Where will your DR be months from now? A prediction with Microsoft Excel & Kudos Prime (w/tutorial)

[John Crellin]

I guess that's how perspectives work because where I am coming from, this isn't even close to science! I consider something science when it involves a hypothesis, the scientific method, and lots of data from the natural sciences (Geological, Biological, Chemical, Physical).

While I do have a hypothesis and data, I consider this just a statistical test. You are correct it is by no means the most accurate way (there are too many variables to depict it), but I think it is a good ballpark figure for a GTS driver who has a lot of racing data in their kudosprime, but not enough data that they are close to plateauing.

As for modeling someone like ORMA_Snow (http://www.kudosprime.com/gts/stats.php?profile=1075382), I think that would work, but there is little to gain from the data as a high S driver is forced into a plateau at 75K.

What are you a professor of?

Not a professor! I'm a PhD student in Chemical Biology. Im also a TA in a biochem course. So we are peers if anything!

I'd say statistical modeling is a scientific tool. I use it all the time in my research.

I think the point is that they are forced to plateau, because there is a maximum, a linear trend line won't be very representative so a different statistical function would be better. Like a kinetic rate function which contains a lag phase and a maximum comes to mind.

Have to play around with it but if we find a profile that contains these characteristics (yeah i know, choosing data like this is a major bias) we could then figure out how to modify the equation for stretching and compression (usually multipling x by some factor that is unique to the individual) which helps fit a relatively fast growth (like ormas) to a slower growth (like mine) and fit a curve that models the Elo systems behavior.

What do you think?

 

[Alpha Cipher]

Not a professor! I'm a PhD student in Chemical Biology. Im also a TA in a biochem course. So we are peers if anything!

I'd say statistical modeling is a scientific tool. I use it all the time in my research.

I think the point is that they are forced to plateau, because there is a maximum, a linear trend line won't be very representative so a different statistical function would be better. Like a kinetic rate function which contains a lag phase and a maximum comes to mind.

Have to play around with it but if we find a profile that contains these characteristics (yeah i know, choosing data like this is a major bias) we could then figure out how to modify the equation for stretching and compression (usually multipling x by some factor that is unique to the individual) which helps fit a relatively fast growth (like ormas) to a slower growth (like mine) and fit a curve that models the Elo systems behavior.

What do you think?

You're too smart : D

 

[John Crellin]

You're too smart : D

Im more like a case study of the Dunning-Kruger effect

 

[TheGeologist]

Not a professor! I'm a PhD student in Chemical Biology. Im also a TA in a biochem course. So we are peers if anything!

I'd say statistical modeling is a scientific tool. I use it all the time in my research.

I think the point is that they are forced to plateau, because there is a maximum, a linear trend line won't be very representative so a different statistical function would be better. Like a kinetic rate function which contains a lag phase and a maximum comes to mind.

Have to play around with it but if we find a profile that contains these characteristics (yeah i know, choosing data like this is a major bias) we could then figure out how to modify the equation for stretching and compression (usually multipling x by some factor that is unique to the individual) which helps fit a relatively fast growth (like ormas) to a slower growth (like mine) and fit a curve that models the Elo systems behavior.

What do you think?

I'll be honest, you're far out of my league! I don't tend to use statistics too much so I cannot comment on how it'd work. It took me a while just to find out how to so what I did :/

And if you're not a professor now, sounds like you will be! I'm currently earning my masters, but I'm unsure if I'll be going beyond that in my education.

I'd be very curious to hear what ypour dissertation is about, but it should be best saved for a different topic (Don't want to drive this one off topic). Hit me up on psn sometime! Psn: pie4july

 

[Alpha Cipher]

Im more like a case study of the Dunning-Kruger effect

You have enough self-awareness about your mental ability to label yourself that, so you may not be a case study of that at all
Something like how you know a person isn't insane if he can label himself insane
--
I want to comment on the formulas to keep this post relevant to the topic... But I don't understand anything

 

[John Crellin]

I'll be honest, you're far out of my league! I don't tend to use statistics too much so I cannot comment on how it'd work. It took me a while just to find out how to so what I did :/

And if you're not a professor now, sounds like you will be! I'm currently earning my masters, but I'm unsure if I'll be going beyond that in my education.

I'd be very curious to hear what ypour dissertation is about, but it should be best saved for a different topic (Don't want to drive this one off topic). Hit me up on psn sometime! Psn: pie4july

They way the grant system is set up, and the rate of current funding, I doubt I'll head towards a career in academia
And I doubt I could keep most people from snoozing through the title of my dissertation

For sure. I'll hit you up on PSN. My PSN:FluffaRaptor

You have enough self-awareness about your mental ability to label yourself that, so you may not be a case study of that at all
Something like how you know a person isn't insane if he can label himself insane
--
I want to comment on the formulas to keep this post relevant to the topic... But I don't understand anything

Don't worry. I'm sure a real statistician would see this and laugh me out of the thread.

