Numpty >> sunday november 27 - 15:45

I've been considering making an effort to try and figure out the player star value formulae. There is a different one for each position. 

It has little practical value other than perhaps being useful in knowing how best to train players for farming purposes. But I think it would be quite cool to be able to work them out. Or at least get fairly close.

It will probably take a lot of effort at some point as they are fairly complex non-linear equations with around 10 or so different variables and also uses data rounding which adds significantly to the difficulty.

What I could use some help with is collecting player data for later analysis. This will be a fairly long term project mostly involve saving up loads of xp on some players before spending it all in one go, or alternatively using position training to release xp that's already been spent on older players before retraining them. It may also be useful if someone has a second team where they don't mind wasting a few players to help out with the research although this is not essential. 

Please send me a PM if you are interested in helping. Or ask questions on here if you want more detail.

baggini >> tuesday july 18 - 12:24

Hi numpty, this is very old thread, but I didn't find any other with more information. Did you do some analyses of data and did you find anything useful so far?

Numpty >> tuesday july 18 - 14:22, Edited tuesday july 18 - 14:23

I've mainly been collecting data into a spreadsheet. 

I've done a little bit of analysis on a few of the variables but no definite results yet, just enough to give me a good idea how some of them work. 

There are 4 different algorithms, one for each playing position. I use the term algorithm rather than formula because at least some of the variables contribute differently depending on it's value. Essentially this makes each algorithm a set of non-contiguous formulae.

Each algorithm appears to use 8 different variables - 9 if you include the age factor -  and most if not all of them are non-linear.   I believe the variables are:

• Power, Speed  
• 5 Trainables attributes (Base values)
• Position xp relevant to the position
• [Age factor needed to convert base values into effective values]

So that makes 4 sets of formulae each with 8 non-linear variables. 

The star value data being collected is rounded. So reverse engineering the formulae from rounded data adds an extra level of difficulty. In addition it may be the case that the source variable data is itself rounded before being fed into each formula. 

Solving that kind of problem - or rather a set of related problems - is going to be very tough indeed. Given the complexity it's going to be very challenging to get any truly accurate results, although I may be able to get a good approximation at some point.

What I need is more data and someone with a PhD in mathematical modelling. 

Any help appreciated.

onuelver >> tuesday july 18 - 22:50

Maybe talent counts too? 

Numpty >> wednesday july 19 - 07:15

What makes you think that?