Guest
home
beatmaps
rankings
community
store
help
forum
  1. Forums
  2. osu!
  3. Gameplay & Rankings
  4. osu!mania

Why not EXScore?

posted
Total Posts
16
Aqo
2,710 posts
Joined April 2010
Topic Starter
Aqo
Many rhythm games I know, whether for PC or mobile or even arcade, use an EXScore scoring system, where basically

maximum judge: 100%
the judge below that 50%
anything else: 0%

in osu!mania this would translate to a play with 600 [rainbow]s, 300 [300]s, and 100 [other judgements] out of 1000 notes being worth 750K score.

this is simple and good. why not just use this. anyone thinks otherwise?

it creates a direct correlation between how well you hit the notes and how much you get rewarded for it, encourages not-mashing, and is "fair" for random misses or other stuff like that.

SPOILER
for the record, osu!standard accuracy is similar to EXScore, except instead of 100%/50%/0%XX it uses 100%/33%/16.6%/0%XX. osu!standard scoring isn't great but the accuracy on it is a fair indicator of how well you did and most people just look at that.
Halogen-
Former osu!mania 4K Champion: 2015
574 posts
Joined September 2009
Halogen-
Ignoring judges beneath regular 300s seems ridiculous to me, given that plenty of songs have scores chock filled with 200s. Wouldn't this also reduce the power of being able to do long notes? Players could get 100s/200s on releases and not be reprimanded any differently according to that example.
Full Tablet
2,542 posts
Joined September 2011
Full Tablet
Ex-Score doesn't differenciate between the judgments worse than the 2nd best (in o!m's case, a 300). A 200 is definitively better than a 100/50/Miss, so it should have a lesser penalty than those other judgments.

Ex-score has it's merits. It is based on judgments counts and doesn't use combo, and the scale of the results uses the possible range of values better than the current accuracy percentage formula (in accuracy percentage, a mashed score gives at least 60% of the possible value, unless the map is so fast you cant even mash fast enough, and the difference between a 95% acc and 97% is quite big even though the numbers are close).

Also, Ex-score is so simple to calculate you can do it in your head without having the computer do it for you.
Aqo
2,710 posts
Joined April 2010
Topic Starter
Aqo
well you can always adjust the rates of the EXScore used.

Something like 100%/50%/25%/0%XX should work for mania. (i.e. 200s still give reward, anything below is a null.)

lets face it a 100 is a miss


100ms off... and that's on OD10. just to give you some perspective, 1/4th notes on 180bpm are 83ms difference. if you get 100s, it literally means you played this pattern:



like this:



I seriously don't think 100s deserve more than 0% score. there's a huge difference between a player who plays patterns the first way or the second way and scores should reflect this. (note that the player on the 2nd example would still get 50% EXScore from this)

Full Tablet wrote:

the difference between a 95% acc and 97% is quite big even though the numbers are close
imo this is one of the biggest strengths of EXScore. it allows for a nice linear progress rate on the map from 80% -> 90% -> 95%, without neglecting the fact that 95%->97% is a huge skill gap.
Last edited by Aqo , edited 1 time in total.
Endaris
5,282 posts
Joined June 2010
Endaris
Should rather rework how the hitareas and especially 50s scale up with high OD though. It's not really a maniaproblem imo cause the same applies in Standard+Taiko.
Redon
302 posts
Joined November 2013
Redon
‎
Last edited by Redon , edited 1 time in total.
Full Tablet
2,542 posts
Joined September 2011
Full Tablet
If we used a system where each judgment had a value, and accuracy was determined by adding their values together (not the best system IMO, but it is simple), I would rather have the values of each judgment dependent on OD, and with ratios like this (based on the linearization of the normal distribution fit near 100% accuracy, using the timing windows of scorev1):

OD10:
Rainbow/300/200/100/50/Miss -> 1000/981/919/691/356/0
OD0:
Rainbow/300/200/100/50/Miss -> 1000/988/818/586/292/0

(2 of the judgments values must be arbitrary, in this case a Rainbow was made 1000, and a Miss 0; if, instead, made the value of a 300 be 500, the other judgments would have become negative).

Then, the ratio could be scaled with a monotonic function so the results use the range of values more evenly in common plays.
Kempie
51 posts
Joined September 2013
Kempie
StepMania's scoring system is similar to EXScore, with the biggest difference being lower judgements giving negative score. Implementing this in osu!mania might look like this:

r300: 2 points
300: 1 point
200: 0 points
100: -2 points
50: -4 points
Miss: -8 points

Bad timing and misses are harshly punished, meaning that spamming and inconsistent gameplay will give you a lower score. Consider this shitty PP farm score of mine on Everlasting Message [GRAVITY]:
Score : 730,048 (93.93%)
Max Combo: 358
MAX / 300 / 200: 897 / 487 / 150
100 / 50 / Misses: 29 / 10 / 19

That's 73.0% of the maximum achievable score. Using the previously mentioned scoring system, I would've got a score that's only 63.8% of the maximum score. Turn the misses into 200's and the 50's into 300's, and the score increases to 70.1% of the maximum score. Missing and bad timing is punished very harshly indeed, but I think it should be.

This scoring system can easily be fine tuned to make it less punishing (e.g. 100/50/miss give -1/-3/-5 score), reduce the gap between r300 and 300 (e.g. r300/300 give 5/3 score). It's simple, builds upon scoring systems present in other well respected VSRG's, and can easily be balanced as necessary.
Aqo
2,710 posts
Joined April 2010
Topic Starter
Aqo

Drojoke wrote:

reduce the gap between r300 and 300
actually kind of the point of exscore is the large gap between 300g and 300, that's what creates the correct scale rate between 80% 90% 95% 97% accuracy rates in score distribution
Full Tablet
2,542 posts
Joined September 2011
Full Tablet
The values of the judgments I proposed are made so, while having ~100% accuracy and many notes played, each relative increment on the amount of each bad judgment hurts the overall accuracy relatively about as much as they hurt the Normal Curve Maximum maximum-likelihood fit of the hit errors.

