Stepmania Difficulity Correcter [WIP]

[Dinglesberry]

EDIT: the SML live parser is garbage apparentlly - However, this isn't meant to be a "some amazing tool that every simfile editor should use to rate a song, it has a perfect algorithm that accurately determines difficulty", more of a "I have a bunch of old packs where the difficulties don't really make sense, this can approximately fix them"

If you are going to use it, for now I would recommend just trying it on one pack at time. It can definitely take a very long time for alot of songs, and there isn't a progress bar at the moment - I used it on my full song directory (~1000 songs) and it took a long time (it's around 3-4 seconds per song perhaps, depending on how long the song is, so you could imagine for 1000 songs, (4 * 1000) / 60 = ~60 minutes minutes for 1000 songs).

Here's something I've been working on for stepmania, made some decent progress on it, so I'll post what I've done:

Basically, one thing that's annoyed me about stepmania is that the difficulties are arbitrary, see ya later stuff like "Extratone Pirates - Diff: 69".. It's definitely funny sometimes, but I am looking for something that lets me judge how difficult a song truly is. There are some scales, one of which is the http://smleaderboards.net/difficulty_calculator (the calculator that FFRmania uses, etc). I feel like this one is pretty accurate.

So, the program uses:

-Apache commons i/o library for folder parsing
-Apache commons httpclient for POST commands
-gson to parse JSON results
-Swing for GUI
-http://smleaderboards.net/difficulty_calculator to parse SM files

What it will do is go through your entire song folder, using the stepmania leaderboards file parser to calculate the correct difficulty. After it is done analyzing the files, it will tell you the corrected difficulties, and if you press "Convert" it will update the SM files.

Even though the example is just showing individual packs being converted, it fully supports arbitrary file structures (e.g. you could run it on the main Songs folder, and it would go through every pack, or you could run it on an individual song folder and do just a single song).

Screenshots/Example Output:

Nuclear Blast JS Awesome Bomb Filez 2
Detected 15 songs:
Blockbuster - Old Diff: 10 - New Diff: 23.19
Chipstream - Old Diff: 15 - New Diff: 9.68
Comsten - Old Diff: 20 - New Diff: 35.89
Disturbed Relationship - Old Diff: 17 - New Diff: 27.94
Double K.O. - Old Diff: 10 - New Diff: 21.99
Gravity Blast - Old Diff: 15 - New Diff: 23.69
Ice Candy Pop - Old Diff: 10 - New Diff: 20.94
Lamerina - Old Diff: 10 - New Diff: 25.73
Laser Canon Deth Sentence - Old Diff: 15 - New Diff: 25.24
Love Hurts - Old Diff: 19 - New Diff: 24.66
Ripcord - Old Diff: 10 - New Diff: 21.94
Snow Goose - Old Diff: 10 - New Diff: 24.58
Stinger - Old Diff: 14 - New Diff: 25.04
This Calling - Old Diff: 20 - New Diff: 20.99
Welcome to Drakcore - Old Diff: 13 - New Diff: 26.43

Right now it's just a WIP, there are some things I want to add before I actually release it:

- better log window, where you could potentially check/uncheck folders you want to include in the conversion
- since the smleaderboard parser only does the first song difficulty, I need to find a way around this

At the moment, it's limited by the speed it takes the SMLeaderboard parser to return the file, which is honestly really quick. For a 50~ files, it takes a couple minutes, so if you have a song folder with thousands of songs, it will take a while

Download Link: http://dinglesffr.byethost5.com/tools/Fixer%20beta.zip

Honestly unsure if I'll update this further, it accomplishes its purpose pretty well for what it is. If the people want it, perhaps

[YoshL]

Originally Posted by Dinglesberry

There are some scales, one of which is the http://smleaderboards.net/difficulty_calculator (the calculator that FFRmania uses, etc). I feel like this one is pretty accurate.

..

[hi19hi19]

SML difficulty algorithm :/

[hi19hi19]

I mean I appreciate the attempt, at least

[Dinglesberry]

lol dude have you ever even looked at the output for the smleaderboards parser? It's better than manually setting the difficulty of the song to 9001 anyways, and it factors your left/right hand separately, accounts for things like one handed trills. Even if its ".." its a better approximation than nothing.

I'm really just making it for my own fun/benefit, already I've been using it and it's way easier to pick songs at my level, ran it on the FFR Community packs and boom. I know full well it won't be 100% accurate, but like I said its better than something like Nuclear Blast 4 where "Aim Burst" is Rated a 10 and Still Blastin is like a 13..

