forum

How about a new SR/pp system for taiko?

posted
Total Posts
20

If maps can no longer break the SR/pp system

Yes please
34
62.96%
I don't care
10
18.52%
No thanks the existing system is good
7
12.96%
No thanks I'd like to keep my sweet pp from maniera, x.u., last fortune etc.
3
5.56%
Total votes: 54
Topic Starter
tkdLolly
If a more reasonable pp system came along :D


:idea: Assumption: maps with 1/6 and doublets i.e. 'overrated' pp features should and would contribute less SR and give less pp than they do now. :idea:

:arrow: If it means your pp (and the pp of others) may suffer a terrible, devastating, catastrophic, precipitous drop, would you opt for a better* pp system? :roll:
*As stated in assumption
X a v y
I think that DT should also be weighted less, or rather some 1/4 songs should be weighted more. Either way, I personally feel like only DT scores and 1/6 pp maps are dominating the best performance of most players, especially when it comes to the top 300 players. Some 1/6 maps' star ratings are way too high, and it should be changed. Since mapper discovered the boost in SR once adding 1/6s, the ranked maps usually have condensed 1/6s in a bar. I'm not discouraging 1/6 patterns, in fact I feel that they can be fun, but it strikes me how much the SR changes once adding in the 1/6 patterns.

If I were to suffer a drop from 1/6 and DT weighing less, I am totally fine with that. In fact, I think it makes sense for my DT and 1/6 map scores to weigh less. I understand that some players that only play HD/HR will benefit from that, and I think it only makes sense that the abovementioned maps shouldn't be weighted much higher than the HD/HR plays.

TL;DR: The SR system for boosting from a NM to DT map is way too high, and 1/6 maps have SRs that are too high. I'm supporting a change in the pp system for Taiko.
Raiden
The best way would be to actually come up with a better SR formula and giving it to the devs so they can work with that. Devs are open-minded: they'll work with it if you give them a good idea that they can work with.

As long as they work with what's there, they are obviously going to put priority into more important osu! features for the near future, such as lazer. And maybe after that, we'll have our own complete rework. But for now, unless they receive something they can work with properly, the efforts will go to first priority endeavors.

I personally can't come up with one since I am not proficient in neither maths nor coding, but anyone who is and wants this to be changed should start by doing stuff themselves.
Topic Starter
tkdLolly

Raiden wrote:

But for now, unless they receive something they can work with properly, the efforts will go to first priority endeavors.

I personally can't come up with one since I am not proficient in neither maths nor coding, but anyone who is and wants this to be changed should start by doing stuff themselves.
I've actually had ideas before posting; the poll is just for fun. I haven't much progress, but I'll share a bit about what I'm up to. Just to show that I've been thinking / daydreaming:

Ideas
Here's a short summary of what I have been experimenting with so far.

Close this box for those not interested in my thought process.

As a maths person, I first considered physics formulas learnt in high school in an attempt to find a most objective and deterministic method of assigning difficulty.

As an example, I tried taking the time in between notes and applied KE = 1/2 mv^2, with the 1/2 factor taken away because it's just a scale factor for my purposes and to be fair the hand has to move up and down in between notes, and the m factor just lying around. I multiplied a ∆t to the KE to make for some sort of a weighted sum until I figured out it's actually the same as cancelling v out with ∆t. So finally I calculated the sum of the reciprocals of the intervals in between notes over time = ∑∆t^(-1) T^(-1), i.e. sum of speeds divided by time = ∑v T^(-1) and this makes some sense. DT would be 2.25 times more difficult than NM on this scale, and the difficulty of 1/16 doublets every 1/4 : 1/8 stream according to my calculations would be 4:3.

