ITGmetrics

[J. S. Mill]

This is a long post with a very complicated purpose. I am going to motivate it by telling everyone a story. About thirty years ago, a fellow named Bill James figured out that nobody knew what the hell actually happened in the game of baseball. A bunch of players gave it their all, played 110% and other such cliche nonsense, but nobody could predict the winners with any level of accuracy. Heck, nobody even knew what mattered for winning. He introduced a theory of scientific and mathematical analysis, which he called sabermetrics, which (unless you're afraid of math) revolutionized the game. By developing scientific hypothesis and testing them through a strategy he called risky back-prediction, he discovered that previously unappreciated skills were vital for winning, and that you could win with a low budget if you understood these key aspects of baseball.

I work in baseball sabermetrics, and my job is to develop and apply many of these statistics. Recently, I was sitting at home, reading an ITGFreak topic about which ITG players were better then others. Although some players had good GS scores, others were better because they played 110%, gave it their all, were clutch under pressure, etc etc. Suddenly, a thought struck me. Nobody really has any idea what they're talking about, do they?

Despite about five years of playing, I realized I couldn't answer simple questions, like "What are five middle-range ten footers for score, emphazing diversity of charts over parity?" "What chance does JJK have of beating DukAmok in a fair match?" "What songs should Mike Quintance practice in order to get better fastest?" I could guess, and my hunches were pretty educated (as I'd later find out), but I had no scientific method of checking any of this information. I had hundreds of pages of song statistics, meticulously kept by Groovestats, and advanced tournament data of my own devising, but I was still clueless. Maybe In the Groove was too casual to know the answers to these questions?

I decided to find out. Over the last few weeks, I've applied some of my sabermetric knowledge to In the Groove. First, I developed a formula for integrated difficulty, the difficulty of a song to score on paper. I learned very quickly that this number was relatively meaningless. Variance of scores, the ability to remain consistent, was most important for tournament performance, and so I developed a form of ordinal and composite tournament difficulty.

I used these statistics to derive two different statistics for advantages. One of these I called Paper Advantage (PA), and I defined it as (Normalized Best - mean Normalized Best)(Integrated Difficulty). This was fairly useless. However, Tournament Advantage (TA), which I defined as (Tournament Difficulty)(Variance - mean Variance) seemed more promising. I made both ordinal and composite models of tournament advantage (called oTA and cTA)

Taking a chapter from the discipline called APBRmetrics (sabermetrics for basketball), I developed two statistics:

[Karl Popper]

Wow, 96 percent accuracy is amazing. And that example qualifier list is perfect. Great job. But how well does the accuracy hold up in predicting matches that are actually pretty close, such as in the finals and semifinals of tournaments? I'm assuming that number includes really obvious matches, like LilMegamanXAmok vs. random noob in the first round of a tourney.

[MinN_Limited]

Very impressive. I honestly don't understand the formula but it's quite intriguing, good work. The results you got were also pretty amazing. How long did it take you to work all this out?

[Emptyeye]

Oh **** YES this is interesting.

I'm definitely intrigued by this, and like MinN_Limited, I'd love to know where these statistics are derived from. In particular, I'm curious as to how Tournament Difficulty varies from Integrated Difficulty.

I presume that these formulas, with some manipulation, could be used in other settings, for instance "What are four DDR Standard song qualifiers, emphasizing chart-diversity over parity, and coming from either DDRMAX2 PS2 or DDREX2 PS2 (target audience: casual players whose skill limit is generally passing 9-footers)?"

As I see it, one of my main questions here is "does this have, or can it be modified to have, any practical application outside of tournament situations?" I realize that I'm in a distinct minority on this thinking, but my feeling is that the true "best" arrowsmasher should be adept at a variety of skillsets that tournaments don't (generally) presently test: The popular "tournament" format, yes, but also Marathons, Survivals, and within reason (Defining "within reason" is another problem I admittedly have yet to solves) random mods of combinations of mods on a song. You mention "practice and training" as a potential use for this, but maybe I want to, say, improve my mod reading, or become more proficient at marathons, or just improve my scores in general but without tournaments specifically in mind, as opposed to training for tournaments. Can this be used in that regard?

