Stepfile Arrow Distribution (ermergerd data)

[NuclearShadow]

Didn't know where to post this so I went with this place.

I haven't played in forever but I saw my old pal Dynam0 (sup bruh) still plays so I've been playing the past few days out of nostalgia.

Then I was wondering if the stepfiles we play skew to certain arrows so I used JavaScript to capture only the keys I play on (D,F,K,L) and stored it in a variable. I played for about 30min and played songs in the Dance2 category between 50-80 difficulty.

Not sure if anyone has done this before but below are the results. Clearly the right arrow needs more love guys!



If anyone's a boss and wants to record their results for like an hour two of playing it'd be awesome to see. You can open the console and enter the code below. You'll have to figure out what the keyCodes are for your specific setup but once you do that all you have to do is play and then type "storedData" into the console without the quotes and you'll be able to see the results.

document.addEventListener("keydown", keyDownTextField, false);

var storedData = {
	input: {
		left: 0,
		up: 0,
		down: 0,
		right: 0
	}
};

function keyDownTextField(e) {
  var keyCode = e.keyCode;
  if (keyCode === 68) storedData['input']['left'] += 1
  if (keyCode === 75) storedData['input']['up'] += 1
  if (keyCode === 76) storedData['input']['right'] += 1
  if (keyCode === 70) storedData['input']['down'] += 1
}

[TheSaxRunner05]

That deviation is slightly more than I would expect

[ilikexd]

With such a small sample, the left to right difference isn't statistically significant, although I do suspect there will be at most a slight trend due to c1 bass syndrome and L->R planning. Up/down over left/right dominance is predictable since they're easier to hit.

[choof]

definitely gonna do this with some solo charts
I'll see how bad c1 bass syndrome really is

[NuclearShadow]

What's c1 bass syndrome?

Yeah I was aiming for 20k total arrows at the very least but I couldn't quite hit the mark Ideally it'd be nice to see the results of 2-3 hours or so of playtime but the songs would have to be recorded that were played and replays avoided since playing the same file would skew the data. I guess if someone were willing to go song by song through Dance 1 and Dance 2 over a week I could have the code write to a cookie instead.

[Dynam0]

wassup dude?!??!

c1 bass syndrome is where the chart uses primarily the left column to step the bass sounds which usually makes the chart biased to the left arrow.

I know eze and a few others have looked into file biases and it's typically the inner two columns that get more of the density. Left and right might actually normalize over a large sample size but it would be interesting to see.

[iCeCuBEz v2]

dont play M.A.M.A. it will skew the data to the left !!!

[dell2150]

Actually did the same thing once I downloaded Whatpulse (over a much larger sample) from both FFR and Stepmania. I found similar results, although L and R do move up a little bit over a larger grouping, I found that U was the most common. I will actually pulse before my session tonight and post results too

Come to think of it, for FFR you can't miss or get any boos (unless they perfectly offset each other, i.e. miss a right, get a boo on right), and on StepMania you can't ghost note at all. It seems like it would be tough to get a perfectly accurate reading for a large base but I'll try my best on StepMania later tonight, probably playing easier things from NBJS and ODI.

[_Zenith_]

EDIT: I like the graph so far, I would love to see someone (if I could play FFR I'd volunteer) play every song to see the entire graph on FFR.

Keep it going, definitely worth pursuing just to see the results!

[choof]

Originally Posted by NuclearShadow

What's c1 bass syndrome?

anchoring bass kicks to the left note
it's more blatantly obvious in solo files, specifically the iidx conversions in the bcmps, and some (all) of the hard files that 0 and myself make

[iCeCuBEz v2]

it really kinda irks me in solo having the bass drum on the left column for every beat starts to feel awkward as fuck for solo

i.e. that one part js part in himiko oni..... those patterns suck

[choof]

himiko oni is ez nerd

[Reincarnate]

Why not just use the raw stepchart data and get an aggregate distribution that way? No need to actually record keystrokes or anything like that. Furthermore, you'd be able to see if the distributions change at all if you change certain variables such as difficulty or genre.

[dAnceguy117]

whoaaaa, NuclearShadow! hey man!

not really relevant to the topic, but are you into programming?

[choof]

Originally Posted by Reincarnate

Why not just use the raw stepchart data and get an aggregate distribution that way? No need to actually record keystrokes or anything like that. Furthermore, you'd be able to see if the distributions change at all if you change certain variables such as difficulty or genre.

I think this would only work for stepmania, and not ffr?

[Reincarnate]

Originally Posted by choof

I think this would only work for stepmania, and not ffr?

I believe Velocity can process the FFR stepchart data (same way he computed things like gameStats.htm)

[NuclearShadow]

Damn there's still a lot of oldschool players here... sup danceguy! I am into programming but I'm not a programmer. I work in analytics so I do a lot with javascript but for the most part it's working with cookies, event listeners, plugins, APIs, and things like that. Other than that I'll use it to analyze data or work on personal projects. Are you a programmer?

Well if Velocity has a way to get to this data does anyone know how to pull it? Is it available through the API? I'd actually be most interested in specific step artists to see if they skew to using certain arrows over others and then if, across the entire group of stepfile artists, it ends up flattening out any biases.

