Asunto: »»Youth reform: Changes in youth academies / junior traini

As for talent between 3 and 6, with 4.5 median

Talents from my current YS [not counting 0-4 weeks youths]

4.75
3.74
6.57
5.60
3.85
4.36
5.91
12.00
6.50

But as Mikoos said, these are "the best" I had in xx weeks, worse ones were deleted already

As a side note, yes, I'm extremely lucky with my youth school lately... I have 12 spots taken only because in nearest 2 weeks 4 of my youths go out (to the trash probably) and I need MIN 11 to stay in Youth League. If it wasn't for that I'd have 3/30 spots taken in my youth school, bloody hell...

Thanks for sharing

Nice to read

1. Sure it's another bias, but what's your point ? Like i said its nearly impossible to confirm at 100 % talent using stat tools, that's already the main bias.

(NB fo skT DB : (1) talent 1 0, NULL or 5 means unknown talent, so not necessary bad talent. (2) "talent" means "talent max" in DB, so it could be better)- or not (3) the algorithm never calculate talent above 5 (4) even if a coach sack a junior, records can stay in DB for a season)

2. i know those cases. But if the range was 3 to 30, talent 16 will be "standard" in normal distribution right ? I can't explain those cases, but for sure most of juniors are not talent 16.

3. As i said, the main bias is stats tool. I dont know how Terrion deal with that, certainly sort of linear regression, but we don't know how prediction is accurate for each talents ( 'coefficient of determination' for linear reg).

When almost half of my juniors show talent >6

But are you sure ? Let's say you have a junior at real talent 6. It should pop twice in 12 weeks right ? Or it could pop in 5 weeks then 7 weeks depend on starting level. So, to be accurate in prediction, we need at least 3x6 18 weeks to be "almost sure". No stat tools can't do better with accuracy.

And that's the point with stats : it need much more weeks of data if talent is higher to confirm something. So, most of those talents "above 5" are mostly unpredictable, it could be 30, or maybe 4 ! And that, it's a huge bias !

So its come to your observation :

"let's say we magically know their exact talents"

Ok it's magic, but in the end, you calculate and compare two senior players having 2 different junior's talent, right ? But what if junior talents are not what you expect they are ? If talents are wrong from start, you can't conclude anything in the end.
(editado)

That's a particularly good example of boring talent :-D

Talent is calculate to 12 but at this point its exactly the same chance of having talent 4.5 or 5 or 12 or 50... The algorithm calculate talent 12, just because it "seems not to pop" in 12 weeks, just that.

But let's said the real progression is 5 5 6 6 6 6 7 7 7 77... the best talent could be 4.5 for now.

The biggest flaw in those calculations is the fact you're never sure about the starting level. It can be wrong up to 4 levels! So Borkos' player could actually came into the youth school as level 2 and having a top talent, which it doesn't reflect up to now, because of the maybe, presumably, wrongly made assumptions of the youth coach.

Hi guys.

As you have the data, I have a small comment, don't know if your tools could catch this

When the player is finally promoted and the skills are finally known, it would be amazing to use this as final level and let's force to 0 uncertainty

Assuming the youth level is skill_sum/3.2 and the uncertainty is 8 levels (+1 in each skill) then the final level could be used as (skill_sum/3.2) + 1.25.

1.25 because the uncertainty is 0 - 8 levels in the skill sum.. then using 4/3.2

Well... This number could be adjusted as desired but my point is to ground a final level, then the tops and the bottoms of the channels (min and max level by week) might be adjusted

Don't know if your tools could handle this but would reduce by a lot uncertainty

First thing first, as I said, If talent isn't reliable enough at its base, the result will be even less so.
In SkTables only 100% of accuracy can give you one particular talent, otherwise it's a range of min/max talent (the better the accuracy the shorter will be the range). 100% it is rare, so we have to deal with a range of talents. Most of juniors didn't even reach 80% ( minimum % for good accuracy).

About your idea, It might interest a very few users for a very few "realiable talent", so... I will not spend my time for a tool which won't be use so often. But off course, talents are registered, then users are free to calculate sumskills by themself.

A little story : Last week, base on talent s records in sktables, I predict few random skills for a promoted player for its next senior training, because 1) before promoted this junior have to level up next week in school according to SkT records 2) prediction was base on 95% of accuracy. The next week he pop 4 random skills in senior team.
(editado)

yeah, not sure if it would be used for many users as usually you are interested in projections and level when promoting, just something extra

cheers

Sure, but statistics can correct the starting level. The problem is that it takes a lot more data (weeks) to correct false level assessments. And it takes even longer to calculate if a junior is not only misjudged by the coach, but also has a real talent of 5 or more. So, those stats tools usually fail to give a talent in those cases.

Sktables algorithm stop to look forward talent 5 (after level 5, we don't care it just a weak talent anyway) and it correct/smooth 3 or more level up/down a week. It can help a lot in some cases to correct starting level more quickly, but it's not a magic bullet either.

There are also weird things for a few minority of juniors : it seems to never level up, even in 20-30 weeks. It make no sense in relation to the well known model (linear progression) and no statistical tools can help.

But on the other hand, statistical tools are great when it comes to estimating a junior who is progressing rapidly (3-4 weeks) without too many double or triple ups, starting level is easily corrected ! And that's the main thing: juniors with best talent are very easy to estimate/predict with good level of certitude.
(editado)

I plea for an adjustment to the juniors coming into the youth squad. THAT level should always be 100% correct! The evolution can be based on estimations of your coach and should on their turn be less or more accurate, based on the level of your coach.
An unearthly coach should always give a max error of 1 level, where a magical or brilliant coach should have a max 2 error and so on...
Imo, starting off with presumably wrong information, even though the tools ''correct'' the level, is just plain stupid. So the tool always assumes the starting level was correct because, how can you build a tool in which you don't know where to start from? Yet, the error margin is 4 levels! That's just too massive to start with. Imo, regardless of each algoritm or calculation, youth ''talents'' are pointless to build on in the current case of estimations throughout the entire road within the youth
(editado)

There is no such thing in Dev goals (except maybe a "Junior analyst position") don't except anything more "accurate". Plus, sokker is very linear and predictable in term of player progression, devs introduce a "trick" (coach errors) for a "good reason" there's no way they're going to go back on that.

So the tool always assumes the starting level was correct because, how can you build a tool in which you don't know where to start from?

I understand you can't understand, but that is why stats calculations are used ! With limits I exposed previously.

The error is from -2 to +2.... thats why it can have a difference of 4... so when you suggest 1 or 2, I think you misjudge why they can change 4 levels ... a 4 level jump/drop should though actually help in evalute, because you can exclude things with it
(editado)

---

---

---