forum

Performance Points feedback and suggestions (Taiko)

posted
Total Posts
645
show more
AnFace
i'm curious about how well it'll work for the really good players that don't have great scores/don't play often or whatever

actually i'd like to know how the current broken temporary system is but we're not allowed to see sadface
NoYzE

AnFace wrote:

i'm curious about how well it'll work for the really good players that don't have great scores
Ehh...
I guess the new pp system should scan the players' brain for skill that he didn't put into a high score? ;)
Topic Starter
Tom94

AnFace wrote:

i'm curious about how well it'll work for the really good players that don't have great scores/don't play often or whatever

actually i'd like to know how the current broken temporary system is but we're not allowed to see sadface
Players will still have to get great scores. The metric is called "performance points", not "potential points".
Also the current broken algorithm simply does the following:
  1. Multiply accuracy with star rating of the map (up to a maximum of 500)
  2. Divide result by 3 if converted
  3. Build the diminishing sum of the best scores for every player
That also makes it pretty obvious why it's not working well, heh? Still better than nothing - that's why it exists for now.
Sy[K]es
There should be a way to convert maps better than it does at the moment and still keep the combo score ratio the same.
Streams might be the biggest obstacle but I believe it is possible. :<
AnFace

NoYzE wrote:

AnFace wrote:

i'm curious about how well it'll work for the really good players that don't have great scores
Ehh...
I guess the new pp system should scan the players' brain for skill that he didn't put into a high score? ;)
ideally hell yea

i mean the players that don't specifically have a lot of scores but still have some that are really good blah blah you know what i mean

quality vs quantity how will it play out overall
Sy[K]es
I have got a question.
If x had 10K PP
and y had 3KPP
Let us assume that y got better score than x on the score board, does that affect the PP or just the rank position or none?
RaneFire

Sy[K]es wrote:

I have got a question.
If x had 10K PP
and y had 3KPP
Let us assume that y got better score than x on the score board, does that affect the PP or just the rank position or none?
You are still thinking inside the box, namely that of the old system.

ppv2 has a beatmap difficulty algorithm, meaning this is what needs to be worked out precisely to get accurate results. Your performance is a statistical analysis compared with a perfect performance on the map. Player ranks in comparison to each other on a single map should be irrelevant. That was the problem with the old system for many reasons.

I'm sure there will be a LOT of testing. When tp first came about, it wasn't anything like it is now, and Tom tweaked it a lot, even adding a new variable (aim). I'm not predicting it's initial failure for taiko, but there is only room for improvement upon the algorithm, even if the first results may be completely off.
Topic Starter
Tom94
I've got some initial difficulty lists to show off here:
https://docs.google.com/spreadsheet/ccc ... 0VGc#gid=1

This spreadsheet does not contain every single beatmap, only the ones I selected for testing. Please give me feedback and let me know which version of the algorithm works the best. ;)
Nashmun
Hum, for me it seems like the basic algorithm is the most accurate.

The others one still have some weird results like Freedom Dive's oni > Inner oni while it's the contrary or still a duckling > I'm your daddy Fatal oni which is the opposite aswell.

I still need to look at it more closely though.
BrokenArrow
Yeah, same goes for HIT AND RUN IN 1920 A.D. and NNRT (which should be placed a lot lower normally)
But from first look things don't seem to be very far off, well done so far
EBAWER123
With colour changes w/o converts:



sacrifice and chipscape taikosakis are easier than who's your daddy or akasha, also gothic system is much harder than taikosakis not just a little :?


This is a lengthy post but please bear with me~


Some of these are human some of these are inhuman and some of these are easy, I wanna see what your programm thinks of these maps :3c
Nashmun
Hey, could you include those maps into the difficulty rating ? They have lower bpm but they are actually pretty hard compared to regular 220+ maps, so it would give us a good indication on how you balanced technical vs speed :)

https://osu.ppy.sh/b/204931
https://www.dropbox.com/sh/2f9ibeu4yk8e ... nsekai.osz
https://www.dropbox.com/sh/2f9ibeu4yk8e ... e%2008.osz
https://www.dropbox.com/sh/k22ut3jyjn9j ... 20Ruby.osz
AnFace

Nashmun wrote:

