forum

Performance Points feedback and suggestions (Standard)

posted
Total Posts
2,750
show more
Nyxa
By strain I mean quite literally the amount of effort (based on an average) it costs to do a certain something. I mean, there's so much available data on maps and scores on those maps, it shouldn't be hard to see what kinds of angles are relatively more intensive to play than others, along with rhythms as well. I'll agree that when it comes to pure aim my stamina between aim and tapping also doesn't compare. But I've noticed that in myself (and other players, I've even seen index mention this) as soon as I start running out of stamina in one hand, I run out of stamina in the other as well.

As for touchscreens, those always seemed to have had an apparent advantage when it comes to aiming. Isn't that why Tom disabled TAG4 pp?

Also, do keep in mind I'm just giving a rough idea of the general direction you could go in when considering how to calculate pp more accurately. I purposefully avoided any specific ideas because, in my experience, people tend to argue or neglect any precise propositions unless they reach them together.
B1rd
Strain is caused through having to tense your hand to hit small circles or to accelerate or decelerate your hand quickly. I experience way more strain playing 4.5* old cs5 maps compared to maps with jumps that flow well like Senketsu no Chikai.
Vuelo Eluko
huh, i guess i just don't think that kind of strain has a noticeable impact on playing, anymore than how hard it is to press a certain type of key down. in my experience if someone messes up on aiming it's reading related.
Yuudachi-kun
What is aimif everyone connects misaims to misreads?
Nyxa
Well reading has a strain of its own which is the time it takes for your brain to process what's on the screen and react to it accordingly. I don't get what's so "subjective" about that. The more things on the screen at once, the longer your brain will take to organize everything because it can't rely on reflexes. The less time you have to react, the more trained your reflexes will have to be to react to something. These are both two reading strains determined by the opposite ends of note density, and yet the attention paid to them is rather minimal.
DeletedUser_4329079
I have an idea, if aim speed and accuracy can be independent, why can't a new subcategory be added for flashlight?
Nyxa
Because most people have no clue what actually makes flashlight difficult. Memorization is the smallest factor, not the only one.
Purple
Strain as described by Tess is a real thing. When you have a series of jumps, the algorithm will calculate the aiming difficulty of each individual jump, and then this number is further increased as you have more jumps. That's strain, I think, but of the most simple variety.

If you have a series of jumps end on a spaced stream, that's a different matter. There is clearly a *lot* of effort that you need to put in order to hit all notes on the spaced stream at the end, but the current difficulty algorithm doesn't really take that into consideration, IMO. The effort you have to put to decelerate your cursor like that is definitely real, as B1rd said. I think this is also partly the reason why CS difficulty calculation is flawed and a flat bonus had to be put in place in order to fix star ratings for high CS maps.

That said, I don't think looking at angles alone is the way to implement a reading calculator. There are some good things you can do with it, but I don't think it's enough to fix highly overrated maps like Koigokoro. For that, you would need a pattern detector: an algorithm that looks at mapping patterns as a whole, instead of just calculating the angles from individual jumps. I'd say that if Tom wants to look into that, he should start very small and safe (as is many times the case in software developing), by testing individual patterns that are *for sure* a pain in the ass to everyone who plays this game and are also underrated by the current algorithm. I think that if you add a small batch of those to the current build, you would already be fixing DT farm maps by making everything else give much more PP. Alternatively, you could detect EASY patterns and then add a penalty, but to me that sounds more difficult.
GhostFrog
Regarding the angle thing that got mentioned briefly several posts back, I think a lot of underweighted and overweighted aim patterns could be brought more into line just by considering the angle between the current straight line path and the previous straight line path. If you call that angle theta, something as simple as d^(a-cos(theta))/t^(a+cos(theta)) rather than d^a/t^a (where I think right now a=3) would fix a looot of problems. Would need to use a smaller measuring stick for distance in order to make it actually work, but that would be probably easy enough to find a good approximate value for.

Doing that correctly would nerf "slow" DT farm maps relative to faster ones - DT farm maps have lots of jumps with small angles between consecutive paths and the change I'm suggesting would make such jumps more dependent on effective bpm and less dependent on distance between notes. On the other extreme, large spacing with really wide angles would be buffed, but would scale less with effective bpm - things like the ridiculous quintuples in Genryuu Kaiko or any pattern at all in Worldwide Choppers.
Drezi
I'd just like to drop by to say (and repeat) that I really do think that even simpler and smaller pattern/rhythm complexity evaluation logic could be added, since anything in the right direction would be a step forward and better than not considering these aspects at all. As long as we feel the average error of the algorhythm gets smaller that is.

For rhythm complexity simply considering variability of the rhythm to a small extent would be great, to reward accing stuff that isn't 1/2 spam almost exclusively a bit more.
E m i
now measure the relation between the music and in-game objects, hitsounds...
ChanSenpai_old
i suck at this game and I'm a loser lol 8-)
blahpy
Is this intended behaviour?



