FFR Skill Rank System

[21992]

So this was something I was cogitating about for a long time and hoped to be implemented into FFR but I never wanted to post it until now because I feared it would get too much hate from the community.

So I propose... A Skill Rank be put into FFR. For those that are perplexed by this. I mean to say, a ranking system that accurately assess someone's skill playing FFR. For anyone that has played osu! they probably have a gist of what I'm saying, you may take a look about what I'm talking about here.

Q:But we already have Overall Ranks and Average Ranks. Why would we need yet another of these?

Well many would comply to say that the ranks we currently have can't give us an idea of the person's skill concisely. Overall Rank (or FFR Rank) measures your GT score compared to other users. Average Ranking can only truly be utilized if you have completed all of the songs in the game (with effort). And I know this might disappoint veteran players who actually have done this, but some people just don't have the time... or the patience.

Q: What about Tier Points?

Tier Points are generally reserved for adept players. And are constrained to only a few songs. As much as I do wish to like the tier point system it's lacking some structure that it needs. Some might consider this execrable but that's all left to opinion. Some might compare this to their relationship with halo 4, but that's another story.

Q: Who will come up with this True Ranking Skill component? How will it be incorporated into the game?

Well I'm going to say this once, this is merely just a suggestion. It feel that it would make this game much more competitive and if done correctly make tournament misplacings a thing of the past. As for the equations to how this will be computed, heh I'll be the first to say that there's bound to be people more competent in math that I am.

[TheSaxRunner05]

What might be a more prudent discussion is what should define each division cut off. We could make equations but the lines between divisions are really the sticking point here. What AAA/SDG/sub 20/sub 30 ect cutoffs define each division?

That's a tough question. Everyone has different skillsets. Just as a rough example, an index player could AAA or SDG VB but still be D2 level overall. No matter how you set guidelines, those guidelines will be advantageous to some players and not so for others.

I think an equation might make a good guideline, but not as a hard and fast rule.

[21992]

Originally Posted by TheSaxRunner05

What might be a more prudent discussion is what should define each division cut off. We could make equations but the lines between divisions are really the sticking point here. What AAA/SDG/sub 20/sub 30 ect cutoffs define each division?

That's a tough question. Everyone has different skillsets. Just as a rough example, an index player could AAA or SDG VB but still be D2 level overall. No matter how you set guidelines, those guidelines will be advantageous to some players and not so for others.

I think an equation might make a good guideline, but not as a hard and fast rule.

I see what you're saying but that will only solve one problem of placing people in a division. I think it would be much cooler if you can see where you rank between people. Especially if you can see who is just ahead of you etc. Multiplayer matches could be mitigated by this too! I made this thread because I'm tired of being judged D5 because I have oni. imo I have no place in D5 more of a higher end D4 player. I practically retired from FFR when my keyboard broke.

[Gradiant]

Pretty sure your Oni get, Rottel booflag, multiple FMO SDGs, and 2 clean on Frictional Nevada (wtf) make you D5

EDIT: For the majority of players, it's pretty obvious what division they are in...but it's the players between divisions that's weird. It's good for tournaments though, since they're easily moved around when needed if there's not enough/too many in a certain division.

[TheSaxRunner05]

It would be impossible to see who is just ahead of you because how objective song difficulties are. The closer two players are in skill, the more subjective the differentiation in skill.

[~Zeta~]

I sense spreadsheets in my future (hi kjw) aklsdfjas;lkfdj

It's not impossible to assess skill; you just have to finagle all the variables to fit a general census. Are those values still arbitrary? Yes. You can break it down by a accumulation of skill sets set to songs (and where songs have multiple elements, break those down even further) to be based on raw scoring and then have a general layout of skill based on what they have played. That means, it can be based on what they have played so far and becoming even more unique or accurate with more played songs (perhaps a deviation or correlation measure based on song total).

The idea is interesting to me to try and find a basis or gist of a equation or rule set.

EDIT:

Just a quick list of element breakdowns:
1. bpm
2. skills required (if it has like 2 bursts, it isn't a bursty song; if it has 16 measures or something of jumpstream, that's "strong" enough to be a skillset of that song;if it has 16 measures of jumpstream and 2 measures of jumpjacks, it's considered both)

problem with multiple elements to the song is we can't tell which skill they're better at but just finding the average can help variance/skillset.