Hey, could you include those maps into the difficulty rating ? They have lower bpm but they are actually pretty hard compared to regular 220+ maps, so it would give us a good indication on how you balanced technical vs speed :)
most of the maps listed i've never played because of wtfbpm and me being slowface so i'd like to know this

otherwise goodbye to my rank i'll see you all in hell
XK2238

Nashmun wrote:

https://osu.ppy.sh/b/204931
THIS MAP.
dafuq
Topic Starter
Tom94
Added another algorithm version, considering rhythm changes, to the end. Will add more versions when I get back home.
Here the link again: https://docs.google.com/spreadsheet/ccc ... 0VGc#gid=1
Luna
I think the pattern difficulty needs to be improved quite a bit - right now chipscape and Broken the Moon have the same rating (45). I could FC chipscape if I actually tried (and I'm really not very good at high BPM), but I'm not even able to pass Broken the Moon.
Even with problems like that, it looks SO much better than what we're used to having, so good job.
TimmyAkmed
I know your algorithm version isn't over but, i'm not sure we talked about "hitting big notes" in the difficulty of the maps. When the bpm is high it becomes harder to hit the big notes even if you alternate. The big notes density inside the map should also be considered as a + in the difficulty. (more the bpm is high and more the big notes density is high more difficult the map is) When there are big notes you are supposed to hit them. And even if you try to avoid them they are a pain for you to read correctly the next part of the map. Imo it's important cause I think it's one of the reasons that makes "Unpleasant Sonata " so hard to FC.
lolcubes
While it's understandable that everyone is focusing more on how the converts get converted and how the hard maps get valued, but let's talk about lower maps a little.

We have *** Passionate (TV Size) which is valued the same as Dragons. I FCd Passionate with hard rock where I barely get an A on Dragons, with a low combo. Dragons is so much harder because of really odd rhythms and because it's kinda long compared to the TV Size thing.

Then you have Hades in the Heaven which is valued slightly above but then you have a couple of much easier maps, namely Night of Fire by lepidon which is a lot higher above. Certain pattern chain in the middle of the map make Hades really really hard to FC, while Night of Fire has quite simple patterns and is something I consider a beginner's map. Thank you for playing isn't that difficult either and it's even further above.

Also in the mid class you have Intersect Thunderbolt 7491's Oni which is below Ernst Oni. I highly disagree with this, despite Ernst being a bit more pattern heavy. The patterns in Ernst aren't difficult, it's just slightly a more dense map, but in Intersect Thunderbolt you have high speed pattern changes, namely a quite hard 1/3 stream to perform correctly due to it's bpm which contains multiple hand swaps.

Just some thoughts.
EBAWER123

TimmyAkmed wrote:

I know your algorithm version isn't over but, i'm not sure we talked about "hitting big notes" in the difficulty of the maps. When the bpm is high it becomes harder to hit the big notes even if you alternate. The big notes density inside the map should also be considered as a + in the difficulty. (more the bpm is high and more the big notes density is high more difficult the map is) When there are big notes you are supposed to hit them. And even if you try to avoid them they are a pain for you to read correctly the next part of the map. Imo it's important cause I think it's one of the reasons that makes "Unpleasant Sonata " so hard to FC.

No, score is enough of a measure. Consider big notes as a bonus points not increase in diff.



I like this, it's really close to a real difficulty. Though 51 is harder than chipscape.
Nashmun
I gave you the wrong version of one of the maps I suggested, here is the interesting one :

http://puu.sh/72lL2.zip

@lolcubes, you are right, but focusing on high-tier maps (not including Firces forces and cie though) is good to have a global overview of how each aspect are weighted, which is more difficult to see on standards onis and uber-insane ones.
TimmyAkmed

EBAWER123 wrote:

No, score is enough of a measure. Consider big notes as a bonus points not increase in diff.
I still think the same, when a mapper put big notes in his map he wants us to hit them like big notes. Being able to FC maps like Distorted Lovesong Taikocalypse DX without hitting the big notes doesn't mean to me that you are able to play the map the way it has been thought for (sorry i'm not sure everybody will know what I mean here) If we avoid to hit the big notes it's because it's obviously harder.

And for example even if it's only on Unrankable stuff i'm sure almost everybody who'll see big notes in patterns will have trouble to read it "normally"
Topic Starter
Tom94

TimmyAkmed wrote:

EBAWER123 wrote:

No, score is enough of a measure. Consider big notes as a bonus points not increase in diff.
I still think the same, when a mapper put big notes in his map he wants us to hit them like big notes. Being able to FC maps like Distorted Lovesong Taikocalypse DX without hitting the big notes doesn't mean to me that you are able to play the map the way it has been thought for (sorry i'm not sure everybody will know what I mean here) If we avoid to hit the big notes it's because it's obviously harder.

