Full Tablet wrote:

You lost pp because you replaced your records with worse ones. The score rank you get in each map doesn't matter.
Your maximum combo is the value that matters. One combo of 300 is worth more than 2 combos of 280.

So it was as I thought, that's kinda annoying, but thanks for the clarification.

[AirCoN] wrote:

@Tom: The system should only take into account your best performance on each map, not the best score.

As mentioned like a billion times already ( ) it's not possible to do this at the moment. I wish it was.

I just dont understand why my rank rose before the patch to 41k rank from 53k then after the patch my rank was risen to like 75k wat is this -_-

Kreto wrote:

I just dont understand why my rank rose before the patch to 41k rank from 53k then after the patch my rank was risen to like 75k wat is this -_-

Play Insanes then you understand ...

At your Playcount I was Rank 110k

And ye recalculating ...

Kreto wrote:

I just dont understand why my rank rose before the patch to 41k rank from 53k then after the patch my rank was risen to like 75k wat is this -_-

It was probably recalculating something. Judging by your Performance Graph you were actually at 75K the entire time.

Ekaru wrote:

Kreto wrote:

I just dont understand why my rank rose before the patch to 41k rank from 53k then after the patch my rank was risen to like 75k wat is this -_-

It was probably recalculating something. Judging by your Performance Graph you were actually at 75K the entire time.

i meant before patch V2 came out

Ultrayano wrote:

Kreto wrote:

I just dont understand why my rank rose before the patch to 41k rank from 53k then after the patch my rank was risen to like 75k wat is this -_-

Play Insanes then you understand ...

At your Playcount I was Rank 110k

And ye recalculating ...

my play count yea i know but i was higher rank and stuff just watneed to know why thx for the reply Insanes i do play some but not all plus i only use mouse no keyboard no tablet. Well since ur really good ^-^ gL in your ranking nice uh signature is it ^-^

Kreto wrote:

my play count yea i know but i was higher rank and stuff just watneed to know why thx for the reply Insanes i do play some but not all plus i only use mouse no keyboard no tablet. Well since ur really good ^-^ gL in your ranking nice uh signature is it ^-^

Only Mouse ? o.o Good Luck in your osu! career

And ~thanks~

Ultrayano wrote:

Kreto wrote:

my play count yea i know but i was higher rank and stuff just watneed to know why thx for the reply Insanes i do play some but not all plus i only use mouse no keyboard no tablet. Well since ur really good ^-^ gL in your ranking nice uh signature is it ^-^

Only Mouse ? o.o Good Luck in your osu! career

And ~thanks~

There are mouse only players better than you and I.

Myke B wrote:

There are mouse only players better than you and I.

I know (: Players like ChangE

Ultrayano wrote:

Kreto wrote:

Only Mouse ? o.o Good Luck in your osu! career

And ~thanks~

Mouse is more fun.

rexcannon_iii wrote:

Mouse is more fun.

"Fun" is subjective.

Stay on-topic, please. This thread is for pp feedback.

What kind of modifications are you still looking to do with the ranking system? assuming there are any.

Tom94 wrote:

Stay on-topic, please. This thread is for pp feedback.

Ok then, I'll ask again.

When are you going to implement complex finger work into the pp calculation?

rexcannon_iii wrote:

Tom94 wrote:

Stay on-topic, please. This thread is for pp feedback.

Ok then, I'll ask again.

When are you going to implement complex finger work into the pp calculation?

I can't answer all those "when" questions. Programming and especially coming up with ideas for an algorithm is not something that can be nailed down to Tuesday 14:00.

Myke B wrote:

What kind of modifications are you still looking to do with the ranking system? assuming there are any.

I believe I wrote quite a bit about that kind of stuff in the thread already. I hope I can get pattern difficulty inside, in addition to the obligatory tweaking of weightings of course.

Ultrayano wrote:

Myke B wrote:

There are mouse only players better than you and I.

I know (: Players like ChangE

i added u so u know:) and i will try my best with mouse XD ^-^

Mutual <3

These probably have been answered before somewhere in the thread, but I don't think I would be able to find it in 40+ pages of "mouse is more fun" etc.

-Do you ever intend to display player attributes on profiles or the performance ranking, like the aim/speed/accuracy player ratings shown on the tp rankings?

