1. osu! forums
  2. osu!
  3. Gameplay & Rankings
  4. osu!mania
show more
posted
Have been talking to other people about scorev2 and its issues. The main three issues that people have said are:
- Rainbows are weighted too little.
- HR on v2 is ridiculously difficult to get rainbows on (on Stable at least, on Cutting Edge it seems as easy as accumulating rainbows on nomod - but more on that later).
- LNs generate a lot of misses if they have really tricky releases. They work fine for other types of LNs.

All of those problems can definitely be amended.

Rainbow Accuracy

Shoegazer wrote:

320s are very much underweighted because the only component of the scoring system that takes into account 320 accuracy is the combo component, which only has a 20% prominence. Add on to the fact that the difference between a 300 and 320 is so small and that the absolute difference between juan and Hudo's 320 count isn't that significant, it would make sense that 320s are really underweighted at the moment.

You could mitigate this by including 300gs into accuracy, but from what I've experimented it might create too much emphasis on MAX accuracy with charts that players have issues getting 96%+ on (and as a result would not be an accurate assessment of skill).

Alternatively, you can avoid including MAXes in the accuracy component and just increase the importance of MAXes to like 360 to increase the emphasis of it by a noticeable but not overpowering amount in the combo component, but that requires a bit more experimentation.
I initially wanted to increase the rainbow judgement weightage without embedding rainbows into accuracy, but no matter how much I changed it, the difference is very minor (~600-1,200 points) and a 200 will almost always be too powerful compared to a rainbow 300. So I scrapped that idea and thought that embedding rainbows into accuracy with a reasonable weightage and maybe making the curve more lenient would be the best idea.

I've been experimenting with weightages and discussing with people about how much a 200 should be worth compared to a 300. I initially thought that 310 would be fine (and a 200 would be worth 11 300s), but when it came to matches like this, if accuracy was the only factor, Argentina would win by 21,000 points. I do think that Argentina should win and it's a step in the right direction, but 21,000 seems extremely overwhelming since it undermines the fact that Poland had overall, noticeably less 200s. I tried it with harder charts too and they seem to favour rainbow accuracy a little too much for my liking - especially since when it comes to harder charts (where players struggle with), good rainbow accuracy is usually caused by variance rather than a higher skill level. 200s and worse judgements should determine performance for that.

I wanted to use 307 afterwards, but it still gave a bit too much emphasis for my liking, about 12,500 points for that Argentina/Poland match. I went down to 305, and the difference is about 6,800. I think that's ultimately the most reasonable assessment, and others I've talked to seem to agree with the prenotion that a 200 is about 21 normal 300s. Ignoring the bad judgements (since those values are pretty much set in stone at this point), this is probably (part of) the ideal solution. This does mean that only full rainbow scores are SSs, but I don't see that as a problem as frames of reference can be shifted.

Getting rid of the difference between a rainbow and a normal 300 in the combo scoring component is probably ideal too, since that should be in the accuracy component, not the combo component. If rainbows are included into accuracy, the combo component does not need a rainbow component.

I also wanted to soften the exponential curve a tiny bit when it comes to including rainbows, mainly because at a certain point extremely good accuracy is more caused by variance rather than a very high skill level - unless the performance is consistently done, which is not measurable with just one match and one attempt. The exponential I had in mind was Accuracy^(2 + 2 * Accuracy), but it's essentially Accuracy^4 - so 1 power down.

Similar note, wanted to respond to this:

Drojoke wrote:

To give a slightly over the top example: scoring a 200 on every note in a song is going to give you an accuracy of 66.67% and a score of about ~335k. The score is really low because you barely get any bonus score. The same play would net you a somewhat respectable 733.3k in the current proposal. Add FL to this, and you get a score of 777.3k. In practice, it's going to be less extreme than this, but it's definitely present.
A 66.67% score nets you about 306K (181K for accuracy, 125K for combo). Adding FL increases it to 324K, you might've miscalculated. In any case, I do agree with the fact that bad judgements (non-200/300 judgements) should be penalised more, but I don't think it's necessarily what they have in mind at the moment, since the values are carved in stone. MAX judgements are not.

tl;dr: Embed rainbows into accuracy with a weightage of 305 instead of 320, change the accuracy curve to Accuracy^(2 + Accuracy * 2), remove the differentiation between rainbows and normal 300s in the combo component (both of them should have a HitValue of 30).