I tried calculating the difficulty of various patterns at different lengths. Particularly ones if anyone is interested are
d d d _ d vs d d _ d d (repeating these patterns indefinitely would demonstrate that they are of similar difficulty but are they the exact same? also latter case includes offbeat),
d d d _ d d d vs d _ d d d d d (same as above but many find 5-stream more difficult than 3-stream at beginner level),
d d d d d _ d d d d d vs d d d d d d d d d _ d (extension of above case),
d d d _ d d d _ d d d _ ...vs d d d d d d d d _ _ _ ... _ _ d with same number of d's
and d d d _ d d vs d d _ d d d (resultant from adding one note to first case reveals the same pattern but reversed, which shows that EITHER the both options in the first cases are of the same difficulty OR addition of notes doesn't work this way ON THE ASSUMPTION that reversed maps have the same difficulty).

It should be noted that I ran into cases where adding notes actually decreased the difficulty since the model takes an average, and the addition of the note lowers that average, which reflects that this model is somewhat flawed or incomplete.

A easily noticeable and fatal drawback to this model would be dealing with extremely close doublets which would skyrocket the difficulty e.g. notes 1ms apart. Just before posting, I had yet another insight. Other than considering pairwise intervals of neighbouring notes, we can also try considering two sets of intervals, between odd-numbered notes and even-numbered notes i.e. skipping a note in between i.e. kddk full alt, and we look at what each hand is doing. We then combine both sets to determine the actual difficulty of the map. This is one approach I'm currently considering to actually kill doublet pp.

Another way that I considered but doesn't do doublets justice is to just 'round them off' to a more manageable, regular timescale.

P.S. Note that I never considered the colour of the notes to be a factor because the difficulty would become subjective.

I'm not sure we'd all welcome pp drops though xd and I'm curious to find out how many of us prefer fake, dirty (ahem I mean artificial pp

getting really late and I'm on my phone, later lol
BabySnakes
To be honest, DT doesn't need that big of a nerf, other mods do since it's impossible to progress any further if you can't play HR or HD, that's mainly because there aren't many difficult maps that punish HR or HD so they're basically free pp.
Cqfuj
While I agree on 1/6 give too much pp, I'm less sure about doublets. Actually, if you check the current object difficulty computation, it is not exactly 1/6 but pretty much every noticeable rhythm change.
It should be noted that ddkddk patterns are also a problem.

tkdLolly wrote:

... ON THE ASSUMPTION that reversed maps have the same difficulty).
I don't think this is the case. For example a long stream with bpm increasing will be harder than a long stream with bpm decreasing. Let's say from 120bpm to 200bpm (1/4). In both streams the hardest part is the 200bpm section, but in the first one when you reach it you're tired because of all the previous notes. Meanwhile the second one starts with the hard part when your hands aren't tired. Even though you will get tired faster by starting with the hard part, I think it is still easier to do.

tkdLolly wrote:

It should be noted that I ran into cases where adding notes actually decreased the difficulty since the model takes an average, and the addition of the note lowers that average, which reflects that this model is somewhat flawed or incomplete.
I don't think taking an average value is the right way. You should take the hardest notes and weight them. For example the hardest note represents 50% of the whole beatmap difficulty, the second one, 25% and the third and forth one, 12.5% each. Pretty much the same way a user's global pp are computed. The current star rating has a similar process but uses intervals. This will help you avoiding difficulty drops by adding easy notes ;)
Edgar_Figaro
I honestly feel like 1/6 and doublets are just fine for SR and the problem lies more in long streams of notes being undervalued in comparison. Extremely long patterns of 1/4 are much easier to notelock or fall behind on then quick 1/6 bursts and are harder. Now if there was a long stream of 1/6 this should still be considered hard as it would require you to essentially play the map at a higher bpm (a 20 note 1/4 stream on bpm 300=20 note 1/6 stream on bpm 200 and should equal the exact same SR)
Topic Starter
tkdLolly

Cqfuj wrote:

I don't think taking an average value is the right way. You should take the hardest notes and weight them. For example the hardest note represents 50% of the whole beatmap difficulty, the second one, 25% and the third and forth one, 12.5% each. Pretty much the same way a user's global pp are computed. The current star rating has a similar process but uses intervals. This will help you avoiding difficulty drops by adding easy notes ;)
Interesting... :O I was trying to start from scratch so I didn't look for or look at the existing code, thanks for sharing :)

