Player development makes zero sense to me in this game.

[Speedyol]

1 hour ago, francis#17 said:

A larger sample would actually mean that we would be able to see how players CA changes on average for all age groups in the game (instead of only testing how well you can develop players), while taking into account all these game world variables that will affect development. When the sample size is 1 then we have no idea if you were just unlucky because of the game world variables.

Anyway as you say its a chore for one person to do this. So far I havent had any issues developing 20+ players in low reputation leagues though so I dont think you are right about them not developing.

You could do it retro active by saving every year, simlulate 10-20 years and find the top 100? players in champions league. then load back previous years/saves to see where they started and how they developed. idk. I dont even have the tool to see the hidden CA or PA values you are talking about so I cant do it.

[goranm]

10 hours ago, enigmatic said:

huh, where did I say that? ah I didn't say that

What I did say is tactical instructions to play simple passes would tend to dramatically increase pass completion a player with a particular Passing attribute achieves in the ME. An example of the effect of tactics on ME outcomes, which you were keen to solicit my opinions on

Here: "Players might well achieve dramatically higher pass completion from being instructed to play simple passes, but it would be absurd for this to increase the attribute which improves their range of passing." Here "This" being simple passing leading to higher pass completion, so just extrapolate what that means if no risky passes are attempted. No one is suggesting that buffing the completion rate with simple passes should lead to an increase in the attribute which improves the range of passing. 

10 hours ago, enigmatic said:

huh, where did I say that? ah I didn't say that,

Here: "but clearly it would be absurd for playing mostly against midgets to improve the attribute which makes them better at soaring high into the air to head the ball away." Again, no one is suggesting that just completing easy headers should lead to an increase in jumping or strength or heading.

10 hours ago, enigmatic said:

Personally, I'd rather SI stuck to more sensible assumptions of assuming that Jumping would change largely in response to height, physical conditioning and dedicated training.

No one is arguing that jumping or physicals governed by genetics would change with match performance, unless injury is involved.

10 hours ago, enigmatic said:

And I believe you were also keen to see Ronaldo-esque late career jumps in long shot accuracy?

The Ronaldo example wasn't about his long shots, it was about his overall development between ages 23 and 30. Long shots were just one the attributes (and the most extreme one) in which he improved greatly between ages 23 and 30, which is something the game is not reflecting.

10 hours ago, enigmatic said:

how do you get either of those from average rating 7.32? 

Why would you look at average rating only? Where am I saying that? If you have a player who is training long shots, to whom you give creative freedom to do long shots, and who then attempts long shots in actual match environment and not just the training ground, the development system should account for how well he is actually using that skill. Training means nothing if a player can't convert it to match performance.

[Platinum]

37 minutes ago, goranm said:

The Ronaldo example wasn't about his long shots, it was about his overall development between ages 23 and 30. Long shots were just one the attributes (and the most extreme one) in which he improved greatly between ages 23 and 30, which is something the game is not reflecting.

Why do you think players cant make big improvements between the ages 23 to 30 and how long do you think this has been an issue for?

My understanding is that this is not true and players can carry on having big attribute increases between 23 to 30 (even if it is less than there teenage years it can still be a significant amount that makes the player a much better player, for example there is the FM19 screenshot on the 2nd page).

[goranm]

1 hour ago, francis#17 said:

Why do you think players cant make big improvements between the ages 23 to 30 and how long do you think this has been an issue for?

Player development has been an issue for a long time. Great strides have been made to improve it, however currently I'm talking about FM21, which is about the same as FM20. FM19 I did not play much - usually I play until newgens completely take over, but with players peaking at 23/24 this usually means that I know how the game world will look like after 7 or so seasons for the next 10 seasons due to predictability of development.

Most players at high level and with decent training facilities are already close to PA at ages 23/24, leaving little space for development after that age. This I've confirmed with the in-game editor in FM20, but don't want to do with FM21 yet as I don't want to spoil my game. I'm 9 seasons in in FM21 and haven't seen anything suggesting that development differs from FM20.

1 hour ago, francis#17 said:

My understanding is that this is not true and players can carry on having big attribute increases between 23 to 30 (even if it is less than there teenage years it can still be a significant amount that makes the player a much better player, for example there is the FM19 screenshot on the 2nd page).

From what I see in your screenshot, your player started as a league two player, one season as a league one player, then 5 seasons in championship and only then he starts playing at the highest level, at age 27. Match experience at higher levels will contribute more to development - had this player started playing regularly at the highest level at 19-23, you'd maybe see his development maxed out at 23/24ish. This is why one player doesn't really prove or disprove anything.

Here is an example @francis#17 : I won't say who the player is to not spoil it for others, but here are his attribute changes since age 22, age 23 and age 24. The player is now 28, will turn 29 at the end of the season. You can get this under Training - Development - Compare Attribute Changes Since Age X. I'm doing mentals only since there was almost no changes in technicals or physicals.

Now as you can see since age 22: major increase in Vision, Positioning, OTB, Anticipation, large increases in other attributes (straight up arrow).

Since age 23: slight increases, only OTB and Leadership increased by 1 (OTB was actually increased by 0.6-0.8), only other attribute with a straight up arrow is Vision

Age 24: Only Leadership has a straight up arrow, most arrows are yellow.

The last graphs shows this even more clearly: from ages 21 to 24.5 there is a lot of development and then stagnation. The stagnation occurs actually closer to 23, since only anticipation consistently increases (and leadership, the violet line around the middle), and other values fluctuate by what looks like 0.4-0.6 points with work rate jumping in increments of 1 (it used to be that we could see attributes to decimal places, if this can be enabled, I don't see where).

 

Edited December 31, 2020 by goranm

[enigmatic]

So let me get this straight, @goranm you think attributes should update based on outputs from the match engine to incorporate the effect of tactics and opponent quality on player development, but also you are definitely not saying that the effects of tactics on pass completion or opponent quality on challenges should be incorporated into player development. 

That's... a little strange, but let's discuss the attribute you do say you want updated recursively based on match engine outputs: Long shots. Each shot is a function of existing long shot skill which is the input you wish to 'correct', plus other factors like where they're shooting from and whether they're under pressure, goalkeeper quality, random chance.  Which of these perturbations to a player's long shot accuracy for a given shot or season's worth of shots is actually linked to the long term development of a player's shooting skill? Like the examples above, what actually improves ME shooting accuracy relative to a player's Long Shots is stuff associated with the chance being less testing to a player's skill, plus luck. 

Seems far more plausible that a player with Long Shots 11 enhances his  ability to shoot from distance in future games from spending hours taking shots in training from a variety of distances rather than rare match events like one deflecting in from 19 yards to double his long shot goals and conversion rate for the season.

 

 

Edited December 31, 2020 by enigmatic

[goranm]

2 minutes ago, enigmatic said:

you think attributes should update based on outputs from the match engine to incorporate the effect of tactics and opponent quality on player development, but also you are definitely not saying that the effects of tactics on pass completion or opponent quality on challenges should be incorporated into player development.

Lol no, I'm saying different things carry different weights. I can even quote myself saying literally that: "Completing a short simple pass would carry much less weight than completing a difficult pass." For someone calling other people thick, you really are bad at reading.

10 minutes ago, enigmatic said:

That's... a little strange, but let's discuss the attribute you do say you want updated recursively based on match engine outputs: Long shots. Each shot is a function of existing long shot skill which is the input you wish to 'correct', plus other factors like where they're shooting from and whether they're under pressure, goalkeeper quality, random chance.  Which of these perturbations to a player's long shot accuracy for a given shot or season's worth if shots actually linked to the long term development of a player's shooting skill? Like the examples above, what actually improves ME shooting accuracy relative to a player's Long Shots is stuff associated with the chance being less testing to a player's skill, plus luck. 

I don't know, if I had a development model thought out to the level of detail you're expecting of me, I wouldn't be telling you anyway, I'd make a game. Why not all of them? The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute. The attributes, after all, are what a player can do in a match, not what they can do in training. If a player uses long shots in a match succesfully at an above average capacity (relative to his previous ability), he gets a development bonus for the long shot attribute.

8 minutes ago, enigmatic said:

Seems far more plausible that a player with Long Shots 11 enhances his  ability to shoot from distance in future games from spending hours taking shots in training from a variety of distances rather than rare match events like one deflecting in from 19 yards to double his long shot goals and conversion rate for the season.

It does, but that's your statement, not mine. You are diluting what I'm saying down to non-sense. A player attempting 2 long shots and scoring 1 over the course of a season would obviously not carry much weight in the development of the long shot attribute.

[enigmatic]

2 minutes ago, goranm said:

Lol no, I'm saying different things carry different weights. I can even quote myself saying literally that: "Completing a short simple pass would carry much less weight than completing a difficult pass." For someone calling other people thick, you really are bad at reading.

Yes, I am are talking about how weighting a short simple pass differently in the match engine outputs you wish to use for development affects whether or not the tactics affect development. For someone that consistently claims he is being misinterpreted, you are really bad at reading.

Pass completion in the ME (both the probability of an individual pass and the resulting percentages for all passes made) is a function of pass difficulty, player attributes and all other factors like luck,  with pass completion being inversely related to pass difficulty. The detail is complicated, the basic direction of the functional relationship between pass completion and pass difficulty in the ME really isn't hard to grasp, especially not for somebody with an interest in ML

So if you weight the effect of the pass difficulty inputs on the pass completion outputs to negate the positive effects of pass simplicity on pass completion (e.g "completing a short simple pass would carry much less weight than completing a difficult pass") you're eliminating the effect of all pass difficulty factors on pass completion in your model which drives player development, including the effect of tactics on pass selection. Exactly as I said above.