HardRock
Accumulating rainbows on HR on most charts is really strict already, but since the rainbow window is stricter in scorev2 in ODs beyond 8, it gets even stricter and probably way too difficult. On anything above OD7.5, the HR will be boosted to 10 - which means that it has a rainbow window of +-13ms. Add on the general effects of HR (which makes windows 40% tighter), it gets knocked down to 9ms, truncated. Considering that any hits within +-5ms is caused by computer performance variance, having only 4ms of "controlled" timing is very very low - especially since the difference between an 6ms controlled window (HRv1 window) and a 4ms controlled window is huge. While it is true that certain modes have windows this tight on HR (and maybe DT), 85% of charts used in MWC are OD8, whereas it's much less common in other game modes to have something this tight (in Standard it's some absurd DTHR with a decent OD, and in Taiko it's some absurdly high OD with DTHR, and I'd argue that timing on both games are easier than Mania).

There's also the fact that with the rainbow weightage aspect included, wins on FreeMod are very variance-based rather than performance-based.

My main suggestion is to keep the 40% tighter window (except for early miss judgements) on HR, but not increase OD with HR. 11ms (OD8-OD9 is standard), while I still think is noticeably harsh, is the norm and the 40% seems to fit the other judgement windows quite well anyway. In fact, v1 already does this, this is a OD5 chart on HR. While it appears as if it looks like OD7+HR, is actually aligned with OD5 with 40% tighter windows. Ignore the additional 0.5ms, as it's some byproduct of woc's janky coding.

But that's one thing. The other thing I noticed is that the difference between rainbow difficulty between scorev2 NoMod and scorev2 HR on on Cutting Edge is insignificant. I've experimented this with juan, and his performance on NoMod and HR is similar, variance included. Pictures included:



A 25x300 difference for something that should be a 6ms gap (15ms - 9ms) is absurd, and is most definitely not caused by variance in performance.


I told juan to play on the Stable build as well, and then he noticed a massive difference in accuracy with scorev2 HR. He can usually get a 6:1 to 7:1 ratio on scorev2 HR in CE (and is comparable to his nomod scores consistently), but he can barely break 2.5:1 on Stable. He didn't seem to mention any performance issues either. Here are pics of his scores on the Stable build:



Note that these two charts are also comparable in difficulty as well, both would've been in Group Stages last year (and Sakura Mirage was in last year).

I know you mentioned a HR rainbow fix earlier in the thread, but I'm not sure if it did what it was supposed to do. Seems like it created issues rather than fixed a problem. underjoy's ratio on HR before the fix made sense to me given the relatively low OD and all, my main concern was with the miss count (which I'll talk about in the next section).

tl;dr: Don't increase OD when HR is switched on but keep the stricter timing windows, Accumulating rainbows on OD10+HR is way too strict because the window where a player can control is really really small and it affects an overwhelming majority of charts in MWC. There might be a problem in Cutting Edge where HR currently is as easy as NoMod.


LNs
LNs in v1 seem to be bugged - making them much easier than they should be. While I understand how the LN mechanics work, it doesn't seem to work that way for LNs that you don't let go but you hit the LN head perfectly. For some reason, no matter when you let go of the note, as long as you hit the head perfectly, you will get a 200. If you hit the head a bit earlier or later, you get a 100 instead. Here is video demonstration of this. This is probably (partially) why LNs in v1 are so easy compared to ones in v2 - particularly ones with very tricky releases. Players subconsciously don't let go of LNs properly and they don't get punished for it. In v2, the punishment becomes noticeable. In v2, players get a miss if they don't let go.

Getting rid of this bug is definitely a good start, but since scorev2 is implemented at such short notice (85% of participants probably haven't used scorev2 yet, though you can argue that it's their own fault) and MWC is used as testing grounds, you'd want to make LN releases more lenient than they currently are at the moment for easier transitioning - as players are getting a lot of misses already, even on NoMod. I think a LN leniency of 1.8x would be fine, but this is a bit of an arbitrary figure. I don't really know the effects of this because I don't play LN charts that much, and you're better off asking players like juankristal or _underjoy instead.

tl;dr: Increase LN leniency to 1.8x for easier transitioning, as LNs in v1 contain a bug that makes LN releases much much easier than they should've been in the first place.


