community forum

Performance Points feedback and suggestions (Catch the Beat)

posted
Total Posts
768
show more
F D Flourite
Agree with Asriel on the Acc penalty part (which is opposite to Sorcerer's proposal) that it should be decreased. Because droplets are generated by random functions that slightly deviate the actual position of droplet from the slider position. They're not completely following the music anyways. So adding penalty for such a random algorithm doesn't seem to make sense for me.

And I'm a low acc player xD
PakaChan

PakaChan wrote:

These 2 patterns should have the same length bonus
by should i mean shouldn't (why can't i edit my post?)
Ttobas
Buff EZ, adularescence is 860pp DT and only 643pp EZDT, that is big for only an AR difference,
while it's true that CS makes some paterns easier, in overhall it's way more of a nerf due to hyper dashs not being spawned.
EZ not giving the right pp is more important than what you can think. What makes EZ hard (and low pp) is the same as what gives a lot of convert maps difficulty (and still low pp). So changing how the star diff is handled w/ EZ would benefit convert plays.


HD buff for low ar is a good idea but it's useless, there is no low AR map that would give enough pp to make a significant increase.

AR9 (AR9.5 for the extreme) HD is like nomod, what does it change ? The AR is already fast enough not to notice anything, and AR10 HD is a joke, more of a help than anything else.

Change flashlight pp to represent difficulty please, on AR lower than 6, flashlight shouldn't give anything noticable, especially on short map, and FL buff should increase as the AR increase (the faster the fruits scrolls, the more memory is needed), and increase as the star rating increase (because star rating should give an approximation of how "jumpy" the map is, even if it's not that good).

Asriel wrote:

+ decrease accuracy bonus (It's different with sorcerer)
if this system will keep that the most important thing is combo, accuracy's worth must be decreased
I'm totally against this.
Accuracy is something that should be important in this game, and it should be reflected in the pp.
If you're just a ppfarmer spamming fast DT overdose in hope you'll FC it, yeah good job get pp.
Lets just take this map for exemple, Endless Hatred.
How much person have SS'd it ? 3.
How much person have FC'd it ? 38.

So, maybe there is something hard in this map other than FC ? Maybe avoiding droplet missing isn't that easy ?

But, maybe you could increase accuracy importance when the score is an FC, or just account the number of droplet miss (and not miss, that already nerf you) when you calculate the accuracy importance of the score.


will edit the post if I think of something else
eldnl
Droplets are important imo bye
GiGas
Would be great if you could change the droplets being placed in the editor. Sometimes it doesn't fit the music. Even though you try to change the slider to make the slider fit the music it just doesnt come out the way you think. For high bpm maps it's easier to make the droplet shapes you want but for low bpm maps its a pain.
AAAAAAAAAAAAHHH
Hey. Yesterday i just explored the osu!catch code and I found out that it was really hard to edit anything when you can't understand the half of the values that are used in each functions lol
I think that the only way to change the code is to take a good dev and tell him to explore the whole osu! code to understand everything, or someone from the staff who knows well how the performance works. It's out of reach for me.

So I just did a short simple program to test an idea I had, by just editing the final pp value with a logarithm.
The aim of this is to reduce the insane amount of difference between really over-valued beatmaps (such as Uta or Envision) and beatmaps that worth less PP.

Why? Because I don't want to be forced to play the same 5 maps over and over again to rank up. With this idea, those beatmaps will be less "obligatory" and people will have to do more scores to get the same amount of PP.

How? I just took the final PP value as a PP base and edited it with this short program:
So, higher are the PP, higher the PP reduction will be (it's proportional to the base PP amount).

Obviously, someone could make the same thing more efficiently but it was just to illustrate the idea ^^

Here is the google sheet of my tests: https://docs.google.com/spreadsheets/d/ ... sp=sharing

I'll come back with some other ideas if I can find anything else ^^
Chicken-chan
Personally, I think a star rating calculation change than a pp calculation change would be better considering how much it's tied to pp - especially for maps that use a minimal amount of hyperdashing (or none at all), but use jumps that come just short like https://osu.ppy.sh/b/1486886&m=2 which ends up having a star rating that makes you massively underestimate its actual difficulty.

Basically use this idea here, but apply it to star ratings instead. You could fix two systems at once.