And for example even if it's only on Unrankable stuff i'm sure almost everybody who'll see big notes in patterns will have trouble to read it "normally"
Since the bonus in score is so small and there are no disadvantages over normal notes when hitting them like normal notes I don't think they should coult as "difficulty". If they'd break combo or reduce accuracy if hit wrong, then I'd agree with you more.

But even if I did agree, it'd be impossible to check scores for having hit big notes with 2 buttons, which in turn makes it impossible to have sensible pp calculations if big notes would be factored in as difficulty.
AnFace

Tom94 wrote:

Since the bonus in score is so small
missing a big note is like getting 2x100. i wouldn't call that small
Topic Starter
Tom94

AnFace wrote:

Tom94 wrote:

Since the bonus in score is so small
missing a big note is like getting 2x100. i wouldn't call that small
Relative to a map with maybe 2,000 hit circles it is. But yeah, depending on how many hit circles there are and how many big notes there are the bonus can also be big, I stand corrected.

The rest of the argument still stands, though.
karterfreak
Looking at the most recent one, I'd say its pretty accurate towards the top. Mad Machine being at the top is expected if we're going to weigh converts equally due to the stupidly high speed of the converted streams, and the taiko maps themselves are ordered pretty well. Some of the maps that I'd consider to be harder have harder finisher patterns in them, which aren't weighted by the system now from what I read.

It might be a good idea to come up with some way of weighting finishers because honestly they are a pretty big deal in how maps are read. As an example, The first kiai for Loctav's Oni in this mapset in its current state is absolutely insane to read the finisher section towards the end (and some of it is unrankable, but I digress), and is notably harder than the same pattern without finishers to read.
Topic Starter
Tom94

Tasha wrote:

Looking at the most recent one, I'd say its pretty accurate towards the top. Mad Machine being at the top is expected if we're going to weigh converts equally due to the stupidly high speed of the converted streams, and the taiko maps themselves are ordered pretty well. Some of the maps that I'd consider to be harder have harder finisher patterns in them, which aren't weighted by the system now from what I read.

It might be a good idea to come up with some way of weighting finishers because honestly they are a pretty big deal in how maps are read. As an example, Loctav's Oni for this map in its current state is absolutely insane to read the finisher section towards the end (and some of it is unrankable, but i digress), and is notably harder than the same pattern without finishers to read.
I'd love to give finishers a proper bonus, but since it's impossible to check whether a player hit the finishers or not this would greatly overvalue maps with finishers when applied to pp for people who did not play them correctly.
Also glad to hear, that the recent version looks good.

I'll check the other feedback regarding specific maps when I find time and will try to work around it. :)
karterfreak
What about just applying a bonus for finishers that are in denser parts of the map? Even if the finishers aren't hit, its still technically harder to read them. My previous example (while exaggerated) shows that quite well imo.
Topic Starter
Tom94

Tasha wrote:

What about just applying a bonus for finishers that are in denser parts of the map? Even if the finishers aren't hit, its still technically harder to read them. My previous example (while exaggerated) shows that quite well imo.
As said in the private discussion I just had with you: Finishers can be skinned small, so they don't make reading more challenging. :/
Nashmun
As Tom said, we can't tell if a player hit the big note with both keys or as a regular note. If we give a bonus to a map due to its finishers, some people will just play them as if they don't exist and earn a lot of pp, while others players will try to hit them and fail.

Yes it does affect the reading, but if you play them like regular notes, you can make abstraction of the finisher pretty easily (Also some players can change their skin too)
karterfreak
Already discussed it with Tom in private, and unfortunately there really is no way to work finishers into the algorithm.
lolcubes
It could be possible if the condition is a full combo though. There is extra score after all.
Could have small calculation issues with sliders though, but they give only a certain amount of score so...

Out of curiosity, how are sliders calculated then? You don't have to hit a slider to full combo, but it should show skill too, there are some maps where sliders become absolutely brutal if you have DT activated. Unreal BPMs and all (above 300).
Yuzeyun
They become brutal above 200 1/4 DT @cubes.
And past 250 you come to 1/2 ticks :U
Luna
120 BPM maps get 240 BPM drumrolls, which is increased to 360 BPM 1/4 by DT
Just an example
I don't really see a way to incorporate big notes/drumrolls/spinners into a difficulty algorithm without per-hitobject data on scores tho (unless it's an SS)
lolcubes

_Gezo_ wrote:

They become brutal above 200 1/4 DT @cubes.
And past 250 you come to 1/2 ticks :U
Yes, that's what I said. Any BPM above 200 with DT applied make them over 300 BPM real time.
Luna
Tom, could you possibly add some DT maps to the difficulty list so we can see how those are rated?
Stuff like this maybe:
http://osu.ppy.sh/b/132889&m=1
http://osu.ppy.sh/b/214251&m=1
http://osu.ppy.sh/b/267841&m=1
http://osu.ppy.sh/b/153886&m=1
http://osu.ppy.sh/b/161275&m=1

All of those are hard DT maps, some moreso than others. I'd just like to see what your program thinks of them.