Edgar_Figaro wrote:

I honestly feel like 1/6 and doublets are just fine for SR and the problem lies more in long streams of notes being undervalued in comparison. Extremely long patterns of 1/4 are much easier to notelock or fall behind on then quick 1/6 bursts and are harder. Now if there was a long stream of 1/6 this should still be considered hard as it would require you to essentially play the map at a higher bpm (a 20 note 1/4 stream on bpm 300=20 note 1/6 stream on bpm 200 and should equal the exact same SR)
When I first started thinking, I thought in terms of BPM i.e. for the sake of demonstration a 1/4 stream at BPM 120 for all eternity bears a standard unit of difficulty, and since every pair of adjacent notes can be described as a 1/4 interval in a specific BPM, we can scale the difficulty according to the BPM, something like that, which could also be applied to DT although OD isn't thoroughly thought through :idea:
Jaye
As much as I'd like a new SR system, the problem with it is that people will just find a way to abuse it once more; pp mapping never truly dies.

As for a new pp system, eh. You could make it more like other modes, so 7 star maps are actually worth something, but really it's not necessary unless you're an all-mode player since pp doesn't translate across modes.

Still, a revised SR/pp system would be nice and perhaps add a bit more fairness and polish to the mode.
-Kazu-
looks like this is going to be an unpopular opinion but I think pp in general should be higher
I mean, you need to have SS on Kurokotei - Scattered Faith (the 10 stars version) + DTHD to get those 1k pp the ctb and mania players are getting :^)
Edgar_Figaro

-Anhedonia- wrote:

looks like this is going to be an unpopular opinion but I think pp in general should be higher
I mean, you need to have SS on Kurokotei - Scattered Faith (the 10 stars version) + DTHD to get those 1k pp the ctb and mania players are getting :^)
I personally think Taiko needs a small boost so that it's on par with Standard. CTB and Mania need to be nerfed
BabySnakes

Jaye wrote:

As much as I'd like a new SR system, the problem with it is that people will just find a way to abuse it once more; pp mapping never truly dies.

As for a new pp system, eh. You could make it more like other modes, so 7 star maps are actually worth something, but really it's not necessary unless you're an all-mode player since pp doesn't translate across modes.

Still, a revised SR/pp system would be nice and perhaps add a bit more fairness and polish to the mode.
Yes, people will find a way to abuse it, but it won't get to the point of having 8* maps that are actually 5*.
The SR is so bad, like really bad, that any improvement would be welcome.


PP maps will keep coming and eventually you'll be able to get to 10k pp with no effort, imagine that you could get pp everytime you submitted a play, even if you already did it before. That's eventually happening but with 550 pp scores on 5k pp skill players. Have fun.
BabySnakes

X a v y wrote:

I think that DT should also be weighted less, or rather some 1/4 songs should be weighted more. Either way, I personally feel like only DT scores and 1/6 pp maps are dominating the best performance of most players, especially when it comes to the top 300 players. Some 1/6 maps' star ratings are way too high, and it should be changed. Since mapper discovered the boost in SR once adding 1/6s, the ranked maps usually have condensed 1/6s in a bar. I'm not discouraging 1/6 patterns, in fact I feel that they can be fun, but it strikes me how much the SR changes once adding in the 1/6 patterns.

If I were to suffer a drop from 1/6 and DT weighing less, I am totally fine with that. In fact, I think it makes sense for my DT and 1/6 map scores to weigh less. I understand that some players that only play HD/HR will benefit from that, and I think it only makes sense that the abovementioned maps shouldn't be weighted much higher than the HD/HR plays.

