Player Rating System

Alright, I'm not sure where the thread went, so I'm starting another one. As some of you might know, I'm interested in coming up with a program (I won't call it a "system" yet, since it's not meant to be full-fledged yet) that can allow a player to keep track of his progress.

BACKGROUND
In most sports, there are rating systems in one form or another, to rank/rate the participating entities. I plan to come up with a casual variant of this for our own usage, if it's possible.

Imagine this fictitious scenario: today I played さいたま2000 on oni mode and I finally passed it! I take a photo of it, and then I upload it to ProgT (no name for this program yet). ProgT reads in the image and extracts the necessary information (e.g. difficulty, score). ProgT then shows me my rating for this song, whether I beat any hi-scores, and my cumulative rating for all songs so far.

WHAT I'M GOING TO DO
The image reading algorithm is quite complicated, so I'm not concerned with that for now. Right now I would like to try to implement something to take in user input (integrity for now), and then output the results. I'm posting here because I'm interested in finding out what you think the rating system should be based on.

Here's a bit of maths for the nerds.

SONG RATING
Given the following variables (those in red are provided by user):
n : song difficulty (no. of stars)
p : no. of 良 / total no. of beats, 0 <= p <= 1
q : no. of 可 / total no. of beats, 0 <= q <= 1
r : no. of 連打
t : suggested value for no. of 連打
Rating for song, R = [p + (q / 2)] x n + (r / t)
(if t = 0 then let (r / t) = 0)

Example: Song: 童话 (oni)
Given n = 6, p = 0.8, q = 0.2, r = 100, t = 125,
R = [0.8 + (0.2 / 2)] x 6 + (100 / 125) = 6.2

CUMULATIVE AVERAGE PLAYER RATING
Given the following variables:
n : no. of songs played
C : cumulative rating
R : song rating
Cumulative average player rating, C' = [(C x n) + R] / (n + 1)
(i.e., recomputing the new average rating)

Example
Given n = 9, C = 5.5, R = 6.2,
C' = [(5.5 x 9) + 6.2] / (9 + 1) = 5.57

NOTES
Regarding the above on CAPR, one pitfall is that the user can play the same old song over and over again to keep his rating up. Therefore, an improvement could be to manage average song ratings, then calculate the CAPR from there, but that would be for later. Basics first.

I know I'm making things look too serious, but please give me feedback on this. Thanks!

o.O

You think i understand?? *laughz*
I'm not a maths person lehz.. xD

What I need input on is whether the factors concerned are valid enough, or are there other factors I should account for.

In a nutshell, the harder the song, the higher the rating you will get. You're rewarded for more 良, of course, and 連打 also plays a small part.

Hmm.. how about those doing double play then?
Marks gotten in double play is different for songs played with 2 hands though.
So i a double player is mostly doing double play instead, how can it be calculated?
Should be a different set of algorithm i guess?

And how about getting slightly more scores for those ninja modes like opposite? =D
Eg. Oni bambini 2x speed. but have to see which songs cos' some slow songs the faster the speed, the easier it gets.

Looking at this post at 2:37am is so wrong X_X

I think the algorithm will become rather complicated because (somewhat quoting yuu) some songs get easier in x2 or faster and so there can't be a standardized set of algo for all songs.

Why are these interesting posts coming up when I still have exams!!! argharghargh

Mint brought up a very good point just now (offline discussion).

If we are appealing to the international audience, then it is obvious that our song variety needs to cover all possible versions. Data gathering, in this aspect, is going to be very tough.

Therefore, what I propose is that the variety of songs be restricted to those in the Asian version for now (because it is easiest for me to get data). Once the system is found to be working reliably, it is only a matter of expanding the song selection.

You can't possibly getting all the data for all possible version without any help from the rest. And in terms of rating, some songs are over-rated. As in some of the songs, it stated xx no. of stars but it's actually easier/more difficult than the rest of the series.
haha

Even so, the no. of stars is the closest thing to quantifying the difficulty of a song. I could really make things complicated by coming up with an expert system to poll players on what they think a song should be rated, but that's beyond my concerns for now.

But as you said, even though you find this song is easy, but to the rest, it might not be easy.. So kinda subjective..

Oh, and taiko game series itself, they have different stars rating too, you know? =D
Just to name some:
Eg. Soul Calibur in psp oni is 9* while the arcade is 7*
Bumble bee in DS oni is 9* while arcade is 10*

Pssssstt..... Maybe if it works on the Asia version of Taiko, it can be submitted to NAMCO so they can put it into all games XD since they are the only people who knows the intricacies of the game best XDD

Work on Asian version in arcade?? o.O
How should we go about it then?? Do the real installation of the card reader and stuffs ma? @.@

Nah I mean if ck can program something that is based on the asian version

[I need to stop reading the forum and get back to studying T_T]

But those cards etc are installed on the machine though.. so..
haha

My original playoff of Chinkai's idea was a rating test. it would be a web-based test that would be specific to what console you have available. So there would be a variety of the test for each console and version. The test creator compiles the songs that he/she thinks are fundamental (This is also specific to difficulty level). The rating portion of it is still a branch off of Chinkai's original thoughts, but the songs are given custom weight, since we can all agree that the difficulty is not what it thinks it is. So the test would be able to score you on different skills required for taiko. Let's say you used these songs for the test:

Flight of the bumble bee
Ekiben2000
Soul caliber

Now the test would give weight to certain skills for each song. So ekiben would give weight to stamina. And flight of bb would give weight to ability of long strings etc...