I think that should be all, feel free to ask any questions if you're uncertain about a couple of things that I've pointed out/suggested.
posted
Gameplay: [smoogipooo] Make osu!mania ScoreV2 early miss window unchanged by HR.
Such a small detail, but very effective in punishing spamming. I'm digging this!
posted
Hello,
I decided to do some testing of LN maps on v2 system. As you know, there was previously an issue that caused LN maps on HardRock to be barely playable due to the excessive amount of misses made on LN releases. However, today I have observed that this problem is also really visible on v2 nomod play. We've tested some maps, here is the mp link and some screenshots from our play that show the problem very well:

https://osu.ppy.sh/mp/26731666
https://osu.ppy.sh/ss/5743076
http://puu.sh/qlPHd/c3d92da6d5.png

As you can see, there is one thing those screenshots have in common: abnormal amount of misses and negligible amount of 50 and 100 judgments. As a pretty decent LN player I can easily notice when I'm hitting a LN as I should and when I'm releasing it too early/too late. My observations are that for releasing a LN on normal terms you can get only MAX, 300, 200 and miss. Apparently there is a bug that doesn't allow the other judgements to appear and converts them into misses.

This is very visible on the Funny Shuffle map, which has really bizarre LN endings. If the system worked correctly, I'd get a lot of 50s and 100s for uncorrect releases - however, I got only one 50. How? It seems that while I cannot see a way to get a 100 from releasing a long note, there is a possibility (and only possibility it seems) to get a 50, if you release the LN much too early and repress it.

Anemone LN is another difficult LN map with unintuitive long note patterning, and releases don't follow the music. I struggled with the map, not only because it's quite hard, but mostly because the lack of my precision on releasing gave me a high number of misses which caused me to fail really quickly as well as tank the score and be unable to recover. The amounts of 50 and 100 also look very suspicious. In this map I also tried to repress the LNs as said above, and I got 50s. This leads to an interesting conclusion: it's more beneficial to repress a LN, than to hold it too long (to avoid a miss).

About maps like Inside DT, where the numbers look okay: 100 and 50 judgments are caused by wrong note presses, not by releases. I still felt got two seemingly random misses near the start of the map. At the ending fast jacks combined with noodles made me obtain bad judgments.

It's really easy to observe that LN releases are at the moment not working as intended, and while this might not be so blatant in early stages of the MWC, it will surely cause a lot of trouble in higher-end maps.
posted
Thanks to _underjoy and a few others that notified me of this - yes LN release timings weren't accounting for the 1.5x lenience. They have been fixed now. Please test again!

Holding LNs also gradually restores HP again now.
posted
ok this is too funny not to comment on:

was playing a very long song, too hard, so i stopped playing. watched as my score dropped for a while

2 minutes later into the song, and I haven't hit any more notes, and my score is going up now
posted
I've checked to make sure what was causing this, and it is intended. The reasoning is simple - towards the lower end of the accuracy model, there is less of a difference between 30% accuracy and 20% accuracy (as an example), but this is not the only factor that defines the accuracy portion of your score. To keep things looking progressive it also takes into account how much of the map has been completed, so for very short maps this value increases much more drastically than your accuracy decreases at those lower accuracies.

To simplify it, no, the calculations are not incorrect and you aren't getting "more score by missing", it's just your accuracy being recalculated to consider how much of the map has been completed.
posted
The fix that fixed LN misses does not seem to have made its way to the beta build. A fix before the tournament would be much appreciated
posted
The fix will propagate all the way to Stable (latest) by Friday in time for MWC. Please use Cutting Edge until then.

Edit: It's been propagated back to Beta now.
posted

smoogipooo wrote:

The fix will propagate all the way to Stable (latest) by Friday in time for MWC. Please use Cutting Edge until then.

Edit: It's been propagated back to Beta now.
Thank you. The main issue was that it was only available for supporters
posted
Hello, so i noticed a super major issue on scorev2 related to LNS (I'm using beta btw)

Even with this new update about the ''hold gives hp'' this is mere nothing, like i didnt feel nothing.

Its hard to talk about this, but like on score v1, a map with HP 9, if you KEEP missing and hitting your hp will be like dropping a little and recovering FAST.