I'm not intending for a "this should be some amazing tool that every simfile editor should use to rate a song, it has a perfect algorithm that accurately determines difficulty", more of a "I have a bunch of old packs where the difficulties don't really make sense, this can approximately fix them"

[Fantasticone]

If my files aren't the hardest 1.0 ones then its not correct. No download link boo.

[Dinglesberry]

If anyone wants to try it or use it:

http://dinglesffr.byethost5.com/tools/Fixer%20beta.zip

To use:

- Press browse and select the song directory. Supports arbitrary folder (could do the Songs folder, a specific pack etc).. I suppose you could even do the root folder of your drive or w/e lol
- After selecting the folder, press analyze (or converter), this will output the details of the new difficulties
- After that's done press convert, this rewrites the SM files

EDIT: Also I just realised, you need to go to options -> Reload Courses/Songs ingame for it to re-update

[RenegadeLucien]

SML algorithm was somewhat accurate until some people took a dump all over it

This would of course be much better if you could link it to Mina's difficulty algorithm instead

[Dinglesberry]

Originally Posted by RenegadeLucien

SML algorithm was somewhat accurate until some people took a dump all over it

This would of course be much better if you could link it to Mina's difficulty algorithm instead

Do you have a link to it? could probably do it, I know theres drama or whatever, I just want accurate ratings though, hell I could use both and average the ratings, or let you select which one on a per song basis with minas as default i suppose

[RenegadeLucien]

If you're asking about the actual algorithm, no. No one has that except Mina. You can get data values from Mina's Transcendence theme for the song packs that he has run his algorithm through. I don't think there's an easy database for the values like with SML, unfortunately.

[Dynam0]

Originally Posted by RenegadeLucien

SML algorithm was somewhat accurate until some people took a dump all over it

This would of course be much better if you could link it to Mina's difficulty algorithm instead

yep, it's horrendously inaccurate now

[Walrusizer]

lets start a petition to replace the sml algorithm with http://www.randomnumbergenerator.com/

[Fantasticone]

All of you and your disgusting Comsten abuse.

[SpaceGorilla]

wnd 1.2 abuse is worse than comsten abuse tbh

[YoshL]

Originally Posted by Walrusizer

lets start a petition to replace the sml algorithm with http://www.randomnumbergenerator.com/

*signed*

[Dinglesberry]

Lol I love some good drama, but like it just so happens that parser is public, so I used it.

If there's a better one, I'd use that. I can't see a better way to quickly approximate difficulty at the moment :/

Maybe if we parse a million files it'll time the site out Kappa

[SpaceGorilla]

yo dingleberry i actually like your difficulty thing
something open-source that does its own file parsing would be amazing (good luck finding out how to do that lmao) because it would give people the ability to make their own difficulty scales for each pack

[Dinglesberry]

Originally Posted by SpaceGorilla

yo dingleberry i actually like your difficulty thing
something open-source that does its own file parsing would be amazing (good luck finding out how to do that lmao) because it would give people the ability to make their own difficulty scales for each pack

Writing the parser for sm files to simply read them is peanuts, the problem is actually analyzing the patterns to determine difficulties. You can view a sm file as plaintext, just open it in notepad - the data for notes is 0s and 1s and you can understand it just by looking at it.

What's incredibly hard is determining what makes it difficult, like say a you take two same speed jumpstream, how do you judge which one is harder if the jumps are different, or the notes between the jumps are different?

There are alot of ways to judge difficulties, I've seen posts on the FFR difficulty forums where people cite the number of AAAs relative to plays. In my opinion, the best way to determine difficulties would be where stepmania syncs online and auto records scores, and difficulty would be machine learning where the parser learns to analyze songs with data from how individual players perform relative to their current skill and dynamically adjusts the difficulties, but I'm dumb and can't implement something like that lol, and it would require online integration with stepmania, and obviously it's not so easy to control the development of stepmania.

[SpaceGorilla]

Originally Posted by Dinglesberry

What's incredibly hard is determining what makes it difficult, like say a you take two same speed jumpstream, how do you judge which one is harder if the jumps are different, or the notes between the jumps are different?

one thing sml does (and i believe this is a good approach, in theory) is counts jack density and per-hand stamina

when i made my adjustments to the algorithm i improved jack density by counting not only the note directly before the "current" one, but also the one before that. you'd probably need some sort of exponential formula for jacks by bpm, because obviously an 8th jack at 200 bpm isn't going to make a big difference for most people, but if it's 400 bpm, that same jack can cause a big cbrush (ofc depending on context, but that should have be to analyzed as well)

[Cataclysm-]

just a reminder that TLD .8 is still rated 28