[hi19hi19]

Originally Posted by Reincarnate

I believe Velocity can process the FFR stepchart data (same way he computed things like gameStats.htm)

It is indeed possible with FFR charts.
I'm not sure if Velocity has some store of the raw frame data somewhere on the server, but even if he doesn't, there is an easy method.
_root.beatbox code can be extracted from level .swfs using a decompiler.
It will look like this:

_root.beatBox = [[105,'L','red'],
[109,'D','yellow'],
[116,'U','yellow'],
[123,'D','yellow'],
[127,'L','blue'],
[134,'R','red'],
[145,'U','yellow'],
[152,'D','yellow'],
[155,'U','blue'],
[162,'R','red'],
[170,'U','blue'],
[177,'U','red'],
[184,'U','blue'],
[188,'D','yellow'],
[198,'L','blue']

This is obviously data that can be very easily manipulated to get whatever interesting stats you want.
Dossar used Flare to view decompiled level data and Flasm to inject changes for things such as course mods and frame fixes- not sure if those are still the de-facto tools because I haven't bothered to do frame fixes myself in a few years.

Originally Posted by NuclearShadow

Well if Velocity has a way to get to this data does anyone know how to pull it? Is it available through the API? I'd actually be most interested in specific step artists to see if they skew to using certain arrows over others and then if, across the entire group of stepfile artists, it ends up flattening out any biases.

Here is a link to the FFR API beta. It currently can pull a player's levelranks and recent games.
Other interesting stats like a player's grandtotal and FFR credit count, as well as the arrow data of individual levels, are currently not available.
The frame and arrow placement data is never going to be made publicly available due to the ease with which that would enable botting programs.
I am unsure if this data is ever going to be made privately available to trusted users with API keys, as the risk still remains. You will have to ask Velocity about that.

In the past, users such as qqwref have accessed the FFR arrow data in a different way: by individually copying all the .swf files from the server via "undisclosed methods" (read: I'm not sure if it's actually been fixed yet so I'm not going to explain rofl) This is, fundamentally, not something we want you to try to do. In fact very strictly speaking I'm pretty sure it's actually completely illegal.
It has, however, led to some cool stat dump threads like this one: http://www.flashflashrevolution.com/...d.php?t=121963
There's a lot of data similar to what you described wanting to access in that thread.

Despite the fact I'm not encouraging you to go digging around on the site to find all the .swfs, the fact remains that there are at least two nearly complete sets of all FFR level .swfs (that I know of) out there: AJ (TC_Halogen) and qqwref have them, though I don't think either one has bothered to keep their collection super current. If Velocity is unwilling -for reasons I hope you find understandable- to give you access to the arrow frame data, contacting one of those two people might also get you what you want

[hi19hi19]

There is also a very crude yet easily available method of getting some interesting data: simply ask stepartists for the .sm versions of their FFR files, then get the stats you want by hand.

For example, to get the values for how many arrows there are in each column given the .sm, you can just open it up in DDReam and copy each individual column of arrows to a separate difficulty then check the notecounts.
Obviously time consuming and better done with code, but pretty fun to do nevertheless.
I would certainly be glad to send you the .sm data for all my files.

Some examples of what this method can give you...
If you look at the overall left/right hand arrow ratios (aka split LD vs. UR), the files I consider my "most serious" technical efforts tend to have extremely balanced ratios.

These technically strong files have ratios generally within 3% or less of each other:
The Labyrinth of Skadi is 1561/1577
Schur's Theorem is 672/668
Full Mast is 1479/1437

Some of my files have c1 bass syndrome to varying degrees:
Bloody Tears is 881/843
-+ is 674/637 (honestly impressive considering the amount of intentional one-hand bullshit we put in)
CQ Boss is 647/555, a pretty extreme example where I exclusively put the bass on the left arrow for large portions of the file

And of course compare this to my troll or "intentionally different" files, which tend to have pretty skewed ratios:
Capoeira Sundance is 665/715
Thunder Light is 1117/775
M.A.M.A is 447/816, almost surely the most unbalanced file in FFR by ratio. Though note if you treat the down arrow as the bass, it actually has really bad c1 bass syndrome, being split 0/447/406/410 haha

I remember Nathan (ilikexd) saying his good files are similarly balanced too. In fact, his T&J file is split exactly 2222/2222 between the hands. Interesting to note.

But of course this you can't read too much into this statistic...
For instance Turbo is very technically accurate and hand balanced at 668/657, but it still manages to be quite troll in the context of FFR's engine.

[TC_Halogen]

Originally Posted by hi19hi19

Despite the fact I'm not encouraging you to go digging around on the site to find all the .swfs, the fact remains that there are at least two nearly complete sets of all FFR level .swfs (that I know of) out there: AJ (TC_Halogen) and qqwref have them, though I don't think either one has bothered to keep their collection super current. If Velocity is unwilling -for reasons I hope you find understandable- to give you access to the arrow frame data, contacting one of those two people might also get you what you want

Distributing official FFR SWFs is something that is considered against the rules, and I am especially married to that rule as a staff member. It's not personally worth it for me to take the chance of distributing that content.

I have kept my collection reasonably current for the purpose of archival in the event of a catastrophic failure, and nothing more. It never hurts to have the content if FFR decides to finally disappear, as some will still want to continue to play.