-What about the aim/speed difficulties of modded maps within the client? For example, I scroll through my maps and choose some map. Then I turn on DT EZ FL (or some obscure mod combination). From what I've heard, there are plans to display aim/speed difficulties of maps on their respective webpages, but covering all mod combinations on the webpage would require some significant modifications. What I'm getting at is a numerical value displayed somewhere in song select (most likely in the top left where the length/# of objects, etc are shown) that changes as you switch mods, much like the Global Ranking (Selected Mods) tab.

-Is there anything planned in the difficulty calculations regarding difficult patterns such as tornadoes?

Thanks again for delivering the justice of skill ranking.

Tom94 wrote:

I believe I wrote quite a bit about that kind of stuff in the thread already.

You might consider updating the first post with answers to commonly asked questions, to hopefully decrease the number of already-answered questions.

Doge wrote:

These probably have been answered before somewhere in the thread, but I don't think I would be able to find it in 40+ pages of "mouse is more fun" etc.

-Do you ever intend to display player attributes on profiles or the performance ranking, like the aim/speed/accuracy player ratings shown on the tp rankings?

It'd be cool to do so, but the way the system works right now (calculating a single pp value per player) those can't be displayed seperately at the moment. It would be cool to have something like that, but the top priority is at least making the system functional for all modes first and integrating it into star difficulty.

Doge wrote:

-What about the aim/speed difficulties of modded maps within the client? For example, I scroll through my maps and choose some map. Then I turn on DT EZ FL (or some obscure mod combination). From what I've heard, there are plans to display aim/speed difficulties of maps on their respective webpages, but covering all mod combinations on the webpage would require some significant modifications. What I'm getting at is a numerical value displayed somewhere in song select (most likely in the top left where the length/# of objects, etc are shown) that changes as you switch mods, much like the Global Ranking (Selected Mods) tab.

I hope I can make star rating change with selected mods. Displaying stuff like aim/speed seperately is probably not going to happen sooo soon, if it happens at all.

Doge wrote:

-Is there anything planned in the difficulty calculations regarding difficult patterns such as tornadoes?

Yup, but that's in the far future.

Doge wrote:

Thanks again for delivering the justice of skill ranking.

You're welcome.

This was probably asked already but I don't feel like reading 40 pages (you could maybe add the most asked questions to the 1st post) so here it goes:

I've heard that old beatmaps don't reward pp anymore, is this true? If so, are you pretending to remove its ranked score/accuracy eventually? Also, what's up with people saying that there's a big possibility of ranked score being removed from the game in the future? (since it doesn't show anymore in-game)

Are we close to the final version of this pp system or are there going to be any big changes incoming?

Last night I gained like 9 pp from a couple of James' old BeForU maps from 2008. That should answer your question regarding old beatmaps.

not sure if I would really consider it a problem, but I find it quite easy to just look at people around my rank and play the beatmaps they have in top ranks. typically they're pretty easy... It racks up some ridiculous amounts of pp though. @_@

Keeby wrote:

not sure if I would really consider it a problem, but I find it quite easy to just look at people around my rank and play the beatmaps they have in top ranks. typically they're pretty easy... It racks up some ridiculous amounts of pp though. @_@

This has always been how it is. If you can beat their top performances then you're better than them (or better than what skill level their top performances dictate).

Keeby wrote:

not sure if I would really consider it a problem, but I find it quite easy to just look at people around my rank and play the beatmaps they have in top ranks. typically they're pretty easy... It racks up some ridiculous amounts of pp though. @_@

That would mean that you actually deserve to be much higher than you actually are, and must've been playing easy maps the whole time.

haven't read the last 10 pages or so yet, but is the "difficulty" sorting option in the beatmap list the same as a list of "what songs give more PP"?

0 miss play on one of the songs on the front page with half-decent accuracy so i'm guessing not, but if it isn't what is it

Gray Pigeon wrote:

Tom, you should make ppv2 to the same calculation method as tp.

The map of score(50|50|50) is easier than the map of score(100|0|0).