TL;DR: The SR system for boosting from a NM to DT map is way too high, and 1/6 maps have SRs that are too high. I'm supporting a change in the pp system for Taiko.
I'm sorry, but do you even know how DT works? It just simulates the sped up map and adds OD, if there is something to be balanced, it's the SR rating, not DT. If you complain about how DT boosts OD, you'd also have to bitch about how HR also boosts OD. DT isn't to be blamed it's only the result of a trash SR and PP system.
Sieg

tkdLolly wrote:

That's nice math you've done. Tho as being said not average but weight values based on track duration would be better. Also you could try not to throw away colors but carefully experiment with bonus ratio values on multiple changes. Anyways I recommend to collect different kind of statistics from ranked sets beforehand - density, time\notes values, tracks durations etc. this will help you to get your reference points. huh.. wish i would less lazy and care more ):
dewero
I have actually worked on an alternative SR system for taiko a long while back, https://osu.ppy.sh/forum/t/484562

The main idea that I implemented in the program was to have two measures of difficulty; density and complexity. Density was simply an average of the time between each note in 10 note windows. Complexity was determined by measuring the difference between the original pattern's length and the pattern's length after compression (if we assume dons and kats to be 0's and 1's). I then combined these two measures for each note and performed a weighted sum to determine difficulty.

The results were interesting but still broken. One thing that was mentioned early on was the fact that the calculator had no way of determining stamina, so less dense streams were underweighted even if they were very long. I kinda lost interest around this time though.

Maybe someone could refine my ideas to make a better calculator? I'd be happy to start working on it again if people have suggestions
Yuzeyun
Let's br real; the best way to have a well-balance Star Rating system is to break down a beatmap's different skillsets into atoms - something I've wanted to do with my old SSR project (that died quickly...)

So far, I've got different skillsets - you can also propose a skillset and we can discuss if this is a good idea or not. The list is:
Stamina - I don't need to explain this one
Stream - How well you can keep combos on a stream
(Speed) - I defined it as a combo of Stamina and Stream, but this could be subject to change if there are good arguments in favour
Accuracy - Mainly defined by OD, this can be subject to change or be deleted
Reading - How well you can read different speeds; HR might break this in all honesty with every screen ratio. However, very low as much as very high scrolling speed will be hard while average will be treated as easy. "Comfort" values are not decided, but I suspect anything between 150 and 200 BPM is the most comfortable for most players.
Technical - Mainly for maps that use 1/6 + 1/4 streams, or anything related handswap-heavy patterns and doubles.
Overall - Will be the end value of the map: It is a combination of every other skillset rating which will be calculated to be as close as possible to reality.
It also means that, using this system, you will have to account for SV speed (can be calculated easily as the product of SV change, base SV and BPM).

The "Overall" value is pretty much the end SR value. Imagine, on a 1-10 scale, a beatmap that has these values:
Stamina 9.73 | Stream 7.28 | Reading 5.58 | Technical 3.29 | Overall 6.31
You will get a beatmap with a star rating of 6.31 - but the breakdown will give you a more accurate outlook on how hard a map is. This is something that is currently used in the Etterna build of Stepmania, and so far, it is fairly accurate.


tl;dr : Break down a beatmap to its elemental skill set (stamina, stream, reading, tech) to have a more accurate star rating. it also may be adjusted as well with mods like HR or HD, as reading is affected by these mods.
BabySnakes

Unmei Muma wrote:

Let's br real; the best way to have a well-balance Star Rating system is to break down a beatmap's different skillsets into atoms - something I've wanted to do with my old SSR project (that died quickly...)

So far, I've got different skillsets - you can also propose a skillset and we can discuss if this is a good idea or not. The list is:
Stamina - I don't need to explain this one
Stream - How well you can keep combos on a stream
(Speed) - I defined it as a combo of Stamina and Stream, but this could be subject to change if there are good arguments in favour
Accuracy - Mainly defined by OD, this can be subject to change or be deleted
Reading - How well you can read different speeds; HR might break this in all honesty with every screen ratio. However, very low as much as very high scrolling speed will be hard while average will be treated as easy. "Comfort" values are not decided, but I suspect anything between 150 and 200 BPM is the most comfortable for most players.
Technical - Mainly for maps that use 1/6 + 1/4 streams, or anything related handswap-heavy patterns and doubles.
Overall - Will be the end value of the map: It is a combination of every other skillset rating which will be calculated to be as close as possible to reality.
It also means that, using this system, you will have to account for SV speed (can be calculated easily as the product of SV change, base SV and BPM).