If you weight pass difficulty in a way which doesn't eliminate the effect size of pass difficulty altogether, you do incorporate an effect of tactics on match performance into your development model, but in a way which either rewards less difficult passing circumstances or penalises superior completion rates. 

 

1 hour ago, goranm said:

I don't know, if I had a development model thought out to the level of detail you're expecting of me, I wouldn't be telling you anyway, I'd make a game. Why not all of them?

Why not all of them? Because, as we've been discussing for the last few posts, most of them are pure noise like wind, and others (like easier opportunities) appear to inversely related to a player's skill level and therefore have no logical relationship with future improvement in that skill.

 I keep giving examples and you keep telling me you mean something else. Seems reasonable for me to be asking what you do mean, but seriously, no obligation on you to continue this discussion.

Also entirely unnecessary for you to make a game, but I'd be very impressed if you did tbf  

 

14 minutes ago, goranm said:

The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute. The attributes, after all, are what a player can do in a match, not what they can do in training. If a player uses long shots in a match succesfully at an above average capacity (relative to his previous ability), he gets a development bonus for the long shot attribute.

ugh, we're back to explaining cause and effect again

Let me help you out here.

The match and the training aren't real. The players are just a set of numbers in a database. You don't simulate the effects of training long shots by asking the numbers to take lots of shots in the hope they will be able to deliver the performances of different numbers, you simulate it by incrementing the Long Shots number. That way the ME simulation gets the input it needs to allow the player to perform better long shots in future. Otherwise it doesn't (and any apparent improvement in long shooting actually comes from easier long shots to execute or luck). Therefore development bonuses in the form of attribute changes are the cause of better performance in the match engine, and it makes little sense to suggest they should also be an effect, and no sense to believe they should only be an effect.

 

14 minutes ago, goranm said:

It does, but that's your statement, not mine. You are diluting what I'm saying down to non-sense. A player attempting 2 long shots and scoring 1 over the course of a season would obviously not carry much weight in the development of the long shot attribute.

A player who attempted 50 long shots, scoring two would be entirely compatible with my example (a goal doubling his long shot goals and conversion rate implies  two goals from an indeterminate number of chances) and fairly typical for an offensive player with unremarkable long range shooting.

 

I thought your original desire was to see more attribute changes, but now you're insisting that players shouldn't be able to develop their long shots attribute in training, only through match performances, and a match performance involving a rare long shot goal 'obviously' shouldn't have much impact . Sounds like most players will never achieve the sort of performances that will allow them to improve their long shots   

 

 

 

Oh, and happy new year. Let's hope we're not still going in 2022.

[FlorianAlbert9]

Il 29/12/2020 in 01:03 , enigmatic ha scritto:

No, if you set a player with 20 for pace in the editor he can have 20 for pace regardless of what his age and ability is. But CA is a weighted sum of all the attributes, so if you set all the attributes to add up to more than the number the CA allows, they will all be reduced.

And pace is obviously an important component of footballing ability (IRL and even more in the FM ME) as anyone who watches a team with Salah and Mane up front should be able to easily see.

 

 

 

Pace is overestimated in the game. 

What players like Salah and Mane used mostly are "explosion" (plus of course, acc, agility and balance) that we dont' have in FM (or it's maybe calculated in some way i dont' know). 
Pace, The max speed a player can reach, it can be reached very very rare during a season. 

Even overweight, Ronaldo R9, was one of the fastest player, instead he lost explosion and agility that are more more important. 
 

[bielsadidnothingwrong]

5 hours ago, FlorianAlbert9 said:

explosion

is this not ACC? 

[Seancool]

Hey guys!

While I have been playing Football and Bundesliga Manager Games back in the 80s on the C64 and Amiga 500, I am somewhat new to SI Football Manager and all the depth of CA and PA, and I read this thread with much interest, and I wanna just ask you guys to check if I got this right.
I installed a skin/mod for the Hidden Attributes and got this guy on my second Team.

He is 20 years, so still very young! Thats good, right?
I would have thought he may be a great prospect but it turns out he has a PA of 103 and already is at 100. And that is..."bad"?

My Question:

Does this mean the PA is a hard cap for the CA and there literally and technically is no way all his skills and abilities combined would ever surpass a future 103 CA?

So all training and playing time and even future lub success would keep him at a second division / Zweite Bundesliga - Level? There is no point in "investing" any more time and (contract) money at him right?

 

 

Edited January 1, 2021 by Seancool

[FlorianAlbert9]

1 ora fa, bielsadidnothingwrong ha scritto:

is this not ACC? 

According to description, no. it's maybe similar but:
Acceleration is an attribute that depends on Pace and it is the amount of time(or space) to reach max speed. 
Explosion is the Instant pace at starting, that is the most important Factor for the most of the roles. 
20 in Acc what really means? 20 acc with 1 in Pace is better or worst of 1 acc with 20 in pace at the starting? We dont' know. 

[FlorianAlbert9]

2 minuti fa, Seancool ha scritto:

Hey guys!

While I have been playing Football and Bundesliga Manager Games back in the 80s on the C64 and Amiga 500, I am somewhat new to SI Football Manager and all the depth of CA and PA, and I read this thread with much interest, and I wanna just ask you guys to check if I got this right.
I installed a skin/mod for the Hidden Attributes and got this guy on my second Team.

He is 20 years, so still very young! Thats good, right?
I would have thought he may be a great prospect but it turns out he has a PA of 103 and already is at 100. And that is..."bad"?

My Question:

Does this mean the PA is a hard cap for the CA and there literally and technically is no way all his skills and abilities combined would ever surpass a future 103 CA?

So all training and playing time and even future lub success would keep him at a second division / Zweite Bundesliga - Level? There is no point in "investing" any more time and (contract) money at him right?

 

 

There are some very important attributes that don't cost in term of CA points: Consistency and important Matches. (and aggressive and flair too, but are less importat).
So he could become a player that perform at his max level more ofter than a player with 120 CA (but low consistency). 
But generally, you're right. When he get his peak he could perform well in Hungarian First Division at max, not in Bundesliga.

 

[Seancool]

6 minutes ago, FlorianAlbert9 said:

There are some very important attributes that don't cost in term of CA points: Consistency and important Matches. (and aggressive and flair too, but are less importat).
So he could become a player that perform at his max level more ofter than a player with 120 CA (but low consistency). 
But generally, you're right. When he get his peak he could perform well in Hungarian First Division at max, not in Bundesliga.

 

Oh dang, that's good to know. So he might outperform a "typical" 120 CA player at his position but he would never become as good or elite as players with top Bundesliga/Serie A/Premier League players around 155 to 175, right?

I have another question if you don't mind.

The next pic is a random new generated player in the game of 16 years. His market value is, well, zero.

His PA says 52 and no matter what anyone (me or the AI) would do, his current 29 overall rating will eventually face a roadblock of 52 CA at his early 20 years? 🤔 Thats like a what, 4th or 5th Division player?

Edited January 1, 2021 by Seancool

[Seancool]

And this guy here will be an elite player with 99 CA and 166 (!) PA one day and is one of the best prospects in the FM2021 game iirc?

Spoilertag for the pic link; dont click if you want to find him yourself

 

By the way "who" determined the 166 PA? The SI devs and researchers? 166 is just their educated "guess" and what/how they think this guy could peak? Sorry for all the newbie questions

Spoiler

 

Edited January 1, 2021 by Seancool

[XaW]

1 minute ago, Seancool said:

And this guy here will be an elite player with 99 CA and 166 (!) PA one day and is one of the best prospects in the FM2021 game iirc?

Spoilertag for the pic link; dont click if you want to find him yourself

 

By the way "who" determined the 166 PA? The SI devs and researchers? 166 is just their educated "guess" and why they think this guy could peak? Sorry for all the newbie questions

  Reveal hidden contents

 

Yes, 166 is very high and with a good attribute distribution could be one of the best players in the game.

The research team will try to estimate how highly he is rated from 0-200 in both current ability and maximum potential, but for younger players the potential is often set a "range" of say 160-180 and then each game will put the player within that range, but at a different value. This is due to estimating potential is very hard the younger the player.

I would also note that looking at CA or PA is not something you should do exclusively. The number can be misleading if the players have a poor distribution of attributes and/or poor stats for the "hidden attributes" such as professionalism and consistency.

[Seancool]

Just now, XaW said:

I would also note that looking at CA or PA is not something you should do exclusively. The number can be misleading if the players have a poor distribution of attributes and/or poor stats for the "hidden attributes" such as professionalism and consistency.

This is a VERY great advice and hint how to deal with this, thanks, did not think about how stats allocation could create two really different players with 160 CA. It's so obvious in hindsight but not intuitive if you are new to the FM series and mechanics. Thanks for helping

[XaW]

Just now, Seancool said:

This is a VERY great advice and hint how to deal with this, thanks, did not think about how stats allocation could create two really different players with 160 CA. It's so obvious in hindsight but not intuitive if you are new to the FM series and mechanics. Thanks for helping

No worries! It's the details that make this game so great!