While this is very true, I personnaly dislike the way the tp system handles this. Right now getting 100/100/100 on a map or getting 100/0/0 + 0/100/0 + 0/0/100 on three different maps gives exactly the same number of tps, when the former is much harder. I am fairly sure that sorting maps by stats this way is unavoidably flawed.

To avoid the previous problem I think each stat just needs tweaking. I thought of a simple solution - I'm sure there are better ones than mine, though. That involves giving bonuses or maluses depending on which stat is the highest/2nd highest/lowest.

Let's say it gives +75% to the highest stat, +25% to the second highest and -25% to the lowest when calculating the total pp rating (I omitted the digits after the decimal point). I'll illustrate that with a simple example (two scores giving the same amount of pps initially) :

90/85/80 -> 157/106/60 || 255pps -> 323pps (it's relative, so the increase does not matter)
140/70/45 -> 245/87/33 || 255pps -> 365pps (notice how it's now higher than the previous score)

I chose the percentages randomly, but you get the idea.

KaosFR wrote:

Let's say it gives +75% to the highest stat, +25% to the second highest and -25% to the lowest when calculating the total pp rating (I omitted the digits after the decimal point)

90/85/80 -> 157/106/60 || 255pps -> 323pps (it's relative, so the increase does not matter)
140/70/45 -> 245/87/33 || 255pps -> 365pps (notice how it's now higher than the previous score)

Tom94 wrote:

Currently the pp a score is worth is computed by the following formula: (aim^X + speed^X + acc^X)^(1/X)
Where X at the moment is 1.1 and will likely rise a bit in the future.

90/85/80 -> 230.789
140/70/45 -> 233.266
The value of X can be adjusted to change the worth of the highest stat.

Example: X=0.7 (The play gets "punished" for having a low stat).
90/85/80 -> 408.197
140/70/45 -> 395.121


Example: X=1.5 (The highest stat is given even more importance).
90/85/80 -> 176.909
140/70/45 -> 186.358

I see. I didn't know about the formula, thanks for pointing that out.

I'm not saying I like it that much though; even with X>>1 it doesn't seem to make much of a difference, mostly because of the last part of the formula (the ^(1/X)).

KaosFR wrote:

I see. I didn't know about the formula, thanks for pointing that out.

I'm not saying I like it that much though; even with X>>1 it doesn't seem to make much of a difference, mostly because of the last part of the formula (the ^(1/X)).

X>>1 indeed makes a huge difference. The ^(1/X) merely makes sure, that the result is not scaled horribly asymptotically to O(n^X), but remains in the same asymptotic scale.

KaosFR wrote:

I see. I didn't know about the formula, thanks for pointing that out.

I'm not saying I like it that much though; even with X>>1 it doesn't seem to make much of a difference, mostly because of the last part of the formula (the ^(1/X)).

The value X can be changed to adjust how much "extra" bonus or penalty there is when a stats is considerably better or worse than the rest:
(aim^X + speed^X + acc^X)^(1/X) * (3)^((X-1)/X)
(The (3)^((x-1)/x) part of the formula isn't really needed, since it is just a constant when the performance stats change, it is just there so the magnitudes remain more easily comparable when changing x)
So:
X =1
Then the formula is just the sum of the 3 stats (the arithmetic mean, multiplied by 3)
X -> Infinity
The formula comes close to the maximum stat, multiplied by 3. That way, if X is a big number, only the best stat matters.
X -> 0
The formula comes close to the geometric mean of the stats, multiplied by 3. The geometric mean gives smaller values if one of the stats is considerably lower
X = -1
The formula is equal to the harmonic mean, multiplied by 3. It gives relatively even smaller values if one of the stats is considerably lower (compared to the geometric mean).
X-> -Infinity
The formula comes close to the minimum stat, multiplied by 3. In that case, the worth of a score is determined only by it's worst stat.

Thanks for you answers. Somehow I still feel scores with one or two big stats don't get rewarded enough, have you thought of using something exponential instead of just polynomial ? This kind of formula :

logx(x^aim+x^speed+x^accuracy)

Still scaled down to O(n) but gives high stats a bigger impact (and still tends to the highest stat)

Full Tablet wrote:

KaosFR wrote:

I see. I didn't know about the formula, thanks for pointing that out.

I'm not saying I like it that much though; even with X>>1 it doesn't seem to make much of a difference, mostly because of the last part of the formula (the ^(1/X)).

