Ingame Song Information

[HoneyMelonCalibrator]

Originally Posted by Xx{Midnight}xX

Can we PLEASE put Southern Cross at it's rightful 11? I hate having to play it every time I am randoming 10s.

Edit: how the hell is lazer shooter an 8? It's a bloody 9. 14 23 14 23 patterns appear in almost NO 8's if any of them.

Those jump trills are the only thing somewhat difficult in the song, though. The rest is a easy. I say it stays an 8.

Originally Posted by Mulie

I like XCV's idea for identifying and analyzing a few variables. Also agree with qqwref's observation that variables have to be carefully weighed and relationally defined before balancing an equation. I don't think a rigorous equation is really needed though... Just simplifying the amount of subjectivity in each song is enough to make the task of ranking them considerably easier.

I played around a little and simplified subjectivity into two factors. My numbers could be completely off, but how about something like this?

file placement = relative difficulty * (intensity + length + compensation factor)

relative difficulty: Use an arbitrary scale, like the already established 12 (+1) scale, to approximate the song's difficulty in relation to others. (A slightly larger scale like 15 might work better.) If everything's balanced enough, we could later multiply relative difficulty by the sum of the other factors to determine a solid rank (the equation above).

intensity/density: Some things would be too much of a pain to rigorously derive. Instead, arbitrarily rate density on a small scale (e.g. scale of 5) to approximate density. Since songs have already been categorized by relative difficulty, this should be considerably easier.
I think length should be considered separate because density alone is closer to determining how much you'll have to concentrate per unit of time.

length: Determine on a scale of 4 layers (short, medium, long, very long). Actually define these. Something like 2:19- short, 4:00+ long and 5:30+ very long.

compensation factor: Maybe add points for other difficulty factors? E.g. bad synchronization +1, blue arrow syndrome +1, flaring background +1...

Yes, that's a good idea, I definitely agree.

[XCV]

Originally Posted by Mulie

I like XCV's idea for identifying and analyzing a few variables. Also agree with qqwref's observation that variables have to be carefully weighed and relationally defined before balancing an equation. I don't think a rigorous equation is really needed though... Just simplifying the amount of subjectivity in each song is enough to make the task of ranking them considerably easier.

I played around a little and simplified subjectivity into two factors. My numbers could be completely off, but how about something like this?

file placement = relative difficulty * (intensity + length + compensation factor)

relative difficulty: Use an arbitrary scale, like the already established 12 (+1) scale, to approximate the song's difficulty in relation to others. (A slightly larger scale like 15 might work better.) If everything's balanced enough, we could later multiply relative difficulty by the sum of the other factors to determine a solid rank (the equation above).

intensity/density: Some things would be too much of a pain to rigorously derive. Instead, arbitrarily rate density on a small scale (e.g. scale of 5) to approximate density. Since songs have already been categorized by relative difficulty, this should be considerably easier.
I think length should be considered separate because density alone is closer to determining how much you'll have to concentrate per unit of time.

length: Determine on a scale of 4 layers (short, medium, long, very long). Actually define these. Something like 2:19- short, 4:00+ long and 5:30+ very long.

compensation factor: Maybe add points for other difficulty factors? E.g. bad synchronization +1, blue arrow syndrome +1, flaring background +1...

Suppose we moved the scale up to 20, a la SM, and round everything up.

If we did this, we'd have to cap the compensation factor at x, where x equals the largest amount that the theoretically hardest song under this system could have and not go above 20.

From there it's just simple equation setup to determine what to divide by.

0 < 13 * (x + 5 + 4)/y < 21

Solving gives

y <= x/8 + 9/8,
161 > x > -9.

Okay, nothing's ever going to get a compensation factor of anything much bigger than 20 under any conceivable system. This is obviously ridiculous, so I propose clearing up the ambiguity of the compensation factor.

Flaring shapes thing = +1
Other distracting BG's (Zelda remix, Funk in G, SkellyBones) = arbitrarily assigned a value from 1 to 3 depending on severity
Offsync steps - Take % of steps that aren't on the beat.

< 1% = 0
1% - 2% = +1
>2% = +2

BNS - +1

The maximum value under this system is 6, therefore, we can redo the evaluation:

0 < 13 * 15/y < 21

Solving:

0 < y < 9.29

If the maximum cap under 1-20 is 9.29, then the cap from 1-13 must be 13 * 9.29/20 = 6.04

That isn't too bad. This is entirely theoretical, though, and of course the days of BNS are over. So, I'll evaluate Death Piano:

Length: 2
Note density: 5
Compensation: 2 for offsyncness (there's no way everything in there is on beat)

So, 13 * 9/6.04 = 19.37. Therefore, DP would score a 20 under this new system.

