try to get this up on public hosting after I get home from vacation.

86

So the level gravitates throughout the tournament, but how do you get each calculation? The level was at 0 (loss), then jumped to 5.06 after a win, then 4.38 after losing and being eliminated. What is the mu usually start at? it seems to increase after a win and decrease after a loss; and sigma seems to lower no matter what (or slowly after a win).

Interested in reading what you have to post. Take your time.

87

the early matches will show a big jump in fluctuations, because the initial sigma (uncertainty) is high. As a player has more consistent results, the jumps are smaller. It may be more instructive to look at Umbrellastyle’s more recent matches:

Note that in match 46, even though he loses to Zass, the loss is insignificant (0.01 of a level). This is because Zass has a higher level, so Umbrella is not penalized. His subsequent wins over geesemaster and rayblade, both players of high level, give him a large increase (0.5 of a level each), while his next win over Pablo, who is ranked slightly lower than him gives him a modest increase in level (0.16 of a level).

The full algorithm can be found here:
http://research.microsoft.com/mlp/apg/Details.aspx

In general, if the game outcome is not suprising, there is effectively no change in level. If the outcome is suprising, there is a change in level to account for this new result.

mu starts at 25 (the halfway point between min level 0 and max level 50), and sigma starts at 8.3