[FlorianAlbert9]

Yes, about Niko Mohr. 

As above, maybe with great consistency and right attributes in the right tactic he could perform better of stronger players (with maybe 20 CA points more), but i dont' know if he could reach third division level. 

 

[FlorianAlbert9]

2 minuti fa, Seancool ha scritto:

This is a VERY great advice and hint how to deal with this, thanks, did not think about how stats allocation could create two really different players with 160 CA. It's so obvious in hindsight but not intuitive if you are new to the FM series and mechanics. Thanks for helping

For example: 2 striker with the same CA, one look to have better attributes overall, but one-footed, the other is ambidextrous. So i have no doubt and will pick up the the ambidextrous one 'cause striker is a position where player often doen't have the time to use the prefered foot. (while a for a defender is less important, imho)

[Seancool]

Just another thought, you guys discussed that at page 1 and now it explains so much: The AI usually "finds" the "best PA prospects" first if you don't use that hidden attributes mod like I do, right?

It somehwat spoils the exitcement to find the best young players myself but that's how I only can compete with the CPU/AI after all it seems?

So the result is that you can find the best 16-19 year olds with 100-150+ PA in the 3-5 best Leagues and the youth and amateur teams of the "big clubs"?

I just checked 1 german, 1 itlaian and 1 spanish top club and the under 20 players and second teams and youth teams and wtf do they have a cluster of 120-150+ PA talents, it is absurd.

These clubs are also the ones to buy/contract the new generated 100+ PA talents asap, right? or does the AI in this game "casually" and artificially "miss" these talents, I am speaking of the new generated 15-16 year olds every new season.

The only exceptions I found are "farm teams" ....

Spoiler

like Red Bull Salzburg, and him:

with this real good 17 year old 146 PA player from a "not top league".

[Seancool]

10 minutes ago, FlorianAlbert9 said:

 So i have no doubt and will pick up the the ambidextrous one

Another geat and in hindsight obvious hint, thanks, and shows how importan the little details are.

Just thinking about Bundesliga Manager C64 Sessions lol

How FAR have we come!

 

  

15 minutes ago, FlorianAlbert9 said:

Yes, about Niko Mohr. 

As above, maybe with great consistency and right attributes in the right tactic he could perform better of stronger players (with maybe 20 CA points more), but i dont' know if he could reach third division level. 

 

To make it clear there is NO way to "pass" the PA cap, right? The PA 80 Player could play in my fantasy cheated team with Ronaldo and Messi and Neymar in attacking and I win 30 games in a row and he gets the best coaches and individual training sessions, but he won't surpass that 80 CA after all, ever?

Edited January 1, 2021 by Seancool

[XaW]

1 minute ago, Seancool said:

or does the AI in this game "casually" and artificially "miss" these talents, I am speaking of the new generated 15-16 year olds every new season.

The AI clubs in the game does not see the CA or PA, they have access to something called Perceived Current Ability (PCA) and Perceived Potential Ability (PPA). That is based of a lot of things, both the skills of the scout who have scouted the player, but also based on how long they have scouted him, how much he has played, how high his reputation is (does he play for the U18s of a great team) and many more aspects. Based of that the AI team will have a PPA that can be very accurate or very off, but they will act on it in the same way as your scouts will advise you on how high the Ability or Potential is.

So a player could stay "hidden" for a while. In a previous game I played, I had a highly rated winger in Vanarama National in England and later on he got sold for almost £80m between two AI clubs. I don't know his true CA/PA since I don't use anything to show me it, but I had him at my club for 4 years before anyone tried to buy him off me...

[FlorianAlbert9]

AI should have your same tools. 

But you're, as human, more able to use those tools (example, you will use more scouts, better staff in general)

Scout don't see clearly PA but in some way a glimpse of it: in their reports use a mix of age/current ability/mentality plus a a range of PA. but they can make mistake.

[Sloak]

Outside of the obvious, hit CA early, play at ~PA levels until 30 and then drop off physically, I feel like how individual attributes develop could be changed.
In my mind, physicals would develop rapididly across the board from 15-2(1-3) depending on the player, and then outside of strength and stamina (I choose these as I feel stamina and strength can be worked up over time more than pace or jumping reach in a real life scenario) would be hard to increase more than marginally.

Mentals would essentially progress for a whole players career, but not equally. Anticipation, Decisions, Composure, and Vision stick out to me as things a player will either be born with or develop over a long time, not really something able to be worked on on the training pitch. I’m not sure about aggression, bravery, teamwork, and work rate, but I feel like they should be similar in that they’re not really specifically trainable and should just naturally grow or not grow overtime depending on teamates, the tactical system, and professionalism. Offball and Positioning should be trainable IMO, as they seem trainable in real life. Flair and concentration feel like they should be mostly static with small room for growth throughout a players career.

Technicals should also grow naturally throughout a players career, with some being static. The set piece attributes should be trainable obviously, and I feel that outside of finishing and technique the rest should be trainable too. I feel that technique and finishing should scale over a players career like mentals as technique acts, to my knowledge, as more of an overall boost rather than anything specific and finishing is obviously a talent some are born with and others are not. My reasoning for technique is that If I slap a player on passing training for his whole career, his development due to training will make him not only a better passer, but also indirectly a better crosser, finisher and dribbler, which doesn’t feel right to me. I don’t feel it’s realistic to turn an 8/20into a 13 or 14/20 in about 2-3 years, making a player a poor technical player in most leagues to an average to above average premier technical player.

I also don’t feel the three categories should progress off of the same criteria. In the current system, if you get a player more game time pre 21 than his peers, he will develop more physically than them, which doesn’t make logical sense. 2 twin 19 year old brothers will not develop physically differently if one of them is playing first team games while the other just trains, if anything, the one just training may experience more physical growth due to having to rest less. I feel that physicals should progress based off of professionalism, ambition, consistency, and maybe determination until the player hits the cut off. I feel that mentals, outside of Offball and Positioning, should only really grow based off of competition level and games at that level. This feels more realistic to me than just finding a young player and putting him on final third and attacking movement until he’s mentally solid. I believe technicals should grow based off both game time and training, finishing and technique mainly off game time and the rest more balanced. 
I feel that this system would make players grow more naturally and realistically. Players would grow into technical abilities and I feel it would reflect differences in coaching more. Players playing for lower reputation teams would still progress in their mental and technical attributes based off playing time, but would lack the ability to radically improve their Offball, Positioning, or any trainable technicals due to their coaching letting them down. 
I think this system would work just fine with the current CA and PA system, I just feel that overall development may need to be slowed down to allow it to work properly.

[goranm]

16 hours ago, enigmatic said:

A player who attempted 50 long shots, scoring two would be entirely compatible with my example (a goal doubling his long shot goals and conversion rate implies two goals from an indeterminate number of chances) and fairly typical for an offensive player with unremarkable long range shooting.

This is such a weird criterion to base your argument on, for two things. First, there is only one situation in which a goal doubling long shot goals and the conversion rate works: 1 goal is scored in a large number of attempts and then an additional one is scored. When you do the maths and solve the system g/s=r and (g+1)/(s+1)=2r, here g being goals before the next one is scored, s being shots and r being the ratio, you end up with g=s/(s+2). The only way for g to be approximately an integer is when s is large.

Secondly, you still keep insisting that when N pieces of information are given, only some of those N will be used, when no one is preventing all of them being used. If all N pieces of information are used (#goals, #shots and conversion rate) as opposed to only some of them (#goals and conversion rate), it's too easy to account for your example: if a player scores 1/999 long shots and the 1000th shot is a goal doubling his conversion rate, the fact that he took 1000 shots to double his conversion rate will not result in his LS attribute increasing. If anything, it would be decreasing.

Again, you're painting a caricature with what I'm saying by thinking of extreme worst-case scenarios which are easily accounted for anyway.

 

16 hours ago, enigmatic said:

I thought your original desire was to see more attribute changes, but now you're insisting that players shouldn't be able to develop their long shots attribute in training, only through match performances

Please quote where exactly I'm insisting on that. You won't be able to, because what I've actually written is: "The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute." Do you see how this is different from what you claim I'm saying? Again, please do a better job at reading what I've written.

17 hours ago, enigmatic said:

ugh, we're back to explaining cause and effect again Let me help you out here.

The match and the training aren't real. The players are just a set of numbers in a database. You don't simulate the effects of training long shots by asking the numbers to take lots of shots in the hope they will be able to deliver the performances of different numbers, you simulate it by incrementing the Long Shots number. That way the ME simulation gets the input it needs to allow the player to perform better long shots in future. Otherwise it doesn't (and any apparent improvement in long shooting actually comes from easier long shots to execute or luck).

No one is arguing otherwise, so I'm not sure what you're explaining or helping me out with here.

17 hours ago, enigmatic said:

Therefore development bonuses in the form of attribute changes are the cause of better performance in the match engine, and it makes little sense to suggest they should also be an effect, and no sense to believe they should only be an effect.

Well, no, the way development is encoded now is non-sense because it happens before the experience is undertaken. It's the other way, people learn and develop skills when an experience is reflected upon. No one became a better chef by memorizing recepies and cooking for themselves, they had to actually cook something in a restaurant environment with coworkers and customers, react when things go wrong and/or when a customer is demanding. Similarly in football, not everything is learnt in training, some things happen in a match environment only.