Thanks!
Topic Starter
Tom94
:!:

lolcubes wrote:

While it's understandable that everyone is focusing more on how the converts get converted and how the hard maps get valued, but let's talk about lower maps a little.

We have *** Passionate (TV Size) which is valued the same as Dragons. I FCd Passionate with hard rock where I barely get an A on Dragons, with a low combo. Dragons is so much harder because of really odd rhythms and because it's kinda long compared to the TV Size thing.

Then you have Hades in the Heaven which is valued slightly above but then you have a couple of much easier maps, namely Night of Fire by lepidon which is a lot higher above. Certain pattern chain in the middle of the map make Hades really really hard to FC, while Night of Fire has quite simple patterns and is something I consider a beginner's map. Thank you for playing isn't that difficult either and it's even further above.

Also in the mid class you have Intersect Thunderbolt 7491's Oni which is below Ernst Oni. I highly disagree with this, despite Ernst being a bit more pattern heavy. The patterns in Ernst aren't difficult, it's just slightly a more dense map, but in Intersect Thunderbolt you have high speed pattern changes, namely a quite hard 1/3 stream to perform correctly due to it's bpm which contains multiple hand swaps.

Just some thoughts.
The first issue you mentioned seems to be well fixed in the newest list. Regarding hades in the heaven, the difference between the maps you called and this one seems to have shrunk, but there still is quite a gap.
Pattern difficulty can only be considered to a certain extent due to subjectivity and due to the fact, that for the algorithm there is no way to find out about "1/3" or "1/4". All there is are relative timings, so I can't promise much improvement in that direction without ruining weightings for other maps.

The thunderbolt vs ernst thing is very similar, you even called the 1/3 out... there's not much the algorithm can do in that regard. Also 1/3 can be trained just as easily as 1/4 - the only reason many people are bad at them is because they are so uncommon. :P


Luna wrote:

Tom, could you possibly add some DT maps to the difficulty list so we can see how those are rated?
Stuff like this maybe:
http://osu.ppy.sh/b/132889&m=1
http://osu.ppy.sh/b/214251&m=1
http://osu.ppy.sh/b/267841&m=1
http://osu.ppy.sh/b/153886&m=1
http://osu.ppy.sh/b/161275&m=1

All of those are hard DT maps, some moreso than others. I'd just like to see what your program thinks of them.

Thanks!
Can't selectively apply mods for some maps for now. That'd require some very unnecessary program loogic for the final purpose if weighting pp and (imho) is a waste of time to implement. If the algorithm works well with the current maps you throw at it, then it should work well with pretty much any map and thus also those DT maps. All it does is create a new map with 1.5 the speed anyway.


lolcubes wrote:

It could be possible if the condition is a full combo though. There is extra score after all.
Could have small calculation issues with sliders though, but they give only a certain amount of score so...

Out of curiosity, how are sliders calculated then? You don't have to hit a slider to full combo, but it should show skill too, there are some maps where sliders become absolutely brutal if you have DT activated. Unreal BPMs and all (above 300).
Sliders can't really be factored in for the same reasons as big notes. There might be one way to do it, though, which is to consider score relative to the maximum possible score instead of things like 300, 100, 50 and miss counts.

Do you guys think the taiko scoring system is good enough to be used as a basis for pp?
Nashmun
From what we are sawing, I think we can give it a try so that we can see how it goes for every maps instead of a limited list and then adjust the algorithm little by little
Topic Starter
Tom94

Nashmun wrote:

From what we are sawing, I think we can give it a try so that we can see how it goes for every maps instead of a limited list and then adjust the algorithm little by little
That means implementing pp with it. Can't go and post gigantic lists with 30,000+ entries. :P
But yeah, things are going forward.
Nashmun
Yeah I know, just saying that implementing it with pp would help us understand more how it weight every maps and then adjust the algorithm to be even more accurate :)
AnFace

Nashmun wrote:

Yeah I know, just saying that implementing it with pp would help us understand more how it weight every maps and then adjust the algorithm to be even more accurate :)
show more
Please sign in to reply.

New reply