On the other end of the spectrum, let's evaluate the easiest song in the game, that being Eien Meikyu IMHO:

Length: 1
Note density: 1
Compensation: 0

So, 1 * 2/6.04 = .33, round it up to 1.

Of course, this would allow 21s to exist, but that's not likely.

So, improved equation. This isn't perfect, but it's a better approximation and a heck of a lot easier to work with:

ceil(D * (I + L + C)/6.04)

And we can adjust this as needed.

[jimerax]

ok can you guys show some examples with that system?

(Songs)
100 bar black out
Rondo Alla Turca
TGWP Part 1
Adventures of lolo
Crowdpleaser

[XCV]

*obediently substitutes*

Comparing this to ceil(20/13 * D)

100 Bar - 3 * (2 + 2 + 2)/6.04 = 3/4 as opposed to 5
RAT - 9 * (2 + 3 + 2)/6.04 = 11 as opposed to 14
TGWP - 10 * (2 + 3 + 0)/6.04 = 9 as opposed to 16
Lolo - 12 * (1 + 5 + 2)/6.04 = 16 as opposed to 19
Crowdpleaser - 12 * (2 + 4 + 1)/6.04 = 12 as opposed to 19

So, for most songs it scores quite a bit low. Perhaps 6.04 is a bit too large a number to divide by; if we reduce it to 5, we get scores that are more accurate to the proportionate values. Remember that this is a scale of 1 to 20, and 20 is something like vrofl in terms of difficulty. Well, not that extreme, but you get the gist of what I'm saying.

I might bring up my point from a page or so ago that this should be used when it's extremely hard to tell or controversial (of course with some refining,) that comparison should be the most important word. The system is pretty sound as it is, maybe if we can perfect this it can be used to work out the kinks.

[qqwref]

Originally Posted by XCV

100 Bar - 3 * (2 + 2 + 2)/6.04 = 3/4 as opposed to 5
RAT - 9 * (2 + 3 + 2)/6.04 = 11 as opposed to 14
TGWP - 10 * (2 + 3 + 0)/6.04 = 9 as opposed to 16
Lolo - 12 * (1 + 5 + 2)/6.04 = 16 as opposed to 19
Crowdpleaser - 12 * (2 + 4 + 1)/6.04 = 12 as opposed to 19

The 100BB rating doesn't place enough value on the Synthlight Factor™, plus that one fast jack in it. I'd say 5 is reasonable.
RAT is a very tricky 9 to AAA (offbeat + weird patterns + jacks) so I dunno about 11.
TGWP songs are all tough to PA because of the huge BPM changes, so while they might not be very dense, they're still hard songs to do well on.
Lolo is fine.
95% of CP is not even VC level, but it has a few bursts which are really hard to FC (and very hard to AAA too), so it should have a correspondingly high difficulty ranking just for that.

I guess the compensation factor should have some extra points for jacks (Fei Longer's ending isn't any more dense than a broken stream pattern, but way harder), and offsync or complicated sync should probably add more points (depending on how bad it is of course).


Incidentally I think the 1-100 scale needs to open up WAY more at the higher levels. Maybe we should just abandon the 100 cap. Having fine distinctions is great on the lower songs, but it's restrictive to only have 10 points to describe the difference between Integraation and Death Piano.

[Mulie]

I also think placing a cap is unnecessary and cumbersome to the refining process. At the very least it should be left for after the rest is balanced.

Compensation factor needs more conditionals. I'm not as well informed as a lot of you in this meta, and threw those first ones out there as examples. What I can say, though, is that all those factors need to be identified before assigning them values. At first, a base value of 1 would suffice.

Qqwref's right about songs like Crowdpleaser and TGWP. Ultimately everything falls back to the effort required to AAA songs. Rather than approximating intensity by average density, charts should be judged by their hardest portions. Fatigue should already be sufficiently accounted for with song length.

Since song length is a measure of fatigue we may also want to add another layer for songs longer than ~7 minutes.


I'd like to hear more opinions on this. I feel naked and vulnerable =x

[Patashu]

Ultimately, any algorithmic or objective attempts to determine difficulty should not be done from the armchair, but by trial and refinement.