Skyleia wrote:

idk make it somehow so basically a jump which is not quite a hyperdash but is very far away is considered as more difficult than hyperdashes?

idk now it looks like it's something like this:

right now pp system is (most likely) based purely on distance, so a map focused only on pixel jumps is completely underrated

make it something like this instead:

so that a furthest pixel jump is rated as much as cross-screen jump, that is not the solution but will most likely help with many converts

also ignoring pp is of course an option, but it definitely wouldn't hurt if it was a little better
It'd be better if more technical maps that rely more on challenging patterns had a more accurate star rating to match, and in turn give an appropriate amount of pp.
Sorcerer

Sinnoh wrote:

I feel like the AR bonus could also be applied to non-hd, since EZ is still way under rated,

Ttobas wrote:

Buff EZ, adularescence is 860pp DT and only 643pp EZDT, that is big for only an AR difference,
while it's true that CS makes some paterns easier, in overhall it's way more of a nerf due to hyper dashs not being spawned.

EZ not giving the right pp is more important than what you can think. What makes EZ hard (and low pp) is the same as what gives a lot of convert maps difficulty (and still low pp). So changing how the star diff is handled w/ EZ would benefit convert plays.
EZ is going to remain tragically under rated until we can start making SR changes. A slight buff to the low AR Bonus isn't unreasonable and I'll update the spreadsheets to include one. The effect will be minimal though and we do need to consider not over rewarding lower diffs such as Cups or equivalent where the AR has less of an impact on difficulty.

Ttobas wrote:

HD buff for low ar is a good idea but it's useless, there is no low AR map that would give enough pp to make a significant increase.
Absolutely, the buff to low AR HD is mostly a side effect of nerfing high AR HD. Which is absolutely fine, a buff to low AR HD isn't unwarranted but its impact will be very minimal

Ttobas wrote:

Change flashlight pp to represent difficulty please, on AR lower than 6, flashlight shouldn't give anything noticable, especially on short map, and FL buff should increase as the AR increase (the faster the fruits scrolls, the more memory is needed), and increase as the star rating increase (because star rating should give an approximation of how "jumpy" the map is, even if it's not that good).
I took some time to play about with the FL buff but I can't reach numbers I'm happy with. Ideally the bonus would be a curve reducing at higher AR where memorisation is the primary factor but I cannot get the numbers to bend to my will. Scaling the bonus further to account for short maps (sub 200 combo) should be simple enough. If anyone can help out here that would be really appreciated

Worth pointing out that the current FL bonus doesn't account for AR at all and still applies the base AR bonus, meaning the FL bonus actually increases slightly as AR gets lower when comparing AR values under 8

Ttobas wrote:

But, maybe you could increase accuracy importance when the score is an FC, or just account the number of droplet miss (and not miss, that already nerf you) when you calculate the accuracy importance of the score.
There's no need to really, the penalty for misses and not achieving max combo will decimate pp values on a non-fc play, the acc penalty at that point is tiny in comparison

PakaChan wrote:

i think the way length is calculated needs a small revision, you don't actually move for every object in ctb. It makes streamy maps overrated.
FL gets a little broken because of this since it's jumpy maps that are the hard ones and streamy ones easier.

These 2 patterns shouldn't* have the same length bonus


Absolutely, I did initially look at adjusting the length bonus but was unsure if it was the right thing to do but I hadn't really thought about the point you raised.

In other gamemodes where every object, no matter how hard or easy it is, requires an input then scaling the length bonus with max combo works fairly well. But this doesn't apply to CtB where there's no guarantee that a note requires any sort of movement.

The ideal solution would be to calculate the length bonus in the actual difficulty calculator and only include objects that require movements, but any changes to the difficulty calculator look incredibly unlikely until it's ported to lazer.

Using total drain time instead has similar drawbacks to max combo but is probably more exploitable. It's easy to pad a map with long sliders in slow sections or long spinners.

For lack of a better solution I'll propose a nerf to the length bonus instead. It's quite strong right now and doesn't exactly achieve what it's intending to


F D Flourite wrote:

Agree with Asriel on the Acc penalty part (which is opposite to Sorcerer's proposal) that it should be decreased. Because droplets are generated by random functions that slightly deviate the actual position of droplet from the slider position. They're not completely following the music anyways. So adding penalty for such a random algorithm doesn't seem to make sense for me.
Despite the variation of droplets from the slider path (which makes me very sad), they're still the same for everyone and still require skill to catch. They require more precise and accurate movements across sliders and better control of the catcher's momentum when you hyper into them. I absolutely believe catching droplets takes an amount of skill and that skill should be rewarded

bastoo0 wrote:

So I just did a short simple program to test an idea I had, by just editing the final pp value with a logarithm.
The aim of this is to reduce the insane amount of difference between really over-valued beatmaps (such as Uta or Envision) and beatmaps that worth less PP.
This is quite interesting, atleast as a band-aid proposal to lessen the effects of the very inaccurate difficulty calculator we currently have. This does however go against the nature of SR a bit, where a higher SR map awards more pp under the assumption it's harder to fc. But currently the vast majority of high SR maps, about 8+, get their SR from hilariously over rated patterns which are usually high BPM streams, most others are using 2B elements which straight up breaks SR. This is definitely worth exploring some more