In fact since your taking stats, your probably more of an authority on this topic than me with it so fresh in your mind

 

[Alpha Cipher]

In fact since your taking stats, your probably more of an authority on this topic than me with it so fresh in your mind

Haha, not really, since I broke one of the most basic rules of stat when I plugged my ratings in (Must have a large enough sample size)
And I think the calculation of statistics here is more complicated than the lessons I'm taking (Intro to inferential statistics or something)

 

[John Crellin]

Haha, not really, since I broke one of the most basic rules of stat when I plugged my ratings in (Must have a large enough sample size)
And I think the calculation of statistics here is more complicated than the lessons I'm taking (Intro to inferential statistics or something)

Eh. I doubt it. Basically what I'm getting at is that the DR progression is non linear so we could probably model it in two ways:

an s- curve which has a slow start picks up pace and then slows again making an s- shape on the graph
Or an arc which has a fast growth that slows as it approaches a maximum ( 75k DR)

These are both functions that Excel is capable of modeling which is by no means a high powered statistical software lol

 

[DriftMethod]

Something like how you know a person isn't insane if he can label himself insane

Uncommonly known as the Heller effect.

 

[zzz_pt]

DR doesn't work like that. It's perfectly possible you stop improving at any given point regardless of number of races you finish.

 

[John Crellin]

DR doesn't work like that. It's perfectly possible you stop improving at any given point regardless of number of races you finish.

Yeah, any prediction would have to come with the warning: asymptotes are individually based and the model cannot account for you utter lack of talent

 

[Sven Jurgens]

What can you predict from the data of my last 630 races

That big drop in the middle was intentional though, back to DR.B for a bit until the penalties became a little bit more reasonable.

The higher you get, the less points to earn per race, the more to lose and the more races it takes to 'repair' a single disconnect.

 

[vxr58]

Interesting. I used to think that I had a ceiling, where I would infinitely plateau. But as I keep playing, I keep improving and the plateau never persists.

The issue most likely to skew progress is that over time there will be fewer and fewer new players, increasing the percentage of players who are experienced and with high DR. That will make the game far more competitive.

 

[Nuschel01]

The issue most likely to skew progress is that over time there will be fewer and fewer new players, increasing the percentage of players who are experienced and with high DR. That will make the game far more competitive.

I'm thinking(/hoping) that more and more players will leave the game some day once other (better?) games are released

 

[simsimsheree]

I believe Samuel Clemens' opinion on statistics is most relevant: “There are three kinds of lies: lies, damned lies, and statistics.” Twain himself denied inventing this quote, and claimed Benjamin Disraeli was the one who created it. It is thanks to Twain, however, that the saying became popular in the U.S.

I'm thinking(/hoping) that more and more players will leave the game some day once other (better?) games are released

I am more of the opinion that more players will leave GTS once a better bashfest comes out! The remastering of Burnout Paradise for PS4 should definitely thin the lower ranks...

 

[phil_75]

Nice work and if you put that much dedication into your driving, I'm predicting that you'll get to DR S quickly.

A couple things I want to mention:

I'm not sure if you mentioned a constant number of races per day. That will definitely make a difference in any prediction like this.

The way DR points are calculated have been in your favor up until now. But soon it will not be in your favor. As you get to high A and S, most races will have few other drivers at your level and more at a lower level. So your points per race will be smaller and progress will slow. At the lower levels, getting a thousand or more points in one race isn't unheard of with a good result or a win. A win for a DR S rated driver is usually a couple hundred points if there aren't a lot of DR S drivers in the race. But one bad race can cost you a thousand points at that level.

I got just 8 DR points for winning a Daily C because the rest of the field were mostly DR D. (I am DR S)

I lost 13,500 DR points and 65 SR points in one single race where i got rammed then caught up in a pack of bumper cars which i ended up rage quitting .

 

[Famine]

DR doesn't work like that. It's perfectly possible you stop improving at any given point regardless of number of races you finish.

Among many other reasons why this is a very weird endeavour.

Firstly, DR is capped. This "predictor" will always break the cap, given enough time.
Secondly, everyone is apparently going up all the time and that's not possible. The sum of all DRs in any one lobby is zero - anyone who goes up in ranking must cause a drop in ranking for other drivers. There are only three ways to gain (or lose) DR independently of race results. They are ending your probationary period (applies once), changing ranking (applies every time) and SR-linked DR reset (applies every time). None of these are really applicable to long term trends, and the last requires you to drop at least two levels in SR lower than your DR - you essentially need to have already won the DR points (although not necessarily in every case)
Thirdly, the higher you go in rankings, the less the potential reward. If you're a D-ranked driver you can beat people at your level for some points, or beat people at a higher level than you for several more points. If you're an S-ranked driver, there's few - if any - above you to beat for more points. Point accumulation slows as your points score rises.