3. score (extremely biased and arbitrary but can be related to rank/scores of others or something)
4. length (really not necessary but things like RWOB a;lksfdja;lkfjawdf)

[EzExZeRo7497]

The system for performance points in Osu is extremely flawed, but the idea is there yeah. It's possible to assess skill, but only vaguely. Main issue would be how everyone has different skillsets and such and that some people will find a file with a higher difficulty much easier than the lower ones. For example, a player would find files like AIM Anthem (71) and Club (70) much easier than something say like For FFR (67?) or Rottel-Da-Station (66) because of the skillset. Of course, that's only if you use difficulties as a factor to indicate skill, which I really don't recommend to begin with.

Unless you have some factors which would get rid of subjectivity (look at Zeta's element breakdown for example), it would get rid of the bias issue and would be more accurate. But the only issue is that judging how large of the factor is also subjective to begin with. Someone can find EHHS easier than Makiba, but most people find EHHS harder for example.

It's not really possible to make a completely "objective" skill indicator, but you can make it that's close to being objective I guess.

[noname219]

I did some calculations during the 8th Official Tournament with levelranks and it turned quite good honestly. I had some of my predictions in the finals (PaperclipGames, Fluffy) and three winners (beary, SallowKnight, samurai).

Something like this could work well :

By using the current song difficulties (1 to 99), separate them by sections (which would represent different skill levels) :
1 to 4................x1
5 to 9................x2
10 to 14.............x3
etc.
90 to 94.............x19
95 to 99.............x20

From each section, find the 5 best PA by using this formula : Perfect / (Good + Average + Miss + Boo) = song total.
* If AAA, notecount x 2
* Song must be passed to count
* If there's less than 5 songs passed, that's not important, the total will be smaller

Sum those 5 songs = section total
Use the multiplayer, and sum all section total = Skill grandtotal.

[~Zeta~]

That's a start but if someone's good a jacks but nothing else, that certainly wouldn't rank accurately compared to another good at jumpstream and bursts lets say. If the 5 best are the same (all theory, of course) then they would have equal ranks but of entirely different skillsets.

What I was considering was completely breaking down song data and applying attributes to files which would then define a players skillset based on their scores. That is applying all songs in the game. The less songs completed/scored on will result in inaccuracies which would inflate/deflate their actual ranking.

The toughest and most subjective part is patterns. You really can't measure patterns in ranks which is really putting a hole in this idea but I haven't started messing around with it yet.

I guess it's an all-encompassing tier point list with total "points" (temp label) ranked from highest to lowest.

[noname219]

Originally Posted by ~Zeta~

What I was considering was completely breaking down song data and applying attributes to files which would then define a players skillset based on their scores.

Sounds really like a good plan to me. I don't know exactly how it would turn out in practice. Most, if not all of the files have a ton of different patterns and because of how the scores are recorded, we can't know where a player missed something (or what pattern is he good).

That's a start but if someone's good a jacks but nothing else, that certainly wouldn't rank accurately compared to another good at jumpstream and bursts lets say. If the 5 best are the same (all theory, of course) then they would have equal ranks but of entirely different skillsets.

I was seeing a trend when I was doing some calculations with my method. Compare to other songs in the same difficulty, long songs had better PA. Also, jumpstream-songs.
If there was a way to encompass this in my equation, I dunno.

[~Zeta~]

I'll see what I can do in the morning. Too tired from all this heat (omg) to think right.

[noname219]

Originally Posted by ~Zeta~

Too tired from all this heat (omg) to think right.

Location: South Pole

^^

[EzExZeRo7497]

Originally Posted by noname219

I was seeing a trend when I was doing some calculations with my method. Compare to other songs in the same difficulty, long songs had better PA. Also, jumpstream-songs.
If there was a way to encompass this in my equation, I dunno.

People have better PA on longer songs compared to files in the same difficulty with shorter length because length is also a factor in difficulty. Most files that are very long have either one or two very hard sections (Red Wings Over Baron, White Walls Part 2) or just very easy throughout for its difficulty, but it's constantly at that difficulty (Megaman Rock Medley).

You can reduce the weightage of those long (>4:00) songs by an arbitrary percentage I guess. Not sure about jumpstream though, it's certainly easier to do compared to bursts (at least for most people) but I don't think you should reduce the weightage of it in the equation.

[HalfStep]

Originally Posted by EzExZeRo7497

Not sure about jumpstream though, it's certainly easier to do compared to bursts (at least for most people)

hi

[Doug31]

I've tried doing this, and it seems a significant problem to me that you can't tell if a song is passed or not from the data. To avoid this, my program only gives points for levels that have a * so I know it's an FC, but admittedly this is a flawed system. Is there a list of the total notes, bpm, length, difficulty, etc., of all the songs somewhere? It would be very helpful in doing this. My problem is that my system only gives people points for what they're best at, so it does things like, huh? Kotarouchan AAA'd Aim? Tons and tons of points, even if a user is bad at virtually all other patterns. Even if songs had the values for stream, voltage, air, freeze, and chaos then it'd be possible to come up with something better, but we don't really have stats representing what type of files are on which songs. I'm not manually going to go type them all in. It would be possible to make a formula to accurately calculate skill but only if we had more data on the songs than is currently available.