17 hours ago, enigmatic said:

Why not all of them? Because, as we've been discussing for the last few posts, most of them are pure noise like wind, and others (like easier opportunities) appear to inversely related to a player's skill level and therefore have no logical relationship with future improvement in that skill.

 I keep giving examples and you keep telling me you mean something else. Seems reasonable for me to be asking what you do mean, but seriously, no obligation on you to continue this discussion.

If I'm telling you "something else", the better thing to do is to quote me exactly, rather than conjuring a scenario to which you expect (as a figure of speech) a detailed answer. As with other things, if certain things are noise, code them as noise. If easier opportunities shouldn't impact future improvement, code them like that. Give them little to no weight.

As to your first four paragraphs, I really don't understand what you are talking about, nor how it relates to anything I've said. You originally wrote "So let me get this straight, @goranm you think attributes should ... " without ever quoting me. There you've put words in my mouth, and now continue to argue something that I might have not even said. If I have, quote me exactly with the problematic parts, similarly to how I am quoting you.  For example:

17 hours ago, enigmatic said:

So if you weight the effect of the pass difficulty inputs on the pass completion outputs to negate the positive effects of pass simplicity on pass completion (e.g "completing a short simple pass would carry much less weight than completing a difficult pass") you're eliminating the effect of all pass difficulty factors on pass completion in your model which drives player development, including the effect of tactics on pass selection. Exactly as I said above.

If you weight pass difficulty in a way which doesn't eliminate the effect size of pass difficulty altogether, you do incorporate an effect of tactics on match performance into your development model, but in a way which either rewards less difficult passing circumstances or penalises superior completion rates. 

There's no eliminating of the effect of all pass difficulty factors by weighing a pass. A pass is difficult or it isn't, how we weight its completion for development purposes doesn't affect that. We assign scores to each pass completed and you can do that however you want, it doesn't have to correlate with pass completion. Two players, both with the same X% pass completion rate under different tactical instructions can very well get different development scores if one was doing only simple passes but the other was not.

18 hours ago, enigmatic said:

Oh, and happy new year. Let's hope we're not still going in 2022.

Happy New Year! Let's hope not, but if we were to, there's nothing wrong with that. I have nothing against you and am not taking these arguments we're having personally or in a bad way.

[herne79]

2 hours ago, goranm said:

"The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute."

Players develop their attributes by having effective Training schedules and gaining match experience at a relevant level.  Both of these factors are important for player development, however for players aged 18 and over match experience takes precedence.  For player below this age Training takes precedence.  As you can see, increasing an attribute isn't really down to training "alone".  (SI have told us this is how it works, that's not just speculation).

2 hours ago, goranm said:

Well, no, the way development is encoded now is non-sense because it happens before the experience is undertaken.

I'm not sure where you're getting this from as it is not correct.

Development takes place as players play matches and their performance in those matches, not before.  Gains from individual matches will however tend to be negligible and thus we wouldn't notice them - we'd only start to notice gains after several matches have been played - but players don't develop "before" experience is gained.  Their development is based on their experience gained.

[Seancool]

So i just got the message from my assistant coach:

"Sven Ulreich only is an average Keeper in the Bundesliga" and it was "former good keeper in the Bundesliga".
Bust his CA matched is PA of 137 so he never was higher rated than now/today?
How can he be "worse" when "mathematically" he has the best case personal rating, like today at age 32!? 🤔
 
I don't understand that message. Does this mean a keeper with 137 CA only is an average keeper in the Bundesliga compared to other keepers in the Bundesliga?

So the reference is not is PA but the CA of similiar players in my Leagues?

Edited January 2, 2021 by Seancool

[enigmatic]

6 hours ago, goranm said:

This is such a weird criterion to base your argument on, for two things. First, there is only one situation in which a goal doubling long shot goals and the conversion rate works: 1 goal is scored in a large number of attempts and then an additional one is scored. When you do the maths and solve the system g/s=r and (g+1)/(s+1)=2r, here g being goals before the next one is scored, s being shots and r being the ratio, you end up with g=s/(s+2). The only way for g to be approximately an integer is when s is large.

Secondly, you still keep insisting that when N pieces of information are given, only some of those N will be used, when no one is preventing all of them being used. If all N pieces of information are used (#goals, #shots and conversion rate) as opposed to only some of them (#goals and conversion rate), it's too easy to account for your example: if a player scores 1/999 long shots and the 1000th shot is a goal doubling his conversion rate, the fact that he took 1000 shots to double his conversion rate will not result in his LS attribute increasing. If anything, it would be decreasing.

Again, you're painting a caricature with what I'm saying by thinking of extreme worst-case scenarios which are easily accounted for anyway.

It's not an 'extreme worse case scenario', it's the norm for long shots, the example attribute you chose. Number of long shots scored is low for nearly all players and hence both that and the conversion rate are volatile, with a single event changing the values significantly. A conversion rate changing from 0.03 to 0.06 for the season is quite a considerable change in terms of long shot conversion statistics relative to other players, and yet can be a single in-match event not reflective of any development of a player's shooting ability. 

So long shot goals scored and conversion rates do not sound like useful inputs to determine whether a player's long shot attribute should rise above or fall below the attribute value which generated those shots in the ME. 

 

6 hours ago, goranm said:

Please quote where exactly I'm insisting on that. You won't be able to, because what I've actually written is: "The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute." Do you see how this is different from what you claim I'm saying? 

You are arguing that 'us[ing] long shots in a match succesfully at an above average capacity (relative to his previous ability)' should be a prerequisite for player improvement, and I am observing that this will actually make it more difficult for players to improve an attribute, both because long shots specifically are rare, and because the 'previous ability' is actually the capacity the match engine uses to determine how the long shots turn out

Highlighting the word 'alone' does not change the fact that compared with a model in which a player may improve long shots in training before converting one in a match situation, your proposal is unlikely to generate more attribute changes,

(I'm not sure exactly how you think the training is supposed to contribute towards the improvement of player abilities to perform specific actions without incrementing attributes, but we're coming to that...)

6 hours ago, goranm said:

Well, no, the way development is encoded now is non-sense because it happens before the experience is undertaken. It's the other way, people learn and develop skills when an experience is reflected upon. No one became a better chef by memorizing recepies and cooking for themselves, they had to actually cook something in a restaurant environment with coworkers and customers, react when things go wrong and/or when a customer is demanding. Similarly in football, not everything is learnt in training, some things happen in a match environment only.

No, it's how attribute-driven simulation works: an attribute gives the player a probability of executing an action in the simulation, so incrementing the attribute improves their ability to perform those actions in match. Therefore a development model to affect players' long term ability to perform operates on the principle that attribute improvements drive greater capacity to perform. It's how cause and effect works. This is entirely compatible with directly (and indirectly through the effect of morale on training performances) restricting players' ability to improve the ability to perform in future matches if they do not have enough experience in a match environment and boosting their development speed when they get more game time at a suitable level, which already occurs.

More generally I find it strange that you apparently do not believe people can reflect and learn from training <BLINK>alone</BLINK>. I don't believe that the hours RL footballers spend at training grounds practising free kicks have no effect on their ability to take free kicks (as modelled in game by an attribute called Free Kicks) until after a crowd has seen one go in, and I'm not sure why anyone would even earn the right to take a free kicks in a match if teammates and coaches didn't believe they already had learned to do so up to the relevant ability level. I'm pretty sure the reason they do all this training is to raise their raw ability, and give them lots of opportunities for reflecting on those experiences when called upon to take a set piece in a similar position in a match. As for chefs,  I'm pretty certain that your average Masterchef contestant does a much better job than I would do the first time they enter a professional kitchen to cook for connoisseurs precisely because they spend so much time testing recipes and practising cooking.

 

6 hours ago, goranm said:

If I'm telling you "something else", the better thing to do is to quote me exactly, rather than conjuring a scenario to which you expect (as a figure of speech) a detailed answer. As with other things, if certain things are noise, code them as noise. If easier opportunities shouldn't impact future improvement, code them like that. Give them little to no weight.

So we remove the noise and the ease of opportunity we're left with... the impact of the original attributes. So players don't develop at all! 

If you tell me 'something else' but persistently refuse to specify what it is (whilst apparently being quite happy to spend all year telling people what it isn't!) it's difficult to draw any positive conclusions about there being a more realistic representation of player development hidden in the delta between the outcomes ME calculates a player's existing attributes are capable of and some other specification of their expected output.

[XaW]

12 minutes ago, Seancool said:

So i just got the message from my assistant coach:

"Sven Ulreich only is an average Keeper in the Bundesliga" and it was "former good keeper in the Bundesliga".
Bust his CA matched is PA of 137 so he never was higher rated than now/today?
How can he be "worse" when "mathematically" he has the best case personal rating, like today at age 32!? 🤔
 
I don't understand that message. Does this mean a keeper with 137 CA only is an average keeper in the Bundesliga compared to other keepers in the Bundesliga?

So the reference is not is PA but the CA of similiar players in my Leagues?

This goes back to what I wrote earlier about how the staff in game doesn't see the CA/PA, but the CPA/PPA. And your coaches now rate him differently than before. It could also mean the Bundesliga also have changed slightly in the rankings, but I'd more think it was that you staff know more of him.