On scorev2 a map on HP 8.5, if you keep missing and hitting or even missing a little and start combo'ing the DRAIN is pretty big and the recovery is super small.
https://osu.ppy.sh/ss/5756787 - score v2
https://osu.ppy.sh/ss/5756802 - score v1
i even stopped played for like 2secs on v1 and still got my hp to recover in like 1~2secs, on v2 if i rush any ln pattern my hp will be drained so much and the recovery is super small i cant even reach max hp, score v1 replay ~ http://puu.sh/qo0SF/86fb41302b.osr (score v2 replay is lost, my osu! didnt saved idk why)

so please take a look again on LN HP Drain, and LN HP Recovery.
Thanks

i dont consider me a good LN player, but all players who knows me always tell me i'm a good LN player, so well you should at least read my opnion.
here some scores that maybe be a proof about my ln skills
- https://osu.ppy.sh/ss/5756872
- https://osu.ppy.sh/ss/5756878
- http://osu.ppy.sh/ss/5756892
- https://osu.ppy.sh/ss/5756897
- https://osu.ppy.sh/ss/5756900
posted
@FelipeLink: Have you tried making a simple map of few notes and testing? I have and the HP looks almost exactly the same. The difference is that LNs have two judgements now instead of one, so the punishment is doubled for missing a complete LN. It is also slightly harder to achieve perfect scoring on LNs due to both the reduced leniency of LNs and that they have two judgements.

All according to plan in my opinion?
posted

smoogipooo wrote:

@FelipeLink: Have you tried making a simple map of few notes and testing? I have and the HP looks almost exactly the same. The difference is that LNs have two judgements now instead of one, so the punishment is doubled for missing a complete LN. It is also slightly harder to achieve perfect scoring on LNs due to both the reduced leniency of LNs and that they have two judgements.

All according to plan in my opinion?
Pretty sure he might tested that in stable while the change was only in cutting edge implemented. Not sure tho
posted
Nah he specifically stated beta.
posted

smoogipooo wrote:

@FelipeLink: Have you tried making a simple map of few notes and testing? I have and the HP looks almost exactly the same. The difference is that LNs have two judgements now instead of one, so the punishment is doubled for missing a complete LN. It is also slightly harder to achieve perfect scoring on LNs due to both the reduced leniency of LNs and that they have two judgements.

All according to plan in my opinion?
I believe his main concern was the fact that the HP recovery while holding on to LNs in v1 was far more lenient than what it currently is in v2.

Its hard to talk about this, but like on score v1, a map with HP 9, if you KEEP missing and hitting your hp will be like dropping a little and recovering FAST.

On scorev2 a map on HP 8.5, if you keep missing and hitting or even missing a little and start combo'ing the DRAIN is pretty big and the recovery is super small. i even stopped played for like 2secs on v1 and still got my hp to recover in like 1~2secs, on v2 if i rush any ln pattern my hp will be drained so much and the recovery is super small i cant even reach max hp
From what he told me, the HP drain of LNs in v2 was reasonable, he was talking about the HP recovery itself that is too harsh (regenerates too slowly) and causes him to fail much more easily than on v1. Maybe that's intended?


Also reposting this I guess, because the overemphasis on 200s is still a major problem and this would fix the problem:

Shoegazer wrote:

Rainbow Accuracy

Shoegazer wrote:

320s are very much underweighted because the only component of the scoring system that takes into account 320 accuracy is the combo component, which only has a 20% prominence. Add on to the fact that the difference between a 300 and 320 is so small and that the absolute difference between juan and Hudo's 320 count isn't that significant, it would make sense that 320s are really underweighted at the moment.

You could mitigate this by including 300gs into accuracy, but from what I've experimented it might create too much emphasis on MAX accuracy with charts that players have issues getting 96%+ on (and as a result would not be an accurate assessment of skill).

Alternatively, you can avoid including MAXes in the accuracy component and just increase the importance of MAXes to like 360 to increase the emphasis of it by a noticeable but not overpowering amount in the combo component, but that requires a bit more experimentation.
I initially wanted to increase the rainbow judgement weightage without embedding rainbows into accuracy, but no matter how much I changed it, the difference is very minor (~600-1,200 points) and a 200 will almost always be too powerful compared to a rainbow 300. So I scrapped that idea and thought that embedding rainbows into accuracy with a reasonable weightage and maybe making the curve more lenient would be the best idea.

I've been experimenting with weightages and discussing with people about how much a 200 should be worth compared to a 300. I initially thought that 310 would be fine (and a 200 would be worth 11 300s), but when it came to matches like this, if accuracy was the only factor, Argentina would win by 21,000 points. I do think that Argentina should win and it's a step in the right direction, but 21,000 seems extremely overwhelming since it undermines the fact that Poland had overall, noticeably less 200s. I tried it with harder charts too and they seem to favour rainbow accuracy a little too much for my liking - especially since when it comes to harder charts (where players struggle with), good rainbow accuracy is usually caused by variance rather than a higher skill level. 200s and worse judgements should determine performance for that.