ChickenChanS wrote:

Personally, I think a star rating calculation change than a pp calculation change would be better considering how much it's tied to pp - especially for maps that use a minimal amount of hyperdashing (or none at all), but use jumps that come just short like https://osu.ppy.sh/b/1486886&m=2 which ends up having a star rating that makes you massively underestimate its actual difficulty.
Star rating calculation changes aren't too feasible right now and are more complicated than pp calculation changes so for now I'm working with what I can. Saying that, current SR calculation does infact rate edge dashes quite strong, it's just it only does so for extreme edge dashes. Try it out in the editor using dashes right on the border of being a hyperdash, there'll be a noticeable increase in SR. It looks like the difficulty calculator is too strict with what it considers an edge dash but there could well be a reason for doing so
Abstract-
Sorry for reviving this thread,

Users' Convert records have been considerably reduced since ppv2 I think,
as harder(or more pp-giving) ctb maps has come out, people lost motivation to play convert songs
what do you think if current pp ranking is divided into two (one for osu! catch specific pp ranking, one for osu! convert pp ranking) ?
Nelly
I'm okay with this decision but we need to fix the star rating first because it's broken as hell.
Karym
either we get separate pp or convert mapping pp gets a bit buffed (not only hyper spams but leftrights-fast sliders)
Lay
+1
Ascendance
it's not done in any other mode, and it shouldn't be done for catch the beat. whether it's a specific or a convert, you're still playing catch the beat, so there's no point in splitting it into two different leaderboards. it would be like splitting mania into 4k only and 7k only leaderboards. yeah, 7k gives more pp than 4k, but you're playing mania either way.
Lay

Ascendance wrote:

it's not done in any other mode, and it shouldn't be done for catch the beat. whether it's a specific or a convert, you're still playing catch the beat, so there's no point in splitting it into two different leaderboards. it would be like splitting mania into 4k only and 7k only leaderboards. yeah, 7k gives more pp than 4k, but you're playing mania either way.
I mean, you're not wrong. But I am not opposed to 1, fixing or readjusting the star rating, 2, buffing the pp received from said converts, because too many are underweighted or even under their true star rating.
Just my 10c, I know nothing of what I'm talking about!
Ascendance
I can agree to SR fix or buffing convert pp (especially the former, the latter is kinda meh whatever to me since converts aren't really meant for catch in the first place), but yeah I can't support a separate leaderboard.
Shiroi Lelouch
revert back to score ranking lul
Sorcerer
Saw some rumblings around about the need for a catch PP rework again so figured I’d share some thoughts on what I’ve been messing around with the past few months using a version of lazer that produces “mostly” accurate SR values.


It’s just gone past the four year mark for when catch star rating and performance points were overhauled in late 2014. Whilst the SR changes were at the time a huge improvement over the previous version, it has become painfully clear that catch SR still has big issues with representing the difficulty of a beatmap.

So in an attempt to inspire some hope in a community that has mostly given up on ever seeing a much needed “PP rework” I wanted to talk about some actual possible changes to the SR calculation. Keep in mind I’m no programmer or mathematician, I don’t have a solid understanding of how half this system works but I’m still confident I have found areas that can be improved.

I haven’t included any values here due to the limited setup I’ve been using, this is more for an idea of a starting point and what changes are worth pursuing.

Edge Dash Bonus
Currently maps with numerous edge dashes (regular dashes that are very close to the hyperdash trigger distance) are very underrated. There’s an existing bonus for edge dashes but it’s simply not big enough, increasing this is an easy fix but tuning the exact amount will require some trial and error.

Hyperdash Bonus
There’s currently a small bonus for both hypderdashes and edge dashes on a direction change. This bonus doesn’t do much for edge dashes but the bonus really adds up in higher difficulty specifics where hypderashes are frequent. Whilst hypderdashes are challenging for new players, a bonus like this gets too large on Overdose level difficulties where the hyperdash itself is not difficult but rather how it’s used.

Splitting this bonus into separate bonuses for edge dashes and hyperdashes allows for better tuning of both. Greatly reducing the hyperdash bonus really helps to bring down the inflated SR values for specifics vs converts/hyperdashless maps.

(Hypderdash bonus only applies on direction changes so hypderdash chains don’t get even sillier than they already are)

Speed Scaling
This is a simple issue, catch SR scales too much with speed. High BPM is overrated and part of why maps like Image Material have such high star ratings. Reducing the base speed scaling but keeping the speed scaling on direction changes similar is probably the best approach to balancing this.