So the idea is that the test will ask you how you did on each song, and based on that, it will tell you your rating overall and rating for certain skills. It would just be a simple servey-type test, people can cheat on it, but there is no point. It is for you to know what you need to improve on, and for you to tell others (like in your sig). And getting the weights and song selection is important. If one person takes the test with ver 10 songs, their results should be VERY close to their results if they took the test with ver 11 songs. That's just my idea, anyway.

Pyrozz wrote:My original playoff of Chinkai's idea was a rating test. it would be a web-based test that would be specific to what console you have available. So there would be a variety of the test for each console and version. The test creator compiles the songs that he/she thinks are fundamental (This is also specific to difficulty level). The rating portion of it is still a branch off of Chinkai's original thoughts, but the songs are given custom weight, since we can all agree that the difficulty is not what it thinks it is. So the test would be able to score you on different skills required for taiko. Let's say you used these songs for the test:

Flight of the bumble bee
Ekiben2000
Soul caliber

Now the test would give weight to certain skills for each song. So ekiben would give weight to stamina. And flight of bb would give weight to ability of long strings etc...

So the idea is that the test will ask you how you did on each song, and based on that, it will tell you your rating overall and rating for certain skills. It would just be a simple servey-type test, people can cheat on it, but there is no point. It is for you to know what you need to improve on, and for you to tell others (like in your sig). And getting the weights and song selection is important. If one person takes the test with ver 10 songs, their results should be VERY close to their results if they took the test with ver 11 songs. That's just my idea, anyway.

This is a VERY late quote of pyro's ideas on the rating system, but I think we should continue discussing about this issue because though this issue is long and has much to cover, and is of real great value. This could mean a huge difference in the taiko grading field, whether for competitive players or just players who wish to work on their ability as a self-gauge method.

Firstly, I would just like to say that chinkai you really have some great ideas there, and that that formula does seem quite sound to me at first. and hey thanks for starting this thread xDD

But on closer inspection and after yuu and pyro have posted I realised that yea there are some problems with this (eg. Original namco rating of songs, variety of songs, presence of lian da, overweightage of lian da, skills and abilities tested, constraints involving repitition of same songs over and over again).

I do however, agree with pyro's method of testing because it does cover the various aspects in taiko.
Because if one follows a standard formula, notice that:
1. The presence of lian da has to be seriously minimized because of the presence of few lian da's in higher difficulty songs(most), and it's hard to give a gauge of the "standard no." in each lian da, and that the value of lian da is too much,

YET it's too little if we minimize it in the case of "PaPaPa Love" in Oni Mode, 8*, because the song places GREAT EMPHASIS on the lian da section, with 120 hits within a few seconds.

2. The overrating of songs/ underrating.
I guess many would agree that KAGEKIYO is overrated, the coded version of KAGEKIYO is as well, and so is the coded version of HareHareYukai(which is suited for double drums).

3. Skills Involved/ Limited variety of songs tried
Some people may prefer to keep playing the same few songs. leading us back to the first problem of creating a formula.

So if we look on the other side of things, following Pyro's initiation,
If we do create a test set of songs, a variety of skills can be tested, for a certain machine version and a certain level. Songs from different sections can be simultaneously tested in one test to prevent people from playing the same few songs.
Then, the rating can be obtained by seeing the number of "test-sets" passed, with certain test ratings for each test.
We can also have similar types of songs tested together, and give it a cool name XD and then the rating obtained from popular demand for standard will decide the points obtained from it, then we can have a full system whereby all the tests are named out and the total points obtained upon the total points possible to be obtained will decide how good a player is. However this method will be controversial to many because its based on certain test sets. xD

We could even give the test sets levels according the total difficulty level in average.
Thus, the points obtained from a set will be decided based on the total difficulty level involved, so that a higher difficulty level will have a higher chance of getting more points.

This is based on a Level8 combination. F rank = (failing any song)
F= 1pt, E= 2pt, D = 3pts, C = 4pts, B= 5pts, A= 6pts, S=7pts, SS= 8pts

if its based on a Level10 combination, this is how it works.
F= 1pt, E= 4pts, D= 5pts, C= 6pts, B= 7pts, A= 8pts, S= 9pts, SS= 10pts

If its based on a lower level combination however, the ranks move up, and lower ranks are not included except "F" rank, because failure is still possible, but chances of getting a higher point is impossible because of level limitation.

if its based on a level 4 combination, this is how it works.
F= 1pt, A= 2pts, S= 3pts, SS= 4pts.

E.g:
Mission: Lv.6 "J-POP"
PEACH 6*
Tentai Kansoku 6*
Love So Sweet 7*

F: Any Failure
C: Any Pass, 1.55m>
B: 1.55m+
A: 1.925m+
S: 2.3M+
SS: 2.7M+

sorry this rating makes no sense. -.- my calculations cmi.

Example only.

Hmm, I'm not sure I understand, blyx

Possible to put it in simpler terms? The example isn't very concise...

nvm forget it its too tedious xD
No point No Point.