Terminal Categorization

[xXOpkillerXx]

Re: Terminal Categorization

There is a level of subjectivity that can't be captured by code, but I think for a project like this it would work fine.

Yes that's why I said "semi-formalize" in the sense that there Has to be some base ground to distinguish each category but that there needs not be a 100% rigorious definition.

For example: Streams category is when xx% of the song is single notes at regular intervals, where xx is optimized subjectively.

[qrrbrbirbel]

Re: Terminal Categorization

I was only planning on doing 30-70 difficulty since that is all I trust my own opinion on before things become too easy or hard

If you want to do the coding I'm more than happy to shovel off more work on people haha.

Really I'm just glad that there has been such interest in it already. I was expecting crickets and jackals tbh

[xXOpkillerXx]

Re: Terminal Categorization

I was only planning on doing 30-70 difficulty since that is all I trust my own opinion on before things become too easy or hard

If you want to do the coding I'm more than happy to shovel off more work on people haha.

Really I'm just glad that there has been such interest in it already. I was expecting crickets and jackals tbh

I will only code it for you if you provide more well-defined categories. I won't mind either if you decide not to work on that. Your call.

[rushyrulz]

Re: Terminal Categorization

I was only planning on doing 30-70 difficulty since that is all I trust my own opinion on before things become too easy or hard

If you want to do the coding I'm more than happy to shovel off more work on people haha.

Really I'm just glad that there has been such interest in it already. I was expecting crickets and jackals tbh

It's definitely something FFR is missing.

[aperson]

Re: Terminal Categorization

I will only code it for you if you provide more well-defined categories. I won't mind either if you decide not to work on that. Your call.

Or you can just have people manually rate a subset of the total set of songs and then use that to train a backpropagation network to rate the rest.

[xXOpkillerXx]

Re: Terminal Categorization

Or you can just have people manually rate a subset of the total set of songs and then use that to train a backpropagation network to rate the rest.

I don't think that would work but I might be wrong. Basically I imagine you'd use the notes as the inputs, but I don't see how you'll get anything that converges towards this semi-defined categorization because of the massive amount of possible layer configurations of your network. How would you attempt it ?

Edit: like I said, Mina has already done all that work anyway so why do it all over again that's a bit counterproductive imo. His difficulty calc is based on patterns, it's like exactly what OP is asking for..

[qrrbrbirbel]

Re: Terminal Categorization

So here's what I have so far:

Technique -
Bursts, grace notes, crossover, poly rhythm

Stamina -
Length of song, resting 'hangtime'

Power -
Jacks, trills, gallops, jumptrill

Stream -
Stream, running man, rolls, stairs, jumpstream

Anchor -
Jumps, hands, chains, jumpstream


Questions comments concerns?

[xXOpkillerXx]

Re: Terminal Categorization

Well, have you read the edit to my last post ?
Why do you want to redo it all for and even with less knowledge on the subject ?

Your categories are most likely still way too vague to code; you can't just mix and match patterns like that.

[Antori]

Re: Terminal Categorization

running man should be in power with thrills and jacks in my opinion

[qrrbrbirbel]

Re: Terminal Categorization

If you want to message Mina to get them in on this that would be great!

And as far as the coding goes I told you already I'm bad at that, and for the original purpose of this thread that's exactly what I had in mind. 5 or so elements that make up ffr as a whole. The fancy names I have in place are not set in stone by any means and are more or less an example of concept.

Dont hate me cuz I'm slow :wah:

[xXOpkillerXx]

Re: Terminal Categorization

If you want to message Mina to get them in on this that would be great!

And as far as the coding goes I told you already I'm bad at that, and for the original purpose of this thread that's exactly what I had in mind. 5 or so elements that make up ffr as a whole. The fancy names I have in place are not set in stone by any means and are more or less an example of concept.

Dont hate me cuz I'm slow :wah:

No worries I'm not hating nor thinking you're slow, I'm just trying to save you a lot of time. If people find your categories helpful/meaningful then hey that's awesome. Also Mina and I do not have the best relationship so you might want to ask him yourself. Good luck !