[enigmatic]

33 minutes ago, Seancool said:

So i just got the message from my assistant coach:

"Sven Ulreich only is an average Keeper in the Bundesliga" and it was "former good keeper in the Bundesliga".
Bust his CA matched is PA of 137 so he never was higher raten than now, with 32?
How can he be "worse" when "mathematically" he has the best case personal rating? 🤔
 
I don't understand that message.

Coaches are sometimes wrong

[WizbaII]

6 hours ago, goranm said:

Please quote where exactly I'm insisting on that. You won't be able to, because what I've actually written is: "The point is you can train all you want, but if you don't deliver, then training alone shouldn't contribute to an increase in the long shot attribute." Do you see how this is different from what you claim I'm saying? Again, please do a better job at reading what I've written.

goranm, you're a terrible communicator and you don't even know it. You shouldn't have a discussion with someone if you don't even know what you yourself are saying. I read your post and you are spewing nonsense and not allowing people to tell you why it's nonsense. That means you're also a terrible listener. I'm not going into specifics because others have already done that to no avail. It's one thing to disagree, but if you can't even communicate properly, then you should leave people in peace.

[Platinum]

On 31/12/2020 at 22:34, goranm said:

Player development has been an issue for a long time. Great strides have been made to improve it, however currently I'm talking about FM21, which is about the same as FM20. FM19 I did not play much - usually I play until newgens completely take over, but with players peaking at 23/24 this usually means that I know how the game world will look like after 7 or so seasons for the next 10 seasons due to predictability of development.

Most players at high level and with decent training facilities are already close to PA at ages 23/24, leaving little space for development after that age. This I've confirmed with the in-game editor in FM20, but don't want to do with FM21 yet as I don't want to spoil my game. I'm 9 seasons in in FM21 and haven't seen anything suggesting that development differs from FM20.

From what I see in your screenshot, your player started as a league two player, one season as a league one player, then 5 seasons in championship and only then he starts playing at the highest level, at age 27. Match experience at higher levels will contribute more to development - had this player started playing regularly at the highest level at 19-23, you'd maybe see his development maxed out at 23/24ish. This is why one player doesn't really prove or disprove anything.

Here is an example @francis#17 : I won't say who the player is to not spoil it for others, but here are his attribute changes since age 22, age 23 and age 24. The player is now 28, will turn 29 at the end of the season. You can get this under Training - Development - Compare Attribute Changes Since Age X. I'm doing mentals only since there was almost no changes in technicals or physicals.

Now as you can see since age 22: major increase in Vision, Positioning, OTB, Anticipation, large increases in other attributes (straight up arrow).

Since age 23: slight increases, only OTB and Leadership increased by 1 (OTB was actually increased by 0.6-0.8), only other attribute with a straight up arrow is Vision

Age 24: Only Leadership has a straight up arrow, most arrows are yellow.

The last graphs shows this even more clearly: from ages 21 to 24.5 there is a lot of development and then stagnation. The stagnation occurs actually closer to 23, since only anticipation consistently increases (and leadership, the violet line around the middle), and other values fluctuate by what looks like 0.4-0.6 points with work rate jumping in increments of 1 (it used to be that we could see attributes to decimal places, if this can be enabled, I don't see where).

 

On average most players in the real world don't have massive improvements after 25 and some even get worse. There are of course players that do massively improve but an example of 1 player that hasnt improved doesn't prove that it doesnt happen in-game. 

(Btw the screenshots werent mine they were someone elses)

[Seancool]

@enigmatic@XaW Thanks for the hints, I am still trying to get "into" all that metrics and how the AI rates situations and allocates or caluclates certain numbers and puts them into a context with all that messages and information

[goranm]

10 hours ago, herne79 said:

Players develop their attributes by having effective Training schedules and gaining match experience at a relevant level.  Both of these factors are important for player development, however for players aged 18 and over match experience takes precedence.  For player below this age Training takes precedence.  As you can see, increasing an attribute isn't really down to training "alone".  (SI have told us this is how it works, that's not just speculation).

The issue is that the match experience is not nuanced enough, it is either 1 or a 0.

10 hours ago, herne79 said:

I'm not sure where you're getting this from as it is not correct.

Development takes place as players play matches and their performance in those matches, not before.  Gains from individual matches will however tend to be negligible and thus we wouldn't notice them - we'd only start to notice gains after several matches have been played - but players don't develop "before" experience is gained.  Their development is based on their experience gained.

I am talking about matches influencing individual attributes. As it is currently stands, this is not how development works in the game, it is following some hard-coded patterns. A player with a 7.4 average rating develops at approx the same pace as the one with 6.9 assuming both have the same number of matches and minutes played. Some players might score 20 goals in a season, but due to the hard-coded pattern they are following, we might not see their finishing increase as much as a player scoring 8 goals in a season that is on a different hard-coded pattern. In FM21 this is particularly visible since it allows AI managed players with 12 finishing to score 30 or more goals per season (w/o penalties).

7 hours ago, francis#17 said:

On average most players in the real world don't have massive improvements after 25 and some even get worse. There are of course players that do massively improve but an example of 1 player that hasnt improved doesn't prove that it doesnt happen in-game. 

There's more than 1 example, but it's impractical to add them all on a forum post. This is a general observation that in the game players stop developing after ages 23/24. I'm not arguing for massive improvements, I'm arguing for improvements, that in-game don't happen after 23/24. Maybe one or two attributes improves to some extent, but others fluctuate around a given value. I disagree about your first statement - take just Liverpool for example. Van Dijk, Henderson, Mane, Wijnaldum or Firmino are all better than they were at 25.

 

Edited January 2, 2021 by goranm

[goranm]

8 hours ago, WizbaII said:

goranm, you're a terrible communicator and you don't even know it. You shouldn't have a discussion with someone if you don't even know what you yourself are saying. I read your post and you are spewing nonsense and not allowing people to tell you why it's nonsense. That means you're also a terrible listener. I'm not going into specifics because others have already done that to no avail. It's one thing to disagree, but if you can't even communicate properly, then you should leave people in peace.

"not going into specifics" like you are here, is terrible communication. Saying that I am "spewing nonsense" without referring to anything is terrible communication. Not even @ me when you're calling me out is terrible communication. I know what I am saying and I always respond to other people with exact quotes. However the same courtesy is not always extended to me, for example like you haven't done here, and instead of replying to what I've said, other users sometime reply to what they think I've said (which is a form of logical fallacy, but lets not go into that). 

Here are some silly metrics: I have been upvoted 126 times in 196 posts, you have been upvoted 26 times in 349 posts. In this thread alone I was upvoted at least 10 times, so someone is agreeing with some of the things I'm saying.

Quote

you should leave people in peace.

You see, this is a sign of bad communication. It is just plain rude to say that I am disturbing people when (i) this is a public forum (ii) I am not forcing anyone to do anything (iii) am following the rules and guidelines (on the other hand I was called stupid and a troll) and (iv) everyone is here by choice and you have the option of blocking me, in which case I believe that you won't see my replies, and I strongly urge you to do that.

 

Edited January 2, 2021 by goranm

[Platinum]

38 minutes ago, goranm said:

There's more than 1 example, but it's impractical to add them all on a forum post. This is a general observation that in the game players stop developing after ages 23/24. I'm not arguing for massive improvements, I'm arguing for improvements, that in-game don't happen after 23/24. Maybe one or two attributes improves to some extent, but others fluctuate around a given value. I disagree about your first statement - take just Liverpool for example. Van Dijk, Henderson, Mane, Wijnaldum or Firmino are all better than they were at 25.

 

Yep thats why I said the majority, Im not saying its impossible or that it doesnt happen; Im saying its much less common. Are you saying that in the game no players develop much after 24? Or are you saying its rare?

[FrazT]

OK- this thread is in danger of starting to go round in circles, but please just all keep the discussion civil.  Allow users who have a different opinion to you to express that opinion and debate the points.  If this deteriorates any further, the thread will be closed.

[goranm]

2 minutes ago, francis#17 said:

Are you saying that in the game no players develop much after 24? Or are you saying its rare?

At the highest level, where players get the most development bonus, they in general don't improve much after 24 because they hit their PA too quickly. If you haven't already, after playing 10 to 20 seasons of top tier management, you'll start observing that. You can verify in-game with the development tab that most of your players won't develop much past 24. I am yet to see a Vardy - in FM20 I attempted to recreate a Vardy by stumbling on a player with 200 PA but only 120ish CA at age 23. He reached a certain CA by age 26, far from the 200 PA, and then stagnated for the rest of his career. I'm not saying which CA to not spoil Vardy's hidden attributes, but he wasn't at Vardy's level despite me doing everything possible to improve him. I'm not saying that development curves like Vardy's don't happen in the game, but they seem very rare and hard to recreate.

[Platinum]

1 hour ago, goranm said:

At the highest level, where players get the most development bonus, they in general don't improve much after 24 because they hit their PA too quickly. If you haven't already, after playing 10 to 20 seasons of top tier management, you'll start observing that. You can verify in-game with the development tab that most of your players won't develop much past 24. I am yet to see a Vardy - in FM20 I attempted to recreate a Vardy by stumbling on a player with 200 PA but only 120ish CA at age 23. He reached a certain CA by age 26, far from the 200 PA, and then stagnated for the rest of his career. I'm not saying which CA to not spoil Vardy's hidden attributes, but he wasn't at Vardy's level despite me doing everything possible to improve him. I'm not saying that development curves like Vardy's don't happen in the game, but they seem very rare and hard to recreate.