I wanted to use 307 afterwards, but it still gave a bit too much emphasis for my liking, about 12,500 points for that Argentina/Poland match. I went down to 305, and the difference is about 6,800. I think that's ultimately the most reasonable assessment, and others I've talked to seem to agree with the prenotion that a 200 is about 21 normal 300s. Ignoring the bad judgements (since those values are pretty much set in stone at this point), this is probably (part of) the ideal solution. This does mean that only full rainbow scores are SSs, but I don't see that as a problem as frames of reference can be shifted.

Getting rid of the difference between a rainbow and a normal 300 in the combo scoring component is probably ideal too, since that should be in the accuracy component, not the combo component. If rainbows are included into accuracy, the combo component does not need a rainbow component.

I also wanted to soften the exponential curve a tiny bit when it comes to including rainbows, mainly because at a certain point extremely good accuracy is more caused by variance rather than a very high skill level - unless the performance is consistently done, which is not measurable with just one match and one attempt. The exponential I had in mind was Accuracy^(2 + 2 * Accuracy), but it's essentially Accuracy^4 - so 1 power down.

tl;dr: Embed rainbows into accuracy with a weightage of 305 instead of 320, change the accuracy curve to Accuracy^(2 + Accuracy * 2), remove the differentiation between rainbows and normal 300s in the combo component (both of them should have a HitValue of 30).
posted
That's the thing - it isn't any more or less lenient, the code is identical. It's just that there are other factors that cause you to lose more HP in ScoreV2. The actual rate of recovery of HP is unchanged. Perhaps if this starts to matter more in the later rounds I can issue a quick update, but I don't think it's good to make last minute changes until that's determined to be a problem by the map selectors.

And likewise I'm pretty sure it's too late now to improve on the weighting of MAXs. That will definitely be looked into for MWC7K, but probably not for MWC4K unless there are very significant results that show a fix is 100% needed. Let's wait until after the first round!
posted
Hmm, actually when we played with underjoy on updated v2 he said that hp recovery was too easy and he was playing on HR so no idea
posted
Am I the only one who thinks it's silly to have a system potentially changed through the tournament (That it'a specifically made for) because of being unfinished?

Unless you redo round1 after a fix if that ever was needed.
posted
I really don't like the idea of waiting for results of the round where the max:300 aspect matters the most by far, but eh, nothing I can do about it at this point.
posted
@smoogi: Fair enough on the HP recovery.

Khelly wrote:

Am I the only one who thinks it's silly to have a system potentially changed through the tournament (That it'a specifically made for) because of being unfinished?

Unless you redo round1 after a fix if that ever was needed.
I do agree that having system changes in midst of a tournament is a bit silly, but I do feel that if it's to ensure a more accurate assesment of ability, I don't think it's a problem in itself. The fact that it is unfinished does justify the change.

I would much rather see the change now, especially since group stages is the round where MAX accuracy matters the most, and cases like this and this (lim vs bumpinho) are going to be very common. I don't think the competitors will mind a change in MAX weightage especially since it'd be similar to scorev1 would look like (which makes for easier transitioning) and from what I've talked to with other players, they certainly wouldn't mind having a change in MAX weightage if it meant for a more accurate assessment of ability. The frame of reference of what defines an SS will be malleable, especially since they know that these context of an SS is now situational.

But that's how I see it really, I w ould much rather have short-term discomfort (for a long-term gain) rather than an implementation afterwards when in the stage that matters, it had a noticeable problem for certain facets of the tournament. Ultimately up to you.
posted
@smoogi:
I made a ''test'' map for trying how LN v1 works vs LN v2 and well i just found that the HP recovery is the same but well the HP drain is kinda a problem, with score v2 LN you can fail twice faster than v1, so well yeah thats the only problem i can find on lnv2, the recovery is not significant because the drain is too big.
And well, on hard LN maps even good skilled players can do rushs and miss like 1~3 lns in a row and on v2 this is like 100% dead and on v1 this isn't a big problem.

Oh and i miss my Hit Lightning, after the new beta update they just disappeared ):
(i talked with some players who use's beta and all of they hit lightning disappeared too)
show more
Please sign in to reply.