The value X can be changed to adjust how much "extra" bonus or penalty there is when a stats is considerably better or worse than the rest:
(aim^X + speed^X + acc^X)^(1/X) * (3)^((X-1)/X)
(The (3)^((x-1)/x) part of the formula isn't really needed, since it is just a constant when the performance stats change, it is just there so the magnitudes remain more easily comparable when changing x)
So:
X =1
Then the formula is just the sum of the 3 stats (the arithmetic mean, multiplied by 3)
X -> Infinity
The formula comes close to the maximum stat, multiplied by 3. That way, if X is a big number, only the best stat matters.
X -> 0
The formula comes close to the geometric mean of the stats, multiplied by 3. The geometric mean gives smaller values if one of the stats is considerably lower
X = -1
The formula is equal to the harmonic mean, multiplied by 3. It gives relatively even smaller values if one of the stats is considerably lower (compared to the geometric mean).
X-> -Infinity
The formula comes close to the minimum stat, multiplied by 3. In that case, the worth of a score is determined only by it's worst stat.

For X -> Infinity the fomula comes close to the maximum stat, not multiplied by 3. Even with the other stats having the optimal weight, that is the same value as the highest stat, it would still converge to the highest stat. For instance (100, 100, 100). If we let X go to infinity we will have (abusing notation)

(100^inf * 3)^(1/inf) = 100 * 3^(1/inf) = 100

Since with (100, 0, 0) we get the same limit it is clear that everything lying inbetween also yield this limit.

Didn't give your other statements much thought since they're not very relevant in this context.

KaosFR wrote:

Thanks for you answers. Somehow I still feel scores with one or two big stats don't get rewarded enough, have you thought of using something exponential instead of just polynomial ? This kind of formula :

logx(x^aim+x^speed+x^accuracy)

Still scaled down to O(n) but gives high stats a bigger impact (and still tends to the highest stat)

Exponential scaling would not work out very well, firstly due to it only working with very small X, because the intermediate results would be come too big otherwise and rounding issues would arise, and secondly, because it'd rise too rapidly. For small aim, speed and acc the effect would be small while for big the effect would be ridiculous while with the polynomial method the effect always is the same.

(100, 50, 50) would yield half of (200, 100, 100) for any X using the polynomial method in contrast to what you proposed.

When the new algorithm is used to calculate star rating, will OD be somehow factored in? Since it plays a big part in determining the pp a score gives (from accuracy), but it isn't reflected in the osu!tp map difficulty, which is solely aim/speed.

Tom94 wrote:

For X -> Infinity the fomula comes close to the maximum stat, not multiplied by 3. Even with the other stats having the optimal weight, that is the same value as the highest stat, it would still converge to the highest stat. For instance (100, 100, 100). If we let X go to infinity we will have (abusing notation)

(100^inf * 3)^(1/inf) = 100 * 3^(1/inf) = 100

Since with (100, 0, 0) we get the same limit it is clear that everything lying in-between also yield this limit.

Didn't give your other statements much thought since they're not very relevant in this context.

I was referring to the formula if you multiplied it by (3)^((X-1)/X). While that factor isn't needed (since it's a constant when varying only the performance stats), it put it there so the magnitude of the values didn't change so much when changing X. With that, scores with equal amount of Acc-Speed-Aim don't change in magnitude, while, scores where one of the stats is better always get a bonus when increasing X.

Kytoxid wrote:

When the new algorithm is used to calculate star rating, will OD be somehow factored in? Since it plays a big part in determining the pp a score gives (from accuracy), but it isn't reflected in the osu!tp map difficulty, which is solely aim/speed.

I most likely will be in the future, but I can't promise it for the initial iteration.

Full Tablet wrote:

Tom94 wrote:

For X -> Infinity the fomula comes close to the maximum stat, not multiplied by 3. Even with the other stats having the optimal weight, that is the same value as the highest stat, it would still converge to the highest stat. For instance (100, 100, 100). If we let X go to infinity we will have (abusing notation)

(100^inf * 3)^(1/inf) = 100 * 3^(1/inf) = 100

Since with (100, 0, 0) we get the same limit it is clear that everything lying in-between also yield this limit.