As you know, a player like Vardy is extremely rare. I dont think its possible to recreate a Vardy when making a player due to hidden values, Seb Wessel has said that each player has a 'key development years'  age range, which we dont have access to. For most players the 'key development  years' will be until they are 24/25 as it is in real life. But for a smaller number of players it will be when they are older. So recreating a Vardy isn't as straight forward as thought.

And as you already acknowledged not many players turn into much better players after they are 25, so this is something that should be rare. I dont have the numbers on how many players this happens to in game but from what Seb has said the FM implementation sounds realistic.

Edited January 2, 2021 by francis#17

[herne79]

1 hour ago, goranm said:

The issue is that the match experience is not nuanced enough, it is either 1 or a 0.

I'm afraid you are mistaken.  The rate of development can be directly influenced by such factors as player performance during matches and the amount of time played during matches.  For example a player who plays well for a whole 90 minutes may benefit from more development than a player who gets subbed on at 60 minutes and plays well.  It is very far from being a 1 or a 0.  We just don't realise it in game because the values attributed by such match performances are very small, however over time they can build into something more tangible.

1 hour ago, goranm said:

I am talking about matches influencing individual attributes. As it is currently stands, this is not how development works in the game, it is following some hard-coded patterns. A player with a 7.4 average rating develops at approx the same pace as the one with 6.9 assuming both have the same number of matches and minutes played. Some players might score 20 goals in a season, but due to the hard-coded pattern they are following, we might not see their finishing increase as much as a player scoring 8 goals in a season that is on a different hard-coded pattern. In FM21 this is particularly visible since it allows AI managed players with 12 finishing to score 30 or more goals per season (w/o penalties).

You are correct, that is not how development works in game.  Development is way more complex than anything anyone has said in this thread.  Yes a player who scores 8 goals a season may well see their Finishing increase more than a player who scores 20.  Then again it could easily be the other way around.  There are a huge amount of factors which get taken into account on an individual basis and it is much too simplistic to state anything otherwise.  Injuries, match performance, training facilities, coaches, personality, squad influence, you the manager, morale, player discussions, existing attribute level, player age and much more besides can all have an influence on the rate at which a player may develop.  And all of those influences can and do differ from player to player.

Just because we might not notice it or something might seem odd doesn't mean it isn't happening.

[goranm]

12 hours ago, francis#17 said:

And as you already acknowledged not many players turn into much better players after they are 25, so this is something that should be rare. I dont have the numbers on how many players this happens to in game but from what Seb has said it sounds realistic.

Players improving past age 24/25 shouldn't be that rare as it is in the game. I'm not talking about massive improvements here, but consistent improvement that's seen with a lot of well performing players in the high tiers.

11 hours ago, herne79 said:

Just because we might not notice it or something might seem odd doesn't mean it isn't happening.

However it is noticeable that players' development stagnates or is very slow past 24, at least in the highest tier. This can be verified with in-game tools in the development tab - I gave a typical example in one of my posts above where this is very visible.

Quote

I'm afraid you are mistaken.  The rate of development can be directly influenced by such factors as player performance during matches and the amount of time played during matches.  For example a player who plays well for a whole 90 minutes may benefit from more development than a player who gets subbed on at 60 minutes and plays well.  It is very far from being a 1 or a 0. 

It's not really that far. I have never had any issues with the development rate of players who I'd use more as subs than as first-choice players. A player who starts 70% of my total matches would develop at approx the same rate as the one starting 30% of my total matches. Players doing okayish develop at approx the same rate as the ones doing good or doing great. In the last couple of iterations of FM I had no problem developing two players of same or similar age in the same position, in parallel, and maxing them out by age 24. Both then request a €300k p/w contract, so I sell one, get another prospect in the same position, develop him, by the time he's 24 the original player is 30-32 and starts declining physically, I sell him, rinse and repeat.

Here is an example of the development of two players since age 18. Currently both are 21, one was bought at age 18, one was promoted to the first team at age 18 with no prior first-team experience. Approx 3.5 seasons elapsed and approx 200 games were played in that time. One is a ball playing defender, the other is a deep lying playmaker.

The BPD: 2 year of previous first-team experience with Gremio (32 starts, 8 subs, he's 17 at the time); with me 58 starts, 17 subs; 19 caps for Brazil; by far a better average rating than the DLP

The DLP:  with me since age 16 (maxed out training/youth facilities), no first-team experience before age 18; 28 starts, 54 subs; 2 caps for Belgium; worse performance than the BPD

The BPD sometimes starts knock-out games in UCL, and rival league games. The DLP mostly starts unimportant cup games, group stage UCL games when I've already progressed, and league games when I've already won it. The BPD by all accounts should at least match the development rate of the DLP, given that he's played much more minutes, played better and at a higher level, even if his starting CA at age 18 I estimate to be higher by some amount, could be anything between 10-30 (I don't know what their CA is or was 4 seasons ago). But the difference is dramatic, in favor of the DLP who has much less minutes, worse average performance, less international experience etc. Technicals: +6 vs. +5 for DLP. Mentals: +14 vs +8 for DLP. Physicals: +13 vs +6 for DLP. Total gains: +33 vs. +19 in favour of DLP.

EDIT: original screenshot for the DLP was incorrect (it was development since 16 y/o, not 18), this is the correct one

 

Edited January 3, 2021 by goranm

[goranm]

13 hours ago, enigmatic said:

It's not an 'extreme worse case scenario', it's the norm for long shots, the example attribute you chose. Number of long shots scored is low for nearly all players and hence both that and the conversion rate are volatile, with a single event changing the values significantly. A conversion rate changing from 0.03 to 0.06 for the season is quite a considerable change in terms of long shot conversion statistics relative to other players, and yet can be a single in-match event not reflective of any development of a player's shooting ability. 

Ok, if the IRL stats are like that, then don't put a strong weight on conversion rate, it's as simple as that. There are other tools IRL and in the game that can indicate improvement, for example xG and performance relative to xG. Again, if the conversion rate is doubled, but the number of shots it took to achieve it was huge, so account for that.

13 hours ago, enigmatic said:

So we remove the noise and the ease of opportunity we're left with... the impact of the original attributes. So players don't develop at all! 

Well you started talking about noise. We have the impact of original attributes, we have the impact of 10 other players, we have the impact of the opposition, we have the impact of playing conditions. We don't have to add or subtract anything that's not already in the ME.

13 hours ago, enigmatic said:

You are arguing that 'us[ing] long shots in a match succesfully at an above average capacity (relative to his previous ability)' should be a prerequisite for player improvement, and I am observing that this will actually make it more difficult for players to improve an attribute, both because long shots specifically are rare, and because the 'previous ability' is actually the capacity the match engine uses to determine how the long shots turn out

I never said it was the only prerequisite. I keep saying that it should be a factor, not the factor.

13 hours ago, enigmatic said:

No, it's how attribute-driven simulation works: an attribute gives the player a probability of executing an action in the simulation, so incrementing the attribute improves their ability to perform those actions in match. Therefore a development model to affect players' long term ability to perform operates on the principle that attribute improvements drive greater capacity to perform. It's how cause and effect works. This is entirely compatible with directly (and indirectly through the effect of morale on training performances) restricting players' ability to improve the ability to perform in future matches if they do not have enough experience in a match environment and boosting their development speed when they get more game time at a suitable level, which already occurs.

Again, you're explaining the game mechanics as they are now. No one is arguing that the game doesn't work like that.

13 hours ago, enigmatic said:

More generally I find it strange that you apparently do not believe people can reflect and learn from training <BLINK>alone</BLINK>.

This is not belief, this is how improvement works in general. People get better at stuff by doing stuff. Training is necessary, but not sufficient for improvement. Footballers get better at football by playing football. Competitive athletes get better by competing. Competitive environments are where a competitors skills and limits are tested, not the training ground. You can have the greatest professional in training, but if that doesn't translate into match performance for some reason or other, that player isn't getting better at playing football, they're just training well. On the other hand, you can have someone who isn't dedicated that much to training, but still excels at playing (e.g. Hazard, if the gossip is to be believed), so all the managers still pick him because they know he will perform, even if training was sub-par.

13 hours ago, enigmatic said:

I don't believe that the hours RL footballers spend at training grounds practising free kicks have no effect on their ability to take free kicks (as modelled in game by an attribute called Free Kicks) until after a crowd has seen one go in

How else do you judge if the effort put into something was worthwhile, especially in a competitive environment? If someone dedicates a significant amount of time to training FK's, but doesn't translate it into match performance for some reason (maybe nervousness, maybe predictability etc.), any sensible coach will ask the player to stop wasting training time and ask someone else to take them.

Quote

and I'm not sure why anyone would even earn the right to take a free kicks in a match if teammates and coaches didn't believe they already had learned to do so up to the relevant ability level.

Because of confirmation bias. The majority of the time you are only seeing those players that are training FK's and can convert them in a match. You're not seeing all the ones that have trained FK's but failed to convert them, because the coaches/managers have already seen them fail and designated someone who can convert a free kick in match to take it.

13 hours ago, enigmatic said:

As for chefs,  I'm pretty certain that your average Masterchef contestant does a much better job than I would do the first time they enter a professional kitchen to cook for connoisseurs precisely because they spend so much time testing recipes and practising cooking.