A linear "you've gone from 0 to some points in this time, so in more time you'll reach more points" doesn't cut it.

 

[John Crellin]

Among many other reasons why this is a very weird endeavour.

Firstly, DR is capped. This "predictor" will always break the cap, given enough time.
Secondly, everyone is apparently going up all the time and that's not possible. The sum of all DRs in any one lobby is zero - anyone who goes up in ranking must cause a drop in ranking for other drivers. There are only three ways to gain (or lose) DR independently of race results. They are ending your probationary period (applies once), changing ranking (applies every time) and SR-linked DR reset (applies every time). None of these are really applicable to long term trends, and the last requires you to drop at least two levels in SR lower than your DR - you essentially need to have already won the DR points (although not necessarily in every case)
Thirdly, the higher you go in rankings, the less the potential reward. If you're a D-ranked driver you can beat people at your level for some points, or beat people at a higher level than you for several more points. If you're an S-ranked driver, there's few - if any - above you to beat for more points. Point accumulation slows as your points score rises.


A linear "you've gone from 0 to some points in this time, so in more time you'll reach more points" doesn't cut it.

That's why you need a non linear model. But you are correct, this prediction would have to work off of a basic set of assumptions (like DR always trends up to some limit) which limits it's usefulness. The wording would work like this "based on your previous rate of improvement and assuming you were to continue improving, the model predicts that you CAN be ___DR in ___days"

But yeah. That is where it would end. It'd be a prediction of potential and it would be wrong in a lot of cases.

What can you predict from the data of my last 630 races

That big drop in the middle was intentional though, back to DR.B for a bit until the penalties became a little bit more reasonable.

The higher you get, the less points to earn per race, the more to lose and the more races it takes to 'repair' a single disconnect.

Technically your data shows a positive trend suggesting you may be still on the rise but your relatively lower slope suggests you may be nearing a plateau. But who knows... Lol I don't think past trends or trends from other accounts are sufficiently representative as you have never been in such a high level so we can't know how you will respond to higher DR levels. Scientifically we need more observations.
Unless we have data from another acct you haven't told us about where you are Dr S

 

[DriftMethod]

I got just 8 DR points for winning a Daily C because the rest of the field were mostly DR D. (I am DR S)

I lost 13,500 DR points and 65 SR points in one single race where i got rammed then caught up in a pack of bumper cars which i ended up rage quitting .

I'm guessing the 13,500 DR loss is a point reset and you could get it back quickly when your SR goes back up.

The 8 point win is ridiculous, no matter who you are driving against. Were you matched with low DR drivers because your SR was low? I think a win should give a minimum of 100 points.

If it's any consolation, a lot of us know from experience that you are a fast driver. Speed and on track manners are more important than these artificial points ratings. They can take away your points, but they can't take away your skill!

 

[Robben]

I'm guessing the 13,500 DR loss is a point reset and you could get it back quickly when your SR goes back up.

The 8 point win is ridiculous, no matter who you are driving against. Were you matched with low DR drivers because your SR was low? I think a win should give a minimum of 100 points.

If it's any consolation, a lot of us know from experience that you are a fast driver. Speed and on track manners are more important than these artificial points ratings. They can take away your points, but they can't take away your skill!

DR points are exchanged between the competitors in a race.. you can't magic up 100 DR points as a token minimum as they have to come from somewhere. If you are 50000 DR and your competitors are 10000 for example you will win:

(10000-50000)/500 + 80 = 0.08

So, at this level if there are 20 opponents you will win a massive 1.6 points. To get to 8 points, we can calculate that the average opposition have 36000 less DR points than @phil_75

 

[Racin-Rev]

Among many other reasons why this is a very weird endeavour.

Firstly, DR is capped. This "predictor" will always break the cap, given enough time.
Secondly, everyone is apparently going up all the time and that's not possible. The sum of all DRs in any one lobby is zero - anyone who goes up in ranking must cause a drop in ranking for other drivers. There are only three ways to gain (or lose) DR independently of race results. They are ending your probationary period (applies once), changing ranking (applies every time) and SR-linked DR reset (applies every time). None of these are really applicable to long term trends, and the last requires you to drop at least two levels in SR lower than your DR - you essentially need to have already won the DR points (although not necessarily in every case)
Thirdly, the higher you go in rankings, the less the potential reward. If you're a D-ranked driver you can beat people at your level for some points, or beat people at a higher level than you for several more points. If you're an S-ranked driver, there's few - if any - above you to beat for more points. Point accumulation slows as your points score rises.

A linear "you've gone from 0 to some points in this time, so in more time you'll reach more points" doesn't cut it.