I'm assuming that perhaps the speed difficulty is being calculated based upon the length of streams based on time, rather than the number of notes? It seems very counterintuitive though.

Here is a spreadsheet of values: https://docs.google.com/spreadsheets/d/ ... edit#gid=0

Here is a copy of the mapset I made while testing if you want to inspect it, I only just updated it though so it hasn't calculated the online difficulty ratings yet: https://osu.ppy.sh/s/298070

edit: This is, of course, a very primitive test, but I tried to minimise the effect that could happen due to aim or accuracy by using OD0 and perfect stacks with stack leniency 0. I don't claim to know how the pp/tar diff systems work but I do fine this rather odd

edit2: It was suggested that I also test this with different length streams to see if it changes and perhaps becomes nullified, I may do that at some point soon. 32 was just an arbitrary power of 2 that's a decent stream length in order to make my copying and pasting easier
Vuelo Eluko
Now i know why Tom is planning to buff high bpm, since 240 seems to be the highest bpm that is being properly rewarded.
Kynan
blahpy's post said it all, just check the star rating of Mad Machine DT if you want a ranked map to test on, it's way too low.
jesse1412
Tom pls why would you do this to us.
Barusamikosu

Nice post count Jesus. Well memed, I know.
GhostFrog
With so few notes, this could be another little quirk related to the fact that tom's difficulty algorithm breaks the map into short time intervals and chooses the highest strain from each. Or at least it did back in the days of tp. Probably still does.
Yuudachi-kun
Try 64 and 128 notes?
DroidBass
Would been a good idea about considering stream lenght by time instead amount of circles at all. Also both cases of very long streams and really fast stacks (355 bpm likely seen at xi - Breakthrough Atmosphere +DT because streams of 1/6 at 237 bpm) are both cases underrated. Seems that the meta for pp is finding a map full of "streams" 5-9 circles frequently, not complicated stream shape (mainly lineally and fluided ones) with higher spacing between circle and OD equal or over to OD8. Maps with streams of +30 circles and over 4 seconds of tapping aren't that seen in people's best performance, mainly because they're highly underrated compared to easier maps that can give more pp for less effort.

I find confusing that on some cases a 5.1 stars map likely Tristam & Braken - Frame of Mind [Chill] can give more pp than maps with 5.48 stars likely Shounen Radio - neu [Gold]...

Facts to consider
1.- Shounen radio has 821 elements, Tristam has 1042 elements
2.- Shounen radio is OD7, Tristam is OD8
3.- Tristam is much more consistent than Shounen radio is
4.- General ranking on top 50's talks by itself.
Personal tought: Tristam [Chill] is being highly farmed because it gives a way too much pp from accuracy value, instead shounen radio [gold] is avoided due it needs strong constancy at fingers on some parts and being massively underewarded because it's worth as very low accuracy on pp.

What I want to highlight from this?
1.- There should been a star system from "average star level" and "hardest star level", this would allow people to find maps to properly train constancy or instead going ahead for maps with high peaks of difficulty.
2.- General pp obtained from a map SHOULD be based on "average star level" and if this is worked finely, it should been balanced on both cases that requires or constancy or a nuke of skill on atleast a single part. This would take lots of effort, but looks promising.
3.- Accuracy is being overrated (mainly OD+8 with many circles). This makes people to avoid maps that they can't do with really good accuracy and instead they go for free accuracy = free pp maps. Needed of more tries to FC 95.50% Shounen Radio [gold] than to FC Tristam Frame of Mind [Chill] with 98.02% accuracy.

Preffered to not extend on too many topics at the same time, just mentioning one and describing another clearly.
blahpy

GhostFrog wrote:

With so few notes, this could be another little quirk related to the fact that tom's difficulty algorithm breaks the map into short time intervals and chooses the highest strain from each. Or at least it did back in the days of tp. Probably still does.
Good theory, could be the reason yeah.

Kheldragar wrote:

Try 64 and 128 notes?
Will try some different numbers of notes later today when I have time
Nyxa
Pretty interested for the results. Also, the pp vs. star rating thing DroidBass mentioned is legit. Some star ratings seem really off, and it still feels counterintuitive to me that a map of a certain SR would give less pp than a map of lower SR than the first.
blahpy

Tess wrote:

it still feels counterintuitive to me that a map of a certain SR would give less pp than a map of lower SR than the first.
This can happen often due to accuracy rating due to OD, length, etc, i'm trying to minimise that effect here though
Topic Starter
Tom94

GhostFrog wrote:

With so few notes, this could be another little quirk related to the fact that tom's difficulty algorithm breaks the map into short time intervals and chooses the highest strain from each. Or at least it did back in the days of tp. Probably still does.
This is probably the reason. In the graph you can see how the amount of BPM difference increases between each of these "steps". Those breakpoints are where the 32 note stream starts being completely contained within one less of these intervals.

If you'd want to compare the difficulties of such streams I suggest you make a map at least 30 seconds long where you just repeat those streams with a short break inbetween. Sadly there could still be aliasing artifacts when the BPM and the time interval share common divisors, but they shouldn't be as bad.