Create an initial method, use it to rate a bunch of songs, and look for discrepancies - in particular, look for the biggest discrepancies, and identify what about them made the rating system go wrong. Edit rating system so it better handles these, rate more songs, repeat ad infinitum.

If you suggest methodologies in a void, then you are essentially groping blindly through the set of all possible methodologies, and giving no results, thus eventually everyone involved will get bored and forget it all.

[qqwref]

Interesting experiment idea:
- start up a new rhythm game site, like a custom FFR or something, with user accounts and some kind of leaderboard/statistics
- at some specific date, add a whole bunch of files with slight enough variations to determine ratings - some should be mostly js that is faster/longer/shorter/slower/denser, some should be more bursty than others, some should have better or more complicated sync than others, etc
- get lots of people to play, wait a while
- look at the stats on various songs and use that to compare their difficulties to actual players

[xXAll-ProXx]

Originally Posted by qqwref

Interesting experiment idea:
- start up a new rhythm game site, like a custom FFR or something, with user accounts and some kind of leaderboard/statistics
- at some specific date, add a whole bunch of files with slight enough variations to determine ratings - some should be mostly js that is faster/longer/shorter/slower/denser, some should be more bursty than others, some should have better or more complicated sync than others, etc
- get lots of people to play, wait a while
- look at the stats on various songs and use that to compare their difficulties to actual players

good idea, but very time-consuming.

[One Winged Angel]

What if we were to link the difficulty of the file to how many AAAs/FCs that file currently has in relation to every other song in the game?

For example, say the number of FCs on a file are worth one point each, and the number of AAAs on the same file are worth three points... every song in the game can be ordered in difficulty with respect to how many 'points' each song has. So, atm (I'm assuming) Trip to the Moon would be 1, and vrofl would be 907...and then instead of arguing if a song should be FMO or VC or whatever when a song has too many AAAs, the song will just move down in difficulty if it begins to accumulate more points than other songs that had a higher number of points previously. Obviously the difficulties would change when new files are introduced into the game but it wouldn't take longer than probably a week or so for those songs to find an area around their rightful difficulties...and if a difficulty scale of close to 1,000 is too high, we could group songs to one difficulty level in groups of 5 or 10, and just the lowest or highest song in each group would move to the next highest/lowest difficulty level when the difficulties change due to increased point totals.

This is probably the closest thing I can think of with respect to a more 'formulaic' approach to difficulty ranking that would address certain files like CP that only have one or two sections of 'wtf' while the rest of file remains pretty much trivial.

[HoneyMelonCalibrator]

I think that the difficulty scale should be expanded. Meaning rather than measuring a song's difficulty on a scale of 1 to 13, maybe songs could be measured from 1 - x. X being whatever new number the difficulties expand to.

This way it would be more distinct which file contains what difficulty. I know that right now people refer to songs as "low 10", "mid VC" or "high 11" etc. Why not create a new system of measurement that distinguishes the differences? You would still be able to refer to songs as VD, VC, FMO and FGO but rather than those titles referring to one level of difficulty, they could refer to a range of difficulties? Like for example you could be like "FMOS are songs ranged from difficulty x to z, but at the same time, a player would be able to distinguish x, y, and z (all FMOS) in terms of increasing difficulty. I don't think this would be too hard, either.. considering songs already have difficulty numbers associated with them. They're just not publicly displayed on the main screen.

Thoughts?

[foxfire667]

Originally Posted by One Winged Angel

What if we were to link the difficulty of the file to how many AAAs/FCs that file currently has in relation to every other song in the game?

For example, say the number of FCs on a file are worth one point each, and the number of AAAs on the same file are worth three points... every song in the game can be ordered in difficulty with respect to how many 'points' each song has. So, atm (I'm assuming) Trip to the Moon would be 1, and vrofl would be 907...and then instead of arguing if a song should be FMO or VC or whatever when a song has too many AAAs, the song will just move down in difficulty if it begins to accumulate more points than other songs that had a higher number of points previously. Obviously the difficulties would change when new files are introduced into the game but it wouldn't take longer than probably a week or so for those songs to find an area around their rightful difficulties...and if a difficulty scale of close to 1,000 is too high, we could group songs to one difficulty level in groups of 5 or 10, and just the lowest or highest song in each group would move to the next highest/lowest difficulty level when the difficulties change due to increased point totals.

