Fun little coding experiment with ratings
 

Decided to take a little break from my studies and wrote up a Python script to rank players by "Top 100 Avg" ratings instead of skill ratings currently used to see how the ranks change. This is primarily done for fun so feel free to post here and see how you rank (searching top 1000 players took 2.5 hours for it to complete, so only D5+ players).

I thought this is interesting because it mainly accounts for all players who actively play sufficiently enough games on FFR to determine one's "comfort" skill level on a good day. For example, my current Top 100 Avg is 87.84. This says that on a good FFR day, I am comfortably able to AAA difficulty 87 - 88. In my opinion from my own gameplay experiences, this is much more accurate than what my recently achieved Level 92 states (it basically says that "on a very lucky day, I can AAA a difficulty 92").

Based on this, here are some fun facts:

Assuming current division thresholds (i.e. D8 = 101+, D7 = 94+, D6 = 83+, D5 = 69+, etc.):

No one qualifies for D8 (opposed to 25 players currently ranked D8)
28 people would be D7+ (opposed to 130 players currently ranked D7+)
156 people would be D6+ (opposed to 529 players currently ranked D6+)
572 people would be D5+ (opposed to 1671 players currently ranked D5+)

Top 10 Highest Top 100 Average Ranked Players:
Matthia: 100.64
TC_Halogen: 98.26
xXOpkillerXx: 98.0
Zoobin4: 97.88
Shxfted: 97.61
klimtkiller: 97.59
Mipha: 97.53
SoFast: 97.48
One Winged Angel: 97.05
XtraFestive: 96.9

22 people can comfortably get OWA Raged at Home Run Derby.
142 people can comfortably get St. Scarhand [Heavy].
538 people can comfortably get Otaku Speedvibe [Oni].

Interesting results overall :) Enjoy and feel free to ask and see how you rank against others via Average Top 100 Ratings!

EDIT: I think it should be possible to predict where other players who don't play enough games (like Myuka and AnMAO) fall in this ranking approach. Would be interesting to get one to play seriously and see how those predictions would shape out to be.

Re: Fun little coding experiment with ratings
 

This is interesting stuff, would be cool to see where I place lol, maybe if possible get a spreadsheet up where we could just search ourselves and see where we are :D

Re: Fun little coding experiment with ratings
 

How long before I become the first ever qualifying D8 player for Top 100 Av Rank ;o

Re: Fun little coding experiment with ratings
 

mathematics

Re: Fun little coding experiment with ratings
 

My average is 69 something and has been for some time, yet it took me a million years to get Oni. Nerves are real.

I suspect no one qualifies for D8 simply because there aren't anywhere close to a hundred 101+ difficulty songs.

Re: Fun little coding experiment with ratings
 

the most interesting thing to me is that the number of qualified players in each division is in the ballpark of the next lowest division average

Re: Fun little coding experiment with ratings
 

The highest possible top 100 right now is 102.22

So you can do it, at least mathematically.

Re: Fun little coding experiment with ratings
 

"EDIT: I think it should be possible to predict where other players who don't play enough games (like Myuka and AnMAO) fall in this ranking approach.”

problem is there aren’t enough songs in-game to decide this.

Re: Fun little coding experiment with ratings
 

Everybody (at a given level) has the same sample of files to play. If we enforce a minimum of 50 or 100 files to be played, we greatly minimize bias. Granted, the very top players will have to compete on a bunch of files that are significantly lower than their level, but there is no way around that if we want to keep the rest of the leaderboards accurate and fair.

Re: Fun little coding experiment with ratings
 

Hey all, this was initially for experimental purposes to see how relying on only unweighted Top 100 pans out. I can probably write a quick follow-up script to show these results too if there's interest.

Here's my proposed weighted setting that seems to be performing well so far: https://colab.research.google.com/dr...o=duy-96GtEK6b

Please let me know if there are any questionable placements here. Thanks! :)