Contestant being key here. They are cooking for other, more knowledgeable people and are open to criticism from more capable chef's who can pinpoint their mistakes and weaknesses, so that they know what to work on for the next contest. Hence it was necessary for the contestants to undertake the experience first, and the correction and improvement comes after that. Had they've done it in their own kitchen, there would be no one to pinpoint the mistakes and what can be improved.

Quote

If you tell me 'something else' but persistently refuse to specify what it is (whilst apparently being quite happy to spend all year telling people what it isn't!) it's difficult to draw any positive conclusions about there being a more realistic representation of player development hidden in the delta between the outcomes ME calculates a player's existing attributes are capable of and some other specification of their expected output.

If I told you 'something else', I've said what it is, so you can quote it - I'm not sure again what are you referring to here. The majority of our arguments have the pattern of you saying something isn't like that but rather like that, and me replying that it doesn't have to be how you are suggesting (see e.g. conversation about pass completion), so it's natural that I'm going to be talking about "what it isn't".

It's not difficult to come up with a very rudimentary model that would account for match performance (that certainly has it flaws):  suppose we isolated a single attribute, say FK. Develop it according to V*(training performance) + W*(match performance), where V and W are some weights, training performance is a score dictated by Professionalism, Work Rate, Facilities, Coaches or whatever plus a random factor, and match performance is a score dictated by whatever happened in the match. It might as well be that V and W are functions of time, so very early on in a player's career we have V(t) > W(t), but later in a player's career it might be V(t)<W(t). Maybe V(t) is also a function of training facility quality, and W(t) is maybe also a function of opposition difficulty etc.

Maybe a FK is scored in 1 try, maybe in 3 takes none are scored but are taken well (go past the wall, are shots on goal), maybe 10 free kicks are attempted and all are badly converted, i.e. they all hit the wall. Assign appropriate scores to those outcomes. Now over X weeks average out (to some mean, not necessarily arithmetic) all the V*(training performance) + W*(match performance) scores - if it hits above a certain threshold by say 0.1, improve the attribute by 0.1. If it hits below the threshold by 0.4, decrease the attribute by maybe 0.2 (the increases and the decreases don't have to be linear functions corresponding to the distance of the score from the threshold - even the threshold doesn't have to be fixed) and notify the manager that over the past X weeks the player hasn't really been converting training into performance and it might be better to focus training elsewhere. This would also automatically clarify those vague "player doesn't feel the additional training focus is having any effect" messages that we're getting in-game.

Edited January 3, 2021 by goranm

[WeePaul]

13 hours ago, goranm said:

At the highest level, where players get the most development bonus, they in general don't improve much after 24 because they hit their PA too quickly. If you haven't already, after playing 10 to 20 seasons of top tier management, you'll start observing that. You can verify in-game with the development tab that most of your players won't develop much past 24. I am yet to see a Vardy - in FM20 I attempted to recreate a Vardy by stumbling on a player with 200 PA but only 120ish CA at age 23. He reached a certain CA by age 26, far from the 200 PA, and then stagnated for the rest of his career. I'm not saying which CA to not spoil Vardy's hidden attributes, but he wasn't at Vardy's level despite me doing everything possible to improve him. I'm not saying that development curves like Vardy's don't happen in the game, but they seem very rare and hard to recreate.

Isn't that realistic though? When discussing that particular kind of development curve, everyone talks about Vardy. He's the go-to example. And that's because there are very, very few players with the kind of development curve like he's had. 

[bowieinspace]

20 hours ago, goranm said:

 

Here is an example of the development of two players since age 18. Currently both are 21, one was bought at age 18, one was promoted to the first team at age 18 with no prior first-team experience. Approx 3.5 seasons elapsed and approx 200 games were played in that time. One is a ball playing defender, the other is a deep lying playmaker.

The BPD: 2 year of previous first-team experience with Gremio (32 starts, 8 subs, he's 17 at the time); with me 58 starts, 17 subs; 19 caps for Brazil; by far a better average rating than the DLP

The DLP:  with me since age 16 (maxed out training/youth facilities), no first-team experience before age 18; 28 starts, 54 subs; 2 caps for Belgium; worse performance than the BPD

The BPD sometimes starts knock-out games in UCL, and rival league games. The DLP mostly starts unimportant cup games, group stage UCL games when I've already progressed, and league games when I've already won it. The BPD by all accounts should at least match the development rate of the DLP, given that he's played much more minutes, played better and at a higher level, even if his starting CA at age 18 I estimate to be higher by some amount, could be anything between 10-30 (I don't know what their CA is or was 4 seasons ago). But the difference is dramatic, in favor of the DLP who has much less minutes, worse average performance, less international experience etc. Technicals: +6 vs. +5 for DLP. Mentals: +14 vs +8 for DLP. Physicals: +13 vs +6 for DLP. Total gains: +33 vs. +19 in favour of DLP.

EDIT: original screenshot for the DLP was incorrect (it was development since 16 y/o, not 18), this is the correct one

 

 

This would be odd if development was linear, but every player is different so it’s a bit odd to compare the two like this. The DLP may have played less but if he’s still getting good training experiences and some match experience why couldn’t he improve more? Ratings in match play a part but it’s important to distinguish form and quality. The idea of attributes and ca is that this forms the blueprint of who they are at their best, regardless of the system they are in and their performances. KDB has had a difficult season this year, if real life was FM his average ratings in the match engine would be lower, but that doesn’t mean his attributes or CA has necessarily changed, just that for whatever reason whether it be form or system, Man City aren’t getting as much out of him. 
 

 

[enigmatic]

18 hours ago, goranm said:

Ok, if the IRL stats are like that, then don't put a strong weight on conversion rate, it's as simple as that. There are other tools IRL and in the game that can indicate improvement, for example xG and performance relative to xG. Again, if the conversion rate is doubled, but the number of shots it took to achieve it was huge, so account for that.

 Performance relative to xG is even more volatile. I'm happy putting zero weight on conversion rate and performance relative to xG, because I'm not the person arguing match engine outputs are useful data to update its inputs .

 

9 hours ago, goranm said:

Well you started talking about noise. We have the impact of original attributes, we have the impact of 10 other players, we have the impact of the opposition, we have the impact of playing conditions. We don't have to add or subtract anything that's not already in the ME.

My point is all these impacts which adjust the impact of attributes on match engine performance are noise with respect to player's attribute development. If you wish to actually make a counter argument that development is plausibly linked to, say, the wind or keeper error resulting in a goal despite the match engine calculating that based on a player's attributes would make them hit that shot badly, or teammates chasing after overhit passes so they register as completed  or anything else in the match engine that causes a player to achieve a different outcome from the expected outcome given the input of their skill level, please actually do so. 

Otherwise I'm going to stick with the assumption that the valuable information about player development from the ME can be reduced to 'are they playing enough at the right level', which additionally incorporates the assumption players can 'learn from their mistakes' and not just from unexpectedly good outcomes.

 

9 hours ago, goranm said:

I never said it was the only prerequisite. I keep saying that it should be a factor, not the factor

 I never said you said it was the only prerequisite. I said that making it a prerequisite would make attribute changes less frequent.

 

10 hours ago, goranm said:

Again, you're explaining the game mechanics as they are now. No one is arguing that the game doesn't work like that.

I am glad that you agree the ME depends on prior attribute improvements to allow a player to sustain performances above his previous capacity. Which is why I am perplexed that your development mechanics proposal being that attribute improvements outside the ME should be driven by the thing they cause. 

 

9 hours ago, goranm said:

Contestant being key here. They are cooking for other, more knowledgeable people and are open to criticism from more capable chef's who can pinpoint their mistakes and weaknesses, so that they know what to work on for the next contest. Hence it was necessary for the contestants to undertake the experience first, and the correction and improvement comes after that. Had they've done it in their own kitchen, there would be no one to pinpoint the mistakes and what can be improved.

Simple question: if we take Person A who has never cooked before and Person B who has cooked a lot in their own kitchen before, but never in a professional kitchen, do we assume both have an equal probability of impressing the judges in the first round Masterchef? If not, why wouldn't we assume that Person B's Frying and Making Souffle attributes had increased above Person A's at the start of the contest.

Especially if we were building Masterchef Simulator and needed attributes for person B to be higher in order for Person B to have a better chance of winning than the person who hadn't cooked....

 

9 hours ago, goranm said:

Because of confirmation bias. The majority of the time you are only seeing those players that are training FK's and can convert them in a match. You're not seeing all the ones that have trained FK's but failed to convert them, because the coaches/managers have already seen them fail and designated someone who can convert a free kick in match to take it.

Confirmation bias from in-match events is literally the attribute update mechanism you have been proposing. See a player convert and assume he will be better at converting in future. See them miss a series of chances and assume they will become worse in future.

And training performances are why David Beckham, a 19 year old playing his first ever game for Preston North End was allowed to take set pieces by teammates who had never seen him in a match (and had plenty of match experience of taking set pieces themselves), and promptly scored. My argument is that this is more realistically represented by famously intensive set piece trainer David Beckham having already trained to be extremely proficient at taking set pieces, thus deserving a strong set piece taking attribute before the match, than Beckham having been unable to develop his set piece taking ability and only improving afterwards because he fortuitously scored. 

