Since mania looks #rip the way things are going with scoreV2, I am "reinventing" mania's scoring system by making my own version of it. This version would be geared more towards VSRGs being simulators rather than games. As such, in this version I am throwing away the discrete scoring system (50, 100, 200, 300, MAX) along with the concept of OD and applying the hit timings to a Gaussian function which determines the acc worth. My question to you is what millisecond range would you consider a miss to be? Also what millisecond range would you consider normal and really good?
forum
Your opinion about OD
posted
Total Posts
6
Topic Starter
Personal opinion:
Some thought(s);
- It's better to discretize the timing errors into score value intervals (timing windows) rather than giving timing errors a continuous treatment when scoring performances in a VSRG.
- While I admit that it is feasible to re-evaluate the current timing windows based on a Gaussian curve, I don't know whether it really is a good idea to abandon timing windows and solely rely on the timing errors.
- The timing window is a comfortable concept for most people here. You have to explain why the timing window system being scaled with OD is a #rip, and provide arguments as to how a new scoring system is beneficial.
Some thought(s);
- There should be a cut-off value of the Gaussian function. Above such a value, no matter how small a hit error is, it's considered to be MAX. The cut-off value would mean that there's still a timing window. The rationale is that: a timing window is still needed because human players are not robots. If this has been your intention all along, we might have been thinking towards the same direction.
The score system I proposed was based on fitting a Gaussian curve to the distribution of timing errors based on the judgment counts.
Not using the judgments and instead using more precise measurements of the timing of each hit would make the accuracy calculation more accurate than in the case of only having the judgment counts. (The game only has finite precision for each hit, but 1ms is enough to make a good approximation).
In the end, there has to be at least 2 timing windows, one for "Miss", and one for "Hit" (the game would consider the exact timing of each hit inside the hit timing window).
If in a play where the player hits each note in the timing window "k" times (with timing errors X= x1, x2, ..., xk), and misses "m" notes (the timing window for a Hit is defined by [-L,L]ms, hits outside that timing window, or not hitting the note at all, are considered misses), then the standard deviation "sigma" of the Gaussian curve that fits the errors is the positive value that solves the equation:
When there are both misses and hits, then solving the equation might require numerical methods.
An approximation for this case (good approximation when sigma is not too big), "s" is the sum of the squares of the timing errors:
When there are only hits, sigma is just the square root of the arithmetic mean of the squares of the timing errors.
When there are only misses, sigma->infinity (numerical methods would always fail at solving for this case).
Not using the judgments and instead using more precise measurements of the timing of each hit would make the accuracy calculation more accurate than in the case of only having the judgment counts. (The game only has finite precision for each hit, but 1ms is enough to make a good approximation).
In the end, there has to be at least 2 timing windows, one for "Miss", and one for "Hit" (the game would consider the exact timing of each hit inside the hit timing window).
If in a play where the player hits each note in the timing window "k" times (with timing errors X= x1, x2, ..., xk), and misses "m" notes (the timing window for a Hit is defined by [-L,L]ms, hits outside that timing window, or not hitting the note at all, are considered misses), then the standard deviation "sigma" of the Gaussian curve that fits the errors is the positive value that solves the equation:
When there are both misses and hits, then solving the equation might require numerical methods.
An approximation for this case (good approximation when sigma is not too big), "s" is the sum of the squares of the timing errors:
When there are only hits, sigma is just the square root of the arithmetic mean of the squares of the timing errors.
When there are only misses, sigma->infinity (numerical methods would always fail at solving for this case).
This is a tricky one. In existing VSRG's, a wide window for the worst judgement actually has a purpose: it's a lot harder to get good score when mashing. This is probably why LR2 has a 200ms window for its 'Bad' judgement (source), regardless of the timing difficulty.abraker wrote:
... what millisecond range would you consider a miss to be? Also what millisecond range would you consider normal and really good?