The "Overall" value is pretty much the end SR value. Imagine, on a 1-10 scale, a beatmap that has these values:
Stamina 9.73 | Stream 7.28 | Reading 5.58 | Technical 3.29 | Overall 6.31
You will get a beatmap with a star rating of 6.31 - but the breakdown will give you a more accurate outlook on how hard a map is. This is something that is currently used in the Etterna build of Stepmania, and so far, it is fairly accurate.


tl;dr : Break down a beatmap to its elemental skill set (stamina, stream, reading, tech) to have a more accurate star rating. it also may be adjusted as well with mods like HR or HD, as reading is affected by these mods.


Great contribution, I just have some concerns.


Stream should be a combo of speed and stamina.
Speed should be based of kps (but make sure to nerf doubles and triples).

Technical is kinda relative to each individual and playstyle.
Alchyr
I made something.

t/656048
Yuzeyun

babysnakes wrote:

Stream should be a combo of speed and stamina.
Speed should be based of kps (but make sure to nerf doubles and triples).
KPS is directly correlating with stream: 20 kps is pretty much a 300 BPM stream. However, it's also a 300 BPM 1/2 doubles section - which asks for less speed and more technique.
Stream is how regular a stream is and how long it lasts. A 340 note 1/4 stream will have a higher value than a 340 note 1/2.
Stamina is how long you keep it for. If you have to keep it for 40 seconds, it will have a higher value than if you keep it for 75 (still with the 340 note example).
Speed is factored in both, which is why having it is redundant with the rest of the sets. It can be mentioned as auxiliary, but never as a proper skillset.

If anything, about bursts, they start to be hard from 5 notes onwards (as it will require at least two finger taps on any note). 4 notes is debatable, 3 and 2 is not.
X a v y
I know I am a little late to the party, and you all are already coming up with better formulas for star rating. Since a lot of you are mentioning about the pp you gain from DT, I would like to add one thing. It would be great if someone changed the mods we have for Taiko.

This one is going to be an unpopular opinion. HDFL is abusable through dual screening, and I myself wouldn't consider it to be a skill since you aren't actually reading anything and the OD has not changed. I'd say pp should be removed for HDFL. If not, only keep the pp for HD mod.

I'm sure some of you will agree with me on this one, as many people, including me, benefit from this perk: The extra star rating by using DT. Even though a lot of people will reduce their pp after this, I think that it is worth mentioning that DT gives way too much pp. (You can interpret it as NM giving way too little pp)

On the discussion of mods, I think it's time for a change of the mods itself. I think it seems like a good idea to be able to change the OD, SV, and BPM of the song, and these properties are all adjustable on a slider.

HR will be split into two. OD and SV. The OD slider can be adjusted from OD 0 to OD10, much like when you create a beatmap. The SV can be changed as well. (I can't decide a range because I've never actually touched the 1.4SV and always kept it that way) The BPM, much like HT and DT, can be changed on a slider from 0.5x speed to, lets say, 2x speed.

Only the OD slider and BPM slider can make a change on the pp you gain from the song though, and the SV does not matter.

With these suggestions mentioned, I would like to ask all of you whether you think it is a good idea or not, and if you think it would benefit the pp system or not. If not, do you think the slider can be implemented anyways, just for Taiko players to have fun? (I've seen a lot of people make the same map except speed it up by 1.2 or 1.25x just because it's more fun that way, and thought this feature would be better for taiko)

TL;DR: HDFL shouldn't give pp. DT gives too much pp. We should be able to change the song's OD, SV, and BPM, but only OD and BPM will change how much pp we get.
Please sign in to reply.

New reply