Didn't give your other statements much thought since they're not very relevant in this context.

I was referring to the formula if you multiplied it by (3)^((X-1)/X). While that factor isn't needed (since it's a constant when varying only the performance stats), it put it there so the magnitude of the values didn't change so much when changing X. With that, scores with equal amount of Acc-Speed-Aim don't change in magnitude, while, scores where one of the stats is better always get a bonus when increasing X.

Oh yeah, I completely overlooked that. My bad. D:

I think I've read before that OD isn't weighted into the pp calculations. Is this true? Let's say I have an Insane SS, and I top that score with HR 97% accuracy, would this mean I would lose pp because my accuracy is lower? This would be imbalanced, since getting >95% accuracy with HR is much harder than getting an SS nomods, or with Hidden. Some Insanes I've SS'd but can not even FC with HR, and it feels like it's not even worth trying, because the 3-4% drop in accuracy would just make me lose pp for a much better performance. I like HR, but I don't try ranking with it purely because I end up losing pp.

Personally, I think HR should give more of a bonus than HD, since it is much harder to do a good performance with. I feel like HR is underrated (and was, even before ppv2) and considered to be about as difficult as HD. Play a map with DTHD, and then with DTHR and it should be obvious why I'm stating this. In my opinion, the order of mod difficulty is: DT > HR >>> HD, excluding FL since it requires an entirely different form of gameplay than the other three. However, to me it seems like pp looks at mods like: DT > HD > HR, which isn't balanced, since a 98% HR FC is much more impressive than an HD SS.

Also, sometimes I'm reluctant to try improving my accuracy on a map, since I've improved my accuracy by 5% on some maps and still lost pp. At first I thought it just happened because I dropped in ranks, not pp, but it keeps happening consistently. Now it feels like unless I FC something I will lose pp, regardless of my accuracy. Or maybe I'm wrong, but I would like to understand why this happens.

I like ppv2, I think the major issues are just some imbalances here and there and a lack of compact information on how it works. I understand that you can't work on the latter yet, though. My apologies if the things I've said have been answered before, but I've read about 75% of the thread and the only thing I've seen about HR was thelewa saying that HR is pretty much pointless below 99% accuracy, which sounds like a heavy imbalance to me.

-Scylla- wrote:

I think I've read before that OD isn't weighted into the pp calculations. Is this true? Let's say I have an Insane SS, and I top that score with HR 97% accuracy, would this mean I would lose pp because my accuracy is lower? This would be imbalanced, since getting >95% accuracy with HR is much harder than getting an SS nomods, or with Hidden. Some Insanes I've SS'd but can not even FC with HR, and it feels like it's not even worth trying, because the 3-4% drop in accuracy would just make me lose pp for a much better performance. I like HR, but I don't try ranking with it purely because I end up losing pp.

Personally, I think HR should give more of a bonus than HD, since it is much harder to do a good performance with. I feel like HR is underrated (and was, even before ppv2) and considered to be about as difficult as HD. Play a map with DTHD, and then with DTHR and it should be obvious why I'm stating this. In my opinion, the order of mod difficulty is: DT > HR >>> HD, excluding FL since it requires an entirely different form of gameplay than the other three. However, to me it seems like pp looks at mods like: DT > HD > HR, which isn't balanced, since a 98% HR FC is much more impressive than an HD SS.

Also, sometimes I'm reluctant to try improving my accuracy on a map, since I've improved my accuracy by 5% on some maps and still lost pp. At first I thought it just happened because I dropped in ranks, not pp, but it keeps happening consistently. Now it feels like unless I FC something I will lose pp, regardless of my accuracy. Or maybe I'm wrong, but I would like to understand why this happens.

I like ppv2, I think the major issues are just some imbalances here and there and a lack of compact information on how it works. I understand that you can't work on the latter yet, though. My apologies if the things I've said have been answered before, but I've read about 75% of the thread and the only thing I've seen about HR was thelewa saying that HR is pretty much pointless below 99% accuracy, which sounds like a heavy imbalance to me.

OD is factored in by a huge margin. And HR below 99% is not pointless. It is pointless for a top-tier player like thelewa, because he already has such good scores.
There will be a wiki article when pp got added for all modes.