[emulord]

Re: Terminal Categorization

So here's what I have so far:

Technique -
Bursts, grace notes, crossover, poly rhythm

Stamina -
Length of song, resting 'hangtime'

Power -
Jacks, trills, gallops, jumptrill

Stream -
Stream, running man, rolls, stairs, jumpstream

Anchor -
Jumps, hands, chains, jumpstream

Questions comments concerns?

This is a good list. I feel like Stream shouldn't have jumpstream, because it doesn't match the rest of the entries in Stream. Keep JS in Anchor.

Then the categories are somewhat easily detectable in programming.
Technique = variation in frames between notes and # of colors.
Stamina = Total # of notes
Power = ?? Hardest to code actually. Highest NPS per column?
Stream = When single notes are the same frame distance as the last note +/- 1 frame.
Anchor = #Jumps + 2*#Triples + 3*#Quads


After the program analyzes the raw file data, the community can refine it, but its good to have an objective baseline.

[qrrbrbirbel]

Re: Terminal Categorization

This is a good list. I feel like Stream shouldn't have jumpstream, because it doesn't match the rest of the entries in Stream. Keep JS in Anchor.

Then the categories are somewhat easily detectable in programming.
Technique = variation in frames between notes and # of colors.
Stamina = Total # of notes
Power = ?? Hardest to code actually. Highest NPS per column?
Stream = When single notes are the same frame distance as the last note +/- 1 frame.
Anchor = #Jumps + 2*#Triples + 3*#Quads


After the program analyzes the raw file data, the community can refine it, but its good to have an objective baseline.

Alright that sounds good. That makes it an even 4 across the board except stamina.

I hear ya about power. I was thinking you could take the highest beat and then look for repeats of the same arrow to avoid the program recognizing beginner-level jacks lol



I also decided against putting triplets in as playing a song with crossovers will get you the same timing memory, but if you guys think there's a place in the categories let me know!



And I'm not worried op lol I was just messing around with not finding a consensus, no ill will taken or given!

[xXOpkillerXx]

Re: Terminal Categorization

I feel like Stream shouldn't have jumpstream, because it doesn't match the rest of the entries in Stream.

How does it not "match" the rest ?

Keep JS in Anchor.

Isn't anchor like {1, 12, 13, 1, 14, 1} (anchor on 1) ? I've never heard an other definition of it on here. JS can have litterally 0 anchor.

Technique = variation in frames between notes and # of colors.

How do you account for pauses (pretty common thing) ? No matter long or short ones they'll add up in the variation without being "technical". You need a quite rigorous formula to code this one.

Stamina = Total # of notes

This can be alright I guess, but it'll be biaised by the difficulty of the files (i.e. harder songs have more notes in less time in general).

Power = ?? Hardest to code actually. Highest NPS per column?

This won't work: any spike will make the nps high without it be the mentionned patterns specifically. I've worked with nps stats already and they need careful handling.

Stream = When single notes are the same frame distance as the last note +/- 1 frame.

That can maybe work yeah, without counting repetitive notes (jacks).

Anchor = #Jumps + 2*#Triples + 3*#Quads

See my point above about anchors. Idk what this would represent.

After the program analyzes the raw file data, the community can refine it, but its good to have an objective baseline.

Although you would now have some way to compute the categories (if fixed the above), you also have to aggregate the numbers in some sort of scale too, since you don't have specific units so like category 1 can give 183 and category 2 would be 1036 for example. You could scale everything by having the max of all 5 be equal to 1.0 and the rest would vary between 0.0 and 1.0, but that loses file-to-file comparision. Gotta think about that step too.

[qrrbrbirbel]

Re: Terminal Categorization

I used anchor like a weight system. Each additional note you have to hit adds more weight or anchors your hands in more definite positions.


Also makes sense technique would be a bitch to code, the category itself is filled with tomfoolery