Community Forum

# Scoring in mania is bad

posted
Total Posts
23
Topic Starter

8 misses > 200+ Rainbows
who knew?
I see what you mean
What else is new?
Or better yet, what isn't new?
Fitting the distribution of judgments and their respective timing windows to a centered normal distribution (each hit note is considered as "hitting anywhere inside the timing window of the judgment obtained", and each miss is considered as "hitting anywhere outside the timing window of all the judgments", and then performing a maximum likelihood estimation based on that), the first play is slightly worse than the second play (34.0ms standard deviation vs 32.4ms, assuming the maps is OD5.5, the OD most commonly used in stepmania converts). If the 2nd play had 9 less Rainbows and 9 more misses, then it would be considered worse than the first play by this method.

While I do agree that the 2nd play should be valued higher than the 1st play, this case in particular doesn't highlight the problem as much as it could. Worse cases would be when a play has strictly better distribution of judgments than other; for example, MAX/300/200/100/50/MISS of 600/200/5/0/0/0 vs 600/197/8/0/0/0, but the 2nd play is scored higher because the 200s were more spread out compared to the first play.

#### Yyorshire wrote:

What the fuck

How does that even work
it doesn't

#### Yyorshire wrote:

What the fuck

How does that even work
It is possible to get a relatively high score with low accuracy if the play is good for the most part, but there are several misses or bad judgments near the end of the map (or a few clusters of misses and bad judgments in different parts of the map, in the case of a map with several thousands of notes).
Score in osu mania is worthless. Don't pay attention to it.

Your acc and judgments are enough.
lol..
acc = score

the other thing is a lie

#### abraker wrote:

acc = score

the other thing is a lie
The acc% formula is not very good for measuring accuracy.

If still using a linear formula (for simplicity), then just the base values of the judgments should be changed.

300/300/200/100/050/000
They should be around:
300/296/262/212/175/000 (or, equivalently, 304/300/265/215/177/0)
Which correspond to the parameters of the linear Minimax approximation of a formula that fits a normal curve to the judgment distribution (for OD8.5, the values vary with OD), giving at most 5.1% error between the formulas (in the case when the linear approximation gives a value 41.93% accuracy with a judgment distribution of only Rainbows and Misses)

The values could then be scaled by an arbitrary monotonic function so the range of usual values is bigger (like scorev2 scales the acc% value in the acc portion of the score).

#### abraker wrote:

acc = score
This would be the correct way to go about things, but apparently we need to factor in combo somewhere because STD does it too. Also consistency something something blah blah blah. Nothing should be a factor except accuracy.

#### abraker wrote:

acc = score

the other thing is a lie
The acc% formula is not very good for measuring accuracy.

If still using a linear formula (for simplicity), then just the base values of the judgments should be changed.

300/300/200/100/050/000
They should be around:
300/296/262/212/175/000 (or, equivalently, 304/300/265/215/177/0)
Which correspond to the parameters of the linear Minimax approximation of a formula that fits a normal curve to the judgment distribution (for OD8.5, the values vary with OD), giving at most 5.1% error between the formulas (in the case when the linear approximation gives a value 41.93% accuracy with a judgment distribution of only Rainbows and Misses)

The values could then be scaled by an arbitrary monotonic function so the range of usual values is bigger (like scorev2 scales the acc% value in the acc portion of the score).
There shouldn't be discrete hit error scores to begin with. Let it be based on the normalized continuous Gaussian function with OD controlling the sigma. Wife scoring is on the right track.

#### johnmedina999 wrote:

This would be the correct way to go about things, but apparently we need to factor in combo somewhere because STD does it too. Also consistency something something blah blah blah. Nothing should be a factor except accuracy.
Fork the client. Establish acc as the recognized scoring method and adjust it to be something like I mentioned in this post. Screw combo.
#screwCombo
VoidCreeper345

#### VoidCreeper345 wrote:

tbh
Before I even bother, have the following issues been fixed?
• - All maps need to be loaded every single time you start the game
- Some of the global setting are lost upon restart
- Is there a good guide/tutorial to making a noteskin?

#### VoidCreeper345 wrote:

tbh
Before I even bother, have the following issues been fixed?
• - All maps need to be loaded every single time you start the game
- Some of the global setting are lost upon restart
- Is there a good guide/tutorial to making a noteskin?
-All maps still need to be loaded
-I've never had my global settings lost
-There's probably a good guide somewhere but just go here tbh https://etternaonline.com/noteskins

Honestly even despite these "issues" it's still worth playing for rates & the difficulty calculator.