Aqo wrote:

Drojoke wrote:

reduce the gap between r300 and 300
actually kind of the point of exscore is the large gap between 300g and 300, that's what creates the correct scale rate between 80% 90% 95% 97% accuracy rates in score distribution
There are several ways to adjust the scale, not just changing the values of the judgments.

For example, in terms of balance:
Rainbow/300/200/100/50/Miss -> 1000/981/919/691/356/0
Is the same (besides rounding errors), but with another scale, as:
Rainbow/300/200/100/50/Miss -> 1000/500/-1130/-7130/-15900/-25300

But when using those values, negative values in the accuracy rate become possible. Another alternative is taking rate calculated as the sum of each judgment value obtained, and then apply a transformation to change the scale. For example:
700 "Rainbows", 200 "300s", and 100 "100s" with the judgment values I mentioned would be a 98.81% rate, which is quite close to 100% or any other high accuracy play. So, instead, the value shown is the accuracy rate to the 15th power (98.81%^15 = 83.56%), for example.
abraker
Global Moderator
8,295 posts
Joined July 2014
abraker
Accuracy based scoring is flawed in the sense it doesn't care about disproportionate difficulty. The whole map can be easy with one hard part and you can get a high score.
Aqo
2,710 posts
Joined April 2010
Topic Starter
Aqo

abraker wrote:

Accuracy based scoring is flawed in the sense it doesn't care about disproportionate difficulty. The whole map can be easy with one hard part and you can get a high score.
but that will always be the case anyway. that's a problem with the map, not with scoring.

you think with combo scoring you can't get an absurd score with something like shuffle heaven where 80% is EZPZ max combo compared to the ending...?

purely based on experience, EXScore has proven itself as a very accurate performance measurement in all of the games where it was employed, whereas any sort of combo-based scoring has proven itself to lead players to not even pay attentions to scores because they're pretty much meaningless in any game where it works that way.
abraker
Global Moderator
8,295 posts
Joined July 2014
abraker

Aqo wrote:

but that will always be the case anyway. that's a problem with the map, not with scoring.
With that mentality, you are almost saying what mappers can and cannot do. Maps can be whatever they come to be to express whatever the mapper wants. Scoring should be made such it works for anything thrown at it.

Aqo wrote:

you think with combo scoring you can't get an absurd score with something like shuffle heaven where 80% is EZPZ max combo compared to the ending...?
I think combo based scoring and accuracy based scoring are flawed and that there is a better scoring method. I and Full Tablet have been trying to push out the idea of difficulty based scoring as of late. My method of difficulty based scoring can be found here.
Kempie
51 posts
Joined September 2013
Kempie

abraker wrote:

Accuracy based scoring is flawed in the sense it doesn't care about disproportionate difficulty. The whole map can be easy with one hard part and you can get a high score.
What's so flawed about a scoring system that doesn't care about disproportionate difficulty? So what if getting a 100% score is a ton harder than getting a 97% score on one specific map, but getting a 100% score is only a little bit harder than getting a 97% score on another map. I'd much rather have a simple scoring system with the player filling in the gaps (e.g. judging whether a 97% score on a specific map is still impressive due to disproportionate difficulty within the map), than a hugely complex scoring system that tries to compensate for every little detail.

The only problem that comes to mind is pp calculation. Perhaps this could be fixed by somehow calculating the difficulty of getting a specific score on a specific map. Using this combined with an accuracy based scoring system even allows generating another score that does compensate for disproportionate difficulty. However, as mentioned earlier, I don't really see why such a score is necessary (except for pp)
abraker
Global Moderator
8,295 posts
Joined July 2014
abraker

Drojoke wrote:

What's so flawed about a scoring system that doesn't care about disproportionate difficulty? So what if getting a 100% score is a ton harder than getting a 97% score on one specific map, but getting a 100% score is only a little bit harder than getting a 97% score on another map. I'd much rather have a simple scoring system with the player filling in the gaps (e.g. judging whether a 97% score on a specific map is still impressive due to disproportionate difficulty within the map), than a hugely complex scoring system that tries to compensate for every little detail.
I will admit accuracy based scoring works if you have the player filling in the gaps. Ultimately, the final result can be compared knowing score and difficulty of the map.

I like difficulty based scoring for two reasons however. One because pattern worth is not neglected and two because you can compare just the scores without needing any extra info. Because getting a 97% score can be difficult or easy, we need to know the difficulty of the map to deduce anything about the number. When using difficulty to calculate score, you can look at that one number, score, and easily compare it to other scores without needing anything else.

I just like the system to represent the correct value and the players have as little roles filling the gaps as possible.
Full Tablet
2,542 posts
Joined September 2011
Full Tablet
To illustrate my idea in how to account for difficulty in performance points:

Map A: 900 easy notes, and 100 hard notes.
Map B: 1000 easy notes.
Map C: 1000 hard notes.
Map D: 900 easy notes, 1000 hard notes.

Then, the performance points achieved in those maps should follow curves similar to this:
score/pp curve

The points show are the ones used for the monotonic cubic interpolation (those points would be the ones that the difficulty algorithm would output, in order to represent a continuous curve with a finite amount of data).

One of the issues I see with including difficulty directly in the score system is feasibility. Since making an accurate difficulty algorithm is not an easy task, and it's probable people come up with a tweaked or completely different algorithms over time, it would be better if those tweaks or new algorithms could be applied to old scores.
Please sign in to reply.

New reply

0
1 / 16