[~Zeta~]

doing it in spreadsheets right now.

[21992]

If only there was a way of determining what songs had which patterns.....

[Doug31]

There is. Just play every song and record what they had. Only problem is it'll take way too long.

[~Zeta~]

Originally Posted by Doug31

There is. Just play every song and record what they had. Only problem is it'll take way too long.

Not really (not about the length). It's extremely subjective and will never be accurate. I find trills, rolls, bursts, speed absolutely no problem. Jumpstream? Friggin hard. It will always be diverse and inaccurate. Inaccurate since even if you find a medium, there will always be high deviation because of huge varying opinions.

[Doug31]

Okay, well, I've created a new and highly improved version of my program although it still fails to take many things into account as mentioned previously. But I think this would probably get about 98% of people into their correct divisions if we could determine numbers to be the cutoff of each version.

The program:

import java.util.ArrayList;
import java.util.List;
import java.util.Scanner;

public class SkillTestV2 {
public static void main(String[] args) {
new SkillTestV2().start();
}
public void start() {
List<Score> theList = new ArrayList<Score>();
Scanner console = new Scanner(System.in);
while (console.hasNextLine()) {
String line = console.nextLine();
if (line.startsWith("Average Rank:")) {
break;
}
Scanner s = new Scanner(line);
if (!s.hasNext()) {
continue;
}
s.next();
if (!s.hasNextInt()) {
continue;
}
int difficulty = s.nextInt();
String current = "";
if (s.hasNext()) {
current = s.next();
}
String songname = "";
while (s.hasNext()) {
songname += current + " ";
current = s.next();
if ((1 == current.length() || current.length() > 6) && (current.endsWith("0*") || current.endsWith("5*") ||current.endsWith("0") || current.endsWith("5"))) {
break;
}
}
if (!current.endsWith("*")) {
continue;
}
int perfects = s.nextInt();
int goods = s.nextInt();
int averages = s.nextInt();
int misses = s.nextInt();
int boos = s.nextInt();
double val = goods + 1.8*averages + 20.0/275*boos;
theList.add(new Score(songname, difficulty, (int) val));
}
// iterate over the list and get the best scores
Score[] best = new Score[10];
int[] weights = {10, 9, 8, 7, 6, 5, 4, 3, 2, 1};
int[] goods = {2, 5, 10, 15, 20, 25, 30, 35, 40, 50, 75};
for (int i = 0; i < 10; ++i) {
best[i] = new Score("You lose", 1, 9999); //really bad score
for (Score s : theList) {
if (s.difficulty > best[i].difficulty && s.goods <= goods[i]) {
best[i] = s;
}
}
theList.remove(theList.indexOf(best[i]));
}
int total = 0;
for (int i = 0; i < 10; ++i) {
System.out.println(best[i]);
total += weights[i] * best[i].difficulty;
}
System.out.println("Total score= " + total);

}

public class Score {
public String songName;
public int difficulty;
public int goods;

public Score(String songName, int difficulty, int goods) {
this.songName = songName;
this.goods = goods;
this.difficulty = difficulty;
}

public String toString() {
return (songName + " (" + difficulty + ") " + goods + " goods.");
}
}
}

Output with my level ranks:
0 (piano version) (66) 1 goods.
K8107 (65) 4 goods.
For FFR (67) 8 goods.
Gynandromorph (66) 15 goods.
Starwolf (72) 20 goods.
Romance in the Club feat. Liquido (70) 18 goods.
Undiscovered Colors (70) 26 goods.
Midnight Dragon (69) 19 goods.
Summer Time Perfume (69) 37 goods.
Caprice (73) 50 goods.
Total score= 3723

Very simple to use and now uses a player's 10 best scores. Just click run, copy and past your level ranks (use select all) into it, and it produces the output.

Edit at 3:50am on 7/21: Improved to this:
0 (piano version) (66) 1 goods.
K8107 (65) 4 goods.
For FFR (67) 8 goods.
Starwolf (72) 13 goods.
Romance in the Club feat. Liquido (70) 18 goods.
Midnight Dragon (69) 19 goods.
Undiscovered Colors (70) 26 goods.
Exciting Hyper Highspeed Star (69) 33 goods.
Summer Time Perfume (69) 37 goods.
Caprice (73) 50 goods.
Total score= 3748