 

10 hours ago, goranm said:

It's not difficult to come up with a very rudimentary model that would account for match performance (that certainly has it flaws):  suppose we isolated a single attribute, say FK. Develop it according to V*(training performance) + W*(match performance), where V and W are some weights, training performance is a score dictated by Professionalism, Work Rate, Facilities, Coaches or whatever plus a random factor, and match performance is a score dictated by whatever happened in the match. It might as well be that V and W are functions of time, so very early on in a player's career we have V(t) > W(t), but later in a player's career it might be V(t)<W(t). Maybe V(t) is also a function of training facility quality, and W(t) is maybe also a function of opposition difficulty etc.

Maybe a FK is scored in 1 try, maybe in 3 takes none are scored but are taken well (go past the wall, are shots on goal), maybe 10 free kicks are attempted and all are badly converted, i.e. they all hit the wall. Assign appropriate scores to those outcomes. Now over X weeks average out (to some mean, not necessarily arithmetic) all the V*(training performance) + W*(match performance) scores - if it hits above a certain threshold by say 0.1, improve the attribute by 0.1. If it hits below the threshold by 0.4, decrease the attribute by maybe 0.2 (the increases and the decreases don't have to be linear functions corresponding to the distance of the score from the threshold - even the threshold doesn't have to be fixed) and notify the manager that over the past X weeks the player hasn't really been converting training into performance and it might be better to focus training elsewhere. 

I'm glad you specified it V*(training performance) + W*(match performance), as this means that where v /= 0 training performance alone (i.e. where an individual has zero match performance score from playing in zero matches) can have an impact (where coefficient on V is not zero).  This makes your protracted insistence above that SI and I am wrong to suggest it should do all the more perplexing...

Exactly how the 'appropriate scores' are determined seems the important bit though. If we correctly weight them all for difficulty (likely possible for someone with access to the ME code that produces the actions) the attributes don't change, because players do not perform them 'above their capability' in a match engine. They might perform some low probability actions in some instances, and fail to perform some high ones in others, but over X weeks the score converges on that implied by their attributes. Only events which are volatile like long shot goals are unlikely to converge on values implied by attributes within X, and they're the ones which tend to be disproportionately affected by fluky opponent events like keeper errors.

The only way you get growth is to assign scores which overestimate the difficulty of performing the action. If you set small values of X and heavy weighting on rare goals you get free kick taking jumps by a point each few weeks a player scores in, which feels a bit artificial and predictable, if you set high values of X and more weighting on just performing the expected good:bad free kick ratio you get steadier and easier improvement. Neither appears to describe a more realistic development curve especially since its far from clear designated set piece takers actually improve their set piece taking broadly proportionately to the number they take (and/or number they score) over the course of their career. 

Alternatively we can scrap the whole thing as intractable and computationally intensive since there's no actual information content for a reinforcement learning process to discover, especially since we already have 'playing matches at an appropriate level' as a factor which gives players a chance of improving from matches, and we don't have to update and retest player development every time someone tweaks the match engine code.

 

10 hours ago, goranm said:

This would also automatically clarify those vague "player doesn't feel the additional training focus is having any effect" messages that we're getting in-game.

Must admit I prefer the "player doesn't feel the additional training focus is having any effect" messages to [vaguely] imply you probably shouldn't bother using the focus as it is unlikely to work rather than essentially meaning "in case you haven't noticed, the player hasn't scored a set piece, this could all change in the next game, leave training as is".

 

[goranm]

1 hour ago, bowieinspace said:

The DLP may have played less but if he’s still getting good training experiences and some match experience why couldn’t he improve more?

Because the BPD is getting the same good training experiences, with the same coaches. They are in the same "defensive unit" so a lot of the time they train the same things, yet the DLP improved more, particularly in Mentals. They both have the same personality, Fairly Professional. So everything is the same or mostly the same, except the BPD playing more than double the minutes of the DLP, yet the DLP progresses more in training - so what can we conclude here? That the first-team match experience, at least in this example, didn't carry much weight with respect to development rate. This is why I say that it is either a 1 or a 0 and not nuanced enough. Both have match experience, but the comparable lack minutes played doesn't seem to hurt DLP's development rate.

[goranm]

2 hours ago, enigmatic said:

Otherwise I'm going to stick with the assumption that the valuable information about player development from the ME can be reduced to 'are they playing enough at the right level', which additionally incorporates the assumption players can 'learn from their mistakes' and not just from unexpectedly good outcomes.

I gave an example above in which two players, of same age (21), training the same amount of time with the first-team (18-21), of the same personality (Fairly Professional), in the same unit (defensive), except one has more than double the minutes of the other, more than 10x the international experience, and performed better on average by far. Intuitively, the one with more experience should improve at least at the same rate as the other, yet with only the "are they playing enough at the right level" we the get counter-intuitive result that the player with less experience, less minutes, less pretty much everything improved more. According to that information, and at least in that example, the ME has little to no impact on development, and the predetermined hard-coded development curve completely dominates whichever development information the ME outputs.

Quote

if we take Person A who has never cooked before and Person B who has cooked a lot in their own kitchen before, but never in a professional kitchen, do we assume both have an equal probability of impressing the judges in the first round Masterchef?

Yes, you can assume that, because "cooking a lot" doesn't give you information about anything except that B was in the kitchen a lot. You don't know if Person B was cooking a lot badly. There is an expectation that a person spending more time cooking will do better than a person that has never cooked, but expectation is not the same as probability. Person A might as well just read the recipe and make better scrambled eggs than B who maybe spent years improvising on the recipe and picked up bad habits because nobody else was tasting their food. Just look up all the X-Factor/Idol/whatever auditions with people saying oh I've been singing for thousands of years, I'm a vocal coach bla bla only to completely bomb.

2 hours ago, enigmatic said:

Confirmation bias from in-match events is literally the attribute update mechanism you have been proposing. See a player convert and assume he will be better at converting in future. See them miss a series of chances and assume they will become worse in future.

No, it's not, because you have all the information across all the period of time the player was with you. Confirmation bias is about having some of the information about some of the period of time. You not only see a player convert, you also see him fail. If someone fails at a much greater rate than they succeed in a skill (in relative terms for the skill that they're undertaking), they generally will not become better in the future.

2 hours ago, enigmatic said:

I'm glad you specified it V*(training performance) + W*(match performance), as this means that where v /= 0 training performance alone (i.e. where an individual has zero match performance score from playing in zero matches) can have an impact (where coefficient on V is not zero).  This makes your protracted insistence above that SI and I am wrong to suggest it should do all the more perplexing...

As I've said, this is a very rudimentary model with flaws, I didn't spend more than 2 minutes thinking about it. Insisting that this is actually how I want it implemented just shows that you're arguing in bad faith. You're right, where V != 0 and W = 0, training performance alone has impact. So then V=0 if #matches=0 at any level. The end. I'm glad that you are glad about finding mistakes in your own assumptions, as I've never assumed that V is non-zero. One other way to easily account for your assumption is to change the + to a *, so when either are 0, development is 0. Maybe you want to use negative weights, so (-1)^p for some odd or even power p has to be put in front. There are many ways to account for your assumptions.

2 hours ago, enigmatic said:

If we correctly weight them all for difficulty (likely possible for someone with access to the ME code that produces the actions) the attributes don't change, because players do not perform them 'above their capability' in a match engine.

Again, you can have a correct weighing that does not have to correlate with how the ME produces actions. The point is to simulate development, not to carbon-copy the probabilities in the ME. For example, there are several types of difficult passes, each has their own probabilities assigned in the ME. For the purposes of development, you could group them all in a single group and assign a single score to that group.

2 hours ago, enigmatic said:

The only way you get growth is to assign scores which overestimate the difficulty of performing the action. If you set small values of X and heavy weighting on rare goals you get free kick taking jumps by a point each few weeks a player scores in, which feels a bit artificial and predictable, if you set high values of X and more weighting on just performing the expected good:bad free kick ratio you get steadier and easier improvement. Neither appears to describe a more realistic development curve especially since its far from clear designated set piece takers actually improve their set piece taking broadly proportionately to the number they take (and/or number they score) over the course of their career. 

Then don't do it like that. In the realm of all possibilities for development models, saying that this is the only way to be done requires some evidence.

Edited January 3, 2021 by goranm

[goranm]

11 hours ago, WeePaul said:

Isn't that realistic though? When discussing that particular kind of development curve, everyone talks about Vardy. He's the go-to example. And that's because there are very, very few players with the kind of development curve like he's had. 

Yes, it's completely realistic, I'm not saying it isn't. However players generally reaching their peak at 24ish is not.

[bowieinspace]

2 hours ago, goranm said:

Because the BPD is getting the same good training experiences, with the same coaches. They are in the same "defensive unit" so a lot of the time they train the same things, yet the DLP improved more, particularly in Mentals. They both have the same personality, Fairly Professional. So everything is the same or mostly the same, except the BPD playing more than double the minutes of the DLP, yet the DLP progresses more in training - so what can we conclude here? That the first-team match experience, at least in this example, didn't carry much weight with respect to development rate. This is why I say that it is either a 1 or a 0 and not nuanced enough. Both have match experience, but the comparable lack minutes played doesn't seem to hurt DLP's development rate.

But you’re talking about two different players who will develop in different ways at different times, with different ceilings. If the game was so simple that players developed exactly as you expected them to it would be very boring 

[Flumpy_85]

Its ridiculous that players dont really develop after the age of 23 - 24