This could be a good idea, although I believe that it is taking a "perfect world" approach in terms of every person having played all of the songs, and the variation on that song being able to be analyzed that way. If everyone who played FFR actively played every single song in the game (which is rather hard considering you'd need to play all 767 public, plus unlock all 102 skill tokens, all 35 regular tokens, and all 3 support tokens) then of coarse because of their varying skill levels, we would probably be able to determine how hard a song is. Considering, however, that this is not the case, it would be tough to base it on FC's and AAA's alone. Let's take token whore, as an example. They're are only 96 actual scores on this song, 92 of which are AAA's...and although this song is easy, it would be ranked much higher in difficulty because so few people actually have the song.

Although, we could average out the total number of different players in relation to the number of FC's and AAA's (which could be assigned different values). This might cancel out the fact that only x amount of players have played a certain song. But then again, everyone who has unlocked Token Whore has the skill to at least AAA an FMO, so this would also be rather one sided. How this would fair in finding a universal difficulty for all of the songs would be tough to say. If you want to average out difficulty, this might work (but might result in some BS calls for songs with sudden difficulty spikes, or for songs that only people who are really good at the game can play) Could be interesting to toy around with to see if something useful could be derived from it in terms of identifying difficulty.

[Patashu]

Originally Posted by One Winged Angel

What if we were to link the difficulty of the file to how many AAAs/FCs that file currently has in relation to every other song in the game?

-snip-

Anything that works on raw totals will obviously be flawed, as it treats newer files/files with less plays explicitly different from those with more.
Please read this and be enlightened as to how bayesian rating systems work:

http://www.andymoore.ca/2010/02/baye...rated-content/
http://www.thebroth.com/blog/118/bayesian-rating
http://www.stat.columbia.edu/~cook/m...an_sortin.html

When reading, replace in your head 'voting' and 'rating' with 'playing' and 'your score' and so on.

[jimerax]

Just integrated ingame song thread with this one, and changed the thread title (yay for sticky -1).

Now we have 13 pages of Misc songs, and only 12 Hip-Hop songs and 6 Funk songs.
So I think we need to do something about those genres.

(Idea 1)
Integrate Funk and Hip-Hop, then use one of those genres as Misc 2.

(Idea 2)
Move jazzy songs from Misc to Funk, then move Hip-Hop songs to Misc, use Hip-Hop genre as Misc 2.

opinions?

[MrMagic5239]

What about just combining Hip-Hop and Funk together?

And then using the left over genre for the legacy genre instead of recreating a whole new genre out of the support token one.

[bmah]

Although I agree on integrating categories with few songs, such as funk, why not create categories for the multitude of songs that share a genre that doesn't have its own category?

Misc 2 is way too vague; it's like some big "blob" of a category that should be divided a bit. Heck, why not make a jazz (J) category? There's actually a lot of jazz songs, and I know people will submit more jazzy songs. I also think there should be a Techno & Trance (T) category, to separate it from "Dance". Dance itself is extremely unspecific for a category.

I'd say only get rid of the funk category and put those songs elsewhere. I don't recommend eliminating the hip hop category. In fact, I'm making a hip hop file soon (and a funk one too, but there's way too little songs for the funk category to stay).

A category too broad like "dance" and "misc" results in a huge amount of files that isn't search-friendly. You might as well use the search list instead.

[OneHandNow]

Lol. The entire genre system needs to be reworked.

Dance, Dance 2, Funk, Hip Hop, Arcade, Classic, Rock, and Misc?

Funk/HipHop are basically empty sections..
Dance 2 is a jumble of anything "electronic/trance-y", while Misc is a jumble of IDM, jazz, electronica, drum and bass, basically anything that's vague.

I say we get some more specific genres like Trance, Breakbeat, etc..

[jimerax]

Yeah Jazz, Pops, DnB(Breakcore) should be there as new genres, but it requires a major change in the engine (won't be done until velo engine at least).

Funk and Hip-Hop don't have enough songs for their own genres so they just could be in Misc, in contrast.

[bmah]

Well ok, then the real answer is that the engine should undergo these major changes.

But for the short term, I guess what you have there is fine. I don't expect it to stay that way though.

[Coolboyrulez0]

I say keep Hip-Hop/Rap/Nerdcore but infuse the Funk Genre into rock instead of misc seeing as that would simply be more logical.

Also I am all for the creation of Hardcore, DnB, Pop genres once the time is appropriate.