Either way, this is definitely some interesting stuff and I'm curious to read more.
_________________

emptyeye.com- Music, writing, life

My Recall- Now with actual, real arcade scores! O.o

[IHYD.dimo]

great stuff, but it's only applicable outside of socal seeing how i win 100% of the time.
_________________

[Synaesthesia]

How well is this system able to account for variability (particularly with the pad) of scenarios? I mean, how well can it factor in possibilities like, say for instance I'm playing someone who is somehow worse than I am, but you know I have allergies and so the probability of me sneezing and breaking focus could play a role in whether or not I lose? Also, does Groovestats track progress of scores entered? That could be useful in seeing how much someone's best score improves over a given interval, but I still think that wouldn't tell the entire story. I'm thinking of my DDRecall, which has the highest score I achieved on particular songs, but doesn't give you the real idea of how consistent I am. For instance, sure I can get 9 on Stoic, but an on-the-spot performance would more likely be 15-24. It seems to me you'd want to have data from every time I played a song so you would get less of a high-end skew of my scores and a better idea of what I consistently manage.
_________________

im a lasagna whale

G_G

[IHYD.DukAmok]

my interest is piqued

tell me more cory
_________________

Sappy_!?! wrote:

just to answer, if someone who stands next to you watching you play PSMO but you get a D on it, versus somebody who understands perfect attacking and stuff, will think you suck. A player is considered good in my opinion when a player of a higher level comments about you or see's you triple A a song. Or if somebody looks up to you. Hope it clarifies.

[J. S. Mill]

[Tyrgannus]

This is more complicated than the stamina formula I made for songs.

(And no, I don't just count tap steps)

My stamina calculator is impossible to calculate to the calorie because people differ in size and shape and they also differ in the movement of a certain song, but it is accurate relative to a person.

In other words, if my formula says that Delirium takes 1029 Dance energy (again, you can't use units like the calorie) than if the guy uses the same style (flatoot, arms on bar, whatever) than Anubis would cost around 734 Dance energy points.

The formula takes into account jumps, hands, footspeed, candles, and length of streams. While it's not perfect, It is fairly accurate.

Sorry, of on a tangent. The whole point of that was to say ITGmetrics are much more complicated and very interesting (though I find my formula interesting as well)
_________________

AA Bob wrote:

Summer is as much of a 12 as PSMO is a 9.

[mydixiewrecked]

[Tyrgannus]

A dance energy point represents a measurement of exertion needed. With my system, the dance energy points for a song should remain the same for everyone even though 1000 may be big for some people while 2000 may not be too bad for another. It's all proportional and that proportionality is based on your stamina and ability to conserve energy.

In other words, if Delirium requires 1029 Dance Energy Points (DEP), then I might find that inordinately tiring while someone else wouldn't notice the drain at all.

Did that answer your question?
_________________

AA Bob wrote:

Summer is as much of a 12 as PSMO is a 9.

[mydixiewrecked]

[J. S. Mill]

[HipHopNotik.]

If you are using GS, how important is it for everyone who is included to keep their GS accounts updated? (it sounds like a stupid question but I ask this for a reason).
_________________

[IHYD.Blake]

And how important is it to not have fake scores on it due to corrected pad excellents turning into 100%s?
_________________

[J. S. Mill]

[Emptyeye]

You had edited a post between the time I looked at it and the time I looked at it again, so apologies for the lateness of this response.

[J. S. Mill]

[Irish.MTA (Retired?)]

As a total math junky this is definitely intriguing work. Assuming I have any decent amount of free time I'll try to get in touch with you about some of this stuff - or feel free to IM me if you want to just talk about any data you think you might want to collect or anything I could do to help or whatever.

I'm going to a tournament next weekend. I can try to collect a whole bunch of data and you/I/we can try plugging it in if you want another set of test cases or something.

Keep it up.

Note: I'm a little worried that a lot of the players at this tournament won't have GS accounts. -_-
_________________

The winds of change are oft unpredictable...

[nasheq]

i think the entire 96% prediction accuracy is bullshit. unless you're comparing like lilQ to mr. scrub.. i feel this is absurd. also i'm sure you aren't claiming that you will predict future matches - only past matches. hence i find it pretty useless.