Famine has it locked down. Rarely does theoretical probability live in the real world after you leave the academic research environment. The OP is extremely intelligent and a deep thinker. However, his situation creates what is known in America as the "Lake Wobegon Effect" with the statement "Lake Wobegon where all the women are strong, all the men are good-looking, and all the children are above average" The statistics just aren't possible.

 

[DriftMethod]

DR points are exchanged between the competitors in a race.. you can't magic up 100 DR points as a token minimum as they have to come from somewhere. If you are 50000 DR and your competitors are 10000 for example you will win:

(10000-50000)/500 + 80 = 0.08

So, at this level if there are 20 opponents you will win a massive 1.6 points. To get to 8 points, we can calculate that the average opposition have 36000 less DR points than @phil_75

Well, since this is a game, of course PD can just assign a minimum point value for a win. Remember there are drivers with the maximum of 75,000 DR points, so what happens when they win a race? Are they excluded from points calculations? Does the formula change to account for them not gaining any points? Do the points disappear?

I have no idea of what actually happens and it doesn't matter, because the points are mainly for matching. I have no delusion that PD is obligated to follow some "law" in fear of fracturing the SR/DR points continuum.

 

[John Crellin]

Famine has it locked down. Rarely does theoretical probability live in the real world after you leave the academic research environment. The OP is extremely intelligent and a deep thinker. However, his situation creates what is known in America as the "Lake Wobegon Effect" with the statement "Lake Wobegon where all the women are strong, all the men are good-looking, and all the children are above average" The statistics just aren't possible.

Statistics apply to large sets of data points, it models the behavior of a group not an individual. For instance, we have all heard the line " the avearge american has 2.5 children, 1 breast, and 1 testicle"
This tells you something about the group but not a single individual in the group. On the surface this information seems useless, but if I were to ask you what the probability is that an individual will have a child, you would say relatively high, if i ask what is the probability that the individual is male, you would say 50% based on the statistics. So while you cant necessarily accurately predict an individuals behavior, you can model a groups behavior and make predictions about the trends in a group.

 

[TheGeologist]

Famine has it locked down. Rarely does theoretical probability live in the real world after you leave the academic research environment. The OP is extremely intelligent and a deep thinker. However, his situation creates what is known in America as the "Lake Wobegon Effect" with the statement "Lake Wobegon where all the women are strong, all the men are good-looking, and all the children are above average" The statistics just aren't possible.

Why thank you , and Famine is correct to a point. As I've stated, this isn't an end all, tell all method. It was more just for fun, but I love how serious everyone is taking this. It's super entertaining. This is my new favorite topic

Statistics apply to large sets of data points, it models the behavior of a group not an individual. For instance, we have all heard the line " the avearge american has 2.5 children, 1 breast, and 1 testicle"
This tells you something about the group but not a single individual in the group. On the surface this information seems useless, but if I were to ask you what the probability is that an individual will have a child, you would say relatively high, if i ask what is the probability that the individual is male, you would say 50% based on the statistics. So while you cant necessarily accurately predict an individuals behavior, you can model a groups behavior and make predictions about the trends in a group.

Guys, we need to all group together and create an incredibly complex model and publish it as scientific data

 

[Alpha Cipher]

Why thank you , and Famine is correct to a point. As I've stated, this isn't an end all, tell all method. It was more just for fun, but I love how serious everyone is taking this. It's super entertaining. This is my new favorite topic



Guys, we need to all group together and create an incredibly complex model and publish it as scientific data

And give it to jasguer or Milouse for them to use on their websites

 

[Macboyilija]

Great work and effort @TheGeologist ! As you said it was not so serious but more for the fun.

Love to see people doing these kind of things and share it with us.

 

[simsimsheree]

I'm not sure how much 'fun' a statistical model that has no real relationship with real data and incorrectly predicts things that cannot happen is...

About as much fun as a model that predicts when elephants will finally design a rocket to the moon would be? LOL

 

[John Crellin]

I'm not sure how much 'fun' a statistical model that has no real relationship with real data and incorrectly predicts things that cannot happen is...

About as much fun as a model that predicts when elephants will finally design a rocket to the moon would be? LOL

You're always such a downer... You ever have anything constructive to add?

 

[Alpha Cipher]

I'm not sure how much 'fun' a statistical model that has no real relationship with real data and incorrectly predicts things that cannot happen is...

About as much fun as a model that predicts when elephants will finally design a rocket to the moon would be? LOL

shoo it's fun for me

 

[Macboyilija]

I'm not sure how much 'fun' a statistical model that has no real relationship with real data and incorrectly predicts things that cannot happen is...

Even if he tried to be accurate and it turned out to be pretty useless, he did make some real effort. And he explained some things about database and connecting stats from the game with Excel and explaining it. A lot to learn from. For us, and also from him based on the given answers.

Its easy to be negative. Fun can be explained in many ways.