The sudden knack at 300 BPM is by the way currently caused by a hard cap avoiding singularities in maps putting some hitobjects 1ms or even 0ms apart. I should probably make that higher now that some people actually start to pull off 300bpm scores.
silmarilen

DroidBass wrote:

Facts to consider
1.- Shounen radio has 821 elements, Tristam has 1042 elements
2.- Shounen radio is OD7, Tristam is OD8
3.- Tristam is much more consistent than Shounen radio is
4.- General ranking on top 50's talks by itself.
you forgot 1 major thing.
tristam has 923 circles
shounen radio has 652 circles
sliders arent considered for acc, only for beatmap length.

so not only is tristam a higher od, it also has almost 300 circles more.
when aim/speed/acc get divided you will see that tristam most likely gives 30-40pp more in the acc part compared to shounen radio. while shounen radio will give more in the combined aim/speed part.

unfortunately pp cant tell how hard a map actually is to acc, it purely looks at od and amount of circles for that part afaik.
Nyxa
IIRC that's the main reason why HR often feels undervalued. If the map doesn't score high in aim or speed but is still hard to acc (like marathons) it won't give any decent pp below 99%
Vuelo Eluko
also because most people don't have high acc.
blahpy

Tom94 wrote:

GhostFrog wrote:

With so few notes, this could be another little quirk related to the fact that tom's difficulty algorithm breaks the map into short time intervals and chooses the highest strain from each. Or at least it did back in the days of tp. Probably still does.
This is probably the reason. In the graph you can see how the amount of BPM difference increases between each of these "steps". Those breakpoints are where the 32 note stream starts being completely contained within one less of these intervals.

If you'd want to compare the difficulties of such streams I suggest you make a map at least 30 seconds long where you just repeat those streams with a short break inbetween. Sadly there could still be aliasing artifacts when the BPM and the time interval share common divisors, but they shouldn't be as bad.

The sudden knack at 300 BPM is by the way currently caused by a hard cap avoiding singularities in maps putting some hitobjects 1ms or even 0ms apart. I should probably make that higher now that some people actually start to pull off 300bpm scores.
Thanks for the explanation, makes perfect sense now
Woobowiz

Riince wrote:

also because most people don't have high acc.
This, the reward curve for HR is incredibly steep so the only way it rewards you is if you get 99+% acc or if you're playing a 3+ minute map.
Nyxa
Even 6 minute maps give little under 98.5% unless it's a map that's hard to FC.
ZenithPhantasm
I feel like acc pp scales too exponentially. It should be more gradual imo 8-)
And I feel like Hidden aim pp should be 0 because its basically preference.
Endaris

ZenithPhantasm wrote:

And I feel like Hidden aim pp should be 0 because its basically preference.
Well, no.
While it's true that maps can become easier to read with Hidden it's still a lot harder to maintain combo. Hidden is clearly my mostplayed mod and the amount of shitmisses I get with hidden compared to nomod on maps i can play very comfortably(99%+ nomod) is immense. I think the difference could get smaller once I get better but I don't think you'll ever be equally consistent when using hidden.
Nyxa
Playing with HD is like synchronised swimming. Either you love the shit out of it or it absolutely disgusts you. I'm in the latter group.
ZenithPhantasm
I still feel like Hidden gives too much bonus pp.
B1rd
It only gives too much pp at high AR. ar9 is where hidden give the proper amount of pp. And yes it does make things harder to read.
ZenithPhantasm

B1rd wrote:

It only gives too much pp at high AR. ar9 is where hidden give the proper amount of pp. And yes it does make things harder to read.
Whether it makes things harder is somewhat subjective. Ik people who cant go without slapping on HD mod.
B1rd
It's not very subjective. It makes the object disappear before you have to hit it, so you have to memorise the location, and it makes it much harder to see stacked objects. This isn't subjective. There are very few people who can read HD better at AR9 who have not almost solely played HD and are not used to nomod.
Endaris

ZenithPhantasm wrote:

B1rd wrote:

It only gives too much pp at high AR. ar9 is where hidden give the proper amount of pp. And yes it does make things harder to read.
Whether it makes things harder is somewhat subjective. Ik people who cant go without slapping on HD mod.
You should know that this is an illness. They do it because they got so used to it that they started sucking when playing with approach circles. It's a similar phenomen to the "cant-play-below AR9.67"-people.
Purple

Endaris wrote:

You should know that this is an illness.

An illness like pokerus maybe AKA the opposite of a problem
Barusamikosu
Not being able to play maps without HD is definitely a problem. Kind of like people who can't read maps without HR. ;)

Nomod is the most basic form of osu gameplay. Anybody who can't play it has a huge gap in their skillset that needs to be addressed.
Also, from my personal experience--multiplayer and friend rankings (which I consider more representative of the 'average' osu player)--HD players tend to stick to maps that are easy with HD. All it takes is a bit of note density or complexity to ruin someone's sightread.
show more
Please sign in to reply.

New reply