Some maps are fairly dependant on speed and reducing the scaling can result in them being relatively underrated so tuning this will take some real trial and error. Though after four years of speed being so dominant I don’t think too many people would mind such a meta shake up.

A big limitation of the setup I used is the inability to test mods. I simply don’t know how changing speed scaling would affect doubletime and halftime but I’d assume reducing speed scaling would nerf doubletime. As far as I understand the difficulty calculator doesn’t currently consider the change in catcher speed, just the change in speed of the map. Adding a catcher speed scaling factor is probably the way to go about balancing DT and HT.

Direction Change Bonus
This would be the biggest change as it’s not just value tweaking but adjusting the logic. Currently there’s a bonus when there’s a direction change, which makes sense but due to the simple logic it leads to flowing patterns getting a much larger bonus than antiflow patterns. Compare these two patterns and where the direction changes are.

The “flow” pattern on the left is easier because the direction change is made on the note after the hyperdash finishes so the speed boost from the hyperdash is gone.

The “antiflow” pattern is noticeably harder, it requires more precise movements to make the direction change whilst hyperdashing because of the speed boost.

Despite this, the bonus is being applied to the movement after the direction change (represented with a black line) and doesn’t consider the movement before it. This leads to the flow pattern getting an enormous boost whilst the antiflow pattern sees very little. By scaling the direction change bonus with the distance and speed of the movement before it we now have a method to consider flow and antiflow patterns.

This seems very promising in the limited testing I’ve been able to do. The infamous flow stream pattern seen above is no longer the ticket to big SR values like it is now and antiflow patterns are actually rewarded.

Balancing this scaling is not easy. I saw good results from having a base direction change bonus that functions like the current but much smaller and then a much larger bonus that scales based on the movement before the direction change (antiflow). This ensures flowing direction changes still contribute to SR which is important for lower difficulties.

Widely Spaced Stacks
This is the best term I could come up with to describe a pattern like this as seen in Sing’s Master on Getty vs. DJ DiA – DropZ-Line-

Check it out in editor at 01:15:002 (1) -. Deleting this one slider reduces the star rating from 9.40 to 5.60

Stacks like these can be caught without requiring any movement from the player yet despite this they massively inflate the overall SR. The tricky bit is the initial positioning of the catcher, after that there’s no input required yet the longer and denser the stack, the greater the SR.

This is an issue currently on live, though only on a handful of ranked beatmaps, but it can become even worse when tweaking values like the edge dash bonus. It’s a bug and it needs a fix.

My proposed solution works like this:

- Check for direction change
- Check if distance of movement is <= catcher size
- Check if distance of current movement is equal to distance of previous movement
- Scale SR addition by time, 1.00x at 100ms down to 0.00x at 50ms

This could be considered a hacky solution but I was surprised by how well it appears to work. The checks try best to isolate the conditions of these stacks. They’re usually created by sliders, hence the requirement for identical distances, and it scales with speed because that’s when this bug really gets silly. The impact on innocent beatmaps seems very minimal if there’s any at all though I’ve not been able to check all too thoroughly.


Hopefully this can help provide a starting point for changes and show that the situation isn’t as hopeless as you might think. These changes won’t fix everything, but they should shake up the meta for the better and bring better balance to converts vs specifics.

Sadly, until catch star rating on lazer matches stable there’s not much work that can be done as all the tools provided for PP development are dependant on lazer. But once that’s fixed I hope there’s enough interest to get a group together with the know-how to actually start making some improvements.
Nelly
One problem. How can osu! team notice your proposal on forums if it's required to be posted on osu!dev Discord server?
AAAAAAAAAAAAHHH

Nelly wrote:

One problem. How can osu! team notice your proposal on forums if it's required to be posted on osu!dev Discord server?
You are right, I tried to access the osu!dev Discord server but my application was refused.
DakkyChan
There are some good points but you also forgot about some more. Especially with Spinners (pp should be also add like in ppv1 what rank you get in a map that would also makes spinning again famous because spinner is currently so underrated .And also some different mods especially easy and flashlight. What I personally understand about your points is just an improvement of the current system what is ok but it wont change that much . In my opinion the whole pp system needs a big rework of the current system because there wasnt any change since years and I can give you 2 examples why no changes kills a game (starcraft 2 and Pubg pc version because there was not really big changes.) in the past especially 2013-2014 there were a lot of changes and it really gave that game I would say the golden time cuz of these often changes .
show more
Please sign in to reply.

New reply