Because of this, I think it's better to have a big window for terribly inaccurate hits. At least 150ms, maybe even up to 250ms. I like LR2's 200ms and StepMania's 180ms. Anything beyond that is probably just inaccurate random mashing or a complete miss.
Based on VSRG's I played and my personal oppinion:abraker wrote:
... what millisecond range would you consider normal and really good?
0-12ms: Insanely accurate
12-18ms: Very good
18-35ms: Good
35-50ms: Decent
50-90ms: Bad
+90ms: Shit
Topic Starter
Technically the concept of a timing window is not entirely thrown away. While I want a constant timing window for the scoring so everything remains relative for the most part, the sigma in the Gaussian function can still be changed to match various sensitivities to hit timings.FrenzyLi wrote:
The timing window is a comfortable concept for most people here.
The only other issue I can think of is that I see the scores light up from a timing window as a helpful indicator of how well I'm hitting. This will diminish with continuous timing windows as there will be little sense for a certain score to light up. This however, can be fixed by having "guards" which would allow the player to customize which discrete timing window. These guards it should light up letting the player know where they are hitting much like the score does now to indicate that.
Is there a reason? I think better resolution doesn't hurt, if anything it will make it more clearer where you stand on terms of accuracy.FrenzyLi wrote:
It's better to discretize the timing errors into score value intervals (timing windows) rather than giving timing errors a continuous treatment when scoring performances in a VSRG.
I would rather rely on timing errors rather than scores with windows which have arbitrary accuracy worth (to my knowledge I have no idea why the timing windows are worth what they are right now). I want to be able to compare any score to another score and various OD makes this rather hard to do without recalculating the score of the replay based on the desired OD.FrenzyLi wrote:
You have to explain why the timing window system being scaled with OD is a #rip
Unlike the current scoring method, I intend to make my scoring system a method of measurement. As a measurement, a 1M would be all 0ms timings, and a score of 0 would be all misses. Where to put hit timing in between on the scale is what I am trying to figure out right now. For those who want to just play a game and have fun, this won't be really beneficial as this would make the game more harder and brutal. For those who interpret this is as a simulator rather than a game and want to put their performance on a metric, this would be beneficial. I do see the need for two types of scoring systems in this case to please both sides.FrenzyLi wrote:
and provide arguments as to how a new scoring system is beneficial.
Humans are not robots, but it is not my intention to have a cut-off. A 1M score will NEVER be achieved by a human, heck 900k sounds barely believable, but striving towards perfection has always been a goal for someone who wants to improve. Much like no human will ever complete a 100m dash in 0 sec, no human will get a 1M score on any map. You don't just declare that 5 seconds would be a cutoff for the 100m dash, that's ludicrous. That time is as much as a measurement as the accuracy is the result by the player.FrenzyLi wrote:
There should be a cut-off value of the Gaussian function. Above such a value, no matter how small a hit error is, it's considered to be MAX. The cut-off value would mean that there's still a timing window. The rationale is that: a timing window is still needed because human players are not robots. If this has been your intention all along, we might have been thinking towards the same direction.
Hmmmm... Let's say you only count some ridiculous narrow range, say 0-12ms, as a hit. You count everything from 12-200ms a miss. And nothing if beyond that. Will this work as against button mashing as well?Drojoke wrote:
This is a tricky one. In existing VSRG's, a wide window for the worst judgement actually has a purpose: it's a lot harder to get good score when mashing. This is probably why LR2 has a 200ms window for its 'Bad' judgement (source), regardless of the timing difficulty.
Yes, but it would be very confusing if hitting way off is a 'miss', but completely missing a note is some other kind of 'miss'. That's were judgements like osu!mania's 50, StepMania's Boo and LR2's Bad come into place.abraker wrote:
Hmmmm... Let's say you only count some ridiculous narrow range, say 0-12ms, as a hit. You count everything from 12-200ms a miss. And nothing if beyond that. Will this work as against button mashing as well?