Yarema said: B teams in non playable leagues actually do play official matches, that is why the assistant isn't scheduling friendlies for majority of the season (only preseason and winter break). You can see the players are getting competitive league appearances but you cannot see the actual matches or schedule.
The issue with B teams is that they are mostly semipro which is kind of awful for development. Different countries do B teams differently though. For example in France a B team is part of the main club, sharing professional status. In most other countries B team is an affiliate. Expand
I am ware of this feature that the team can play games that are no in the schedule. I'm talking here about a case where B team have no games at all.
So basically B Teams that are in non-playable league and play no official matches can be effectively use for player's growth if we just set assistant to 'Arrange a fixture if there is no match in the week'.
As always thank you very much for providing crucial evidence!
ClaudeJ said: I did, and even quoted it a few posts above. You may have missed that context clue. I may not have fully grasped it the way you intended.
Anyway, I understand I'm not the target audience of your work, and I thank you for taking the time to elaborate.
Cheers Expand
Then you'd noticed for example the part with 1000 minutes required to be part of the training data set and combining league rank into players data.
The goal was to make rather universal model that will be a more of a guidance for the player what attributes look for than some absolute solution to players selection. I did it mostly for myself out of curiosity and since I already did the work I've decided to share it with the community so others can benefit from it. And then I've decided to update with it attributes for each position (because first model was for all outfield players) and then expanded it with custom match engines.
I'm currently not planning on doing another models dedicated for top leagues because I believe they won't be that different from general models.
I highly recommend you reading about methodology of this research - the source method posted on SI forum - because clearly you haven't read it.
ClaudeJ said: My main concern is that the average includes non-playing players, possibly those not even on the squad roster. Expand
It definitely does not because: 1) players that don't play won't have average rating at all - so the target value 2) part of players filtering contains minimum of 1000 minutes played in the league in the season
ClaudeJ said: PS: ingame, each competition has a reputation value, regardless of its continent, and that could serve as a global relative ranking. Expand
Each player has added a custom feature that represents their respective league position (that is based on the league reputation) in Europe. Every player attributes are 'transformed' to not use 'absolute' value of the attribute but it shows difference between players actual value and the average value for the league his playing in, for every single attribute considered.
ClaudeJ said: I understand that refining the dataset would require a lot of additional work, but I truly believe it would better serve its intended purpose by providing a clearer and more accurate representation of attribute importance at a competitive level. Expand
Not everyone plays with top leagues - I do not. I never in FM era played with a team in top 5 leagues. I'd also make a bet that among people who lurk on this forum there is higher share of players playing in lower leagues than in general population of FM players. Standardisation should compensate for the 'relativeness' of attributes. Please read those topics mentioned before. It would answer a lot of your questions and concerns.
lordus said: I really like your approach, and I'll give it a try next week.
By the way, I noticed that for the DMC position in the vanilla Match Engine, only ACC has a coefficient, while Pace is 0. However, in all other MEs, these two seem to have the highest coefficients. Compared to all other positions, DMC appears to be quite an outlier in the vanilla ME. Could you maybe check again to see if there's a mistake in there? Expand
This is only an anecdotal evidence but I had in my playing experience cases where DMC was this particular position that did not required great physicals from the player (maybe beside balance - but this is not that affected by ageing). I had a 36yo veteran playing DLP in DMC slot with far below league and even team average physicals while maintaining one of the best average rating - even thou he had almost no goals and assist so his rating was not 'inflated' with g/a.
I had a plan on doing another 10 years simulation for 'vanilla ME' because now in this testing save it's so far in the game that there are almost no real players, only the newgens. So the results would be not skewed towards real players attributes - and we know that for certain positions real player have different attribute distribution than newgens with Fullbacks/Wingbacks being good example since the newgens there have usually relatively low crossing for example.
ClaudeJ said: Could the current dataset be skewed by an overrepresentation of lower-level players, whose attributes may not reflect high-level gameplay? You mentioned yourself: 'the data is flooded with players with relatively low CA.' We all know that in lower leagues, and even near the bottom of top leagues, the game tends to be more primal and physical rather than technical.
By focusing on top teams and excluding young, developing players, the analysis might better capture the attributes that truly influence performance at an elite level. Given that, what do you think about refining the sample to focus on actual top-tier teams and filtering out younger, still-developing players?
I feel this would make the findings more representative of how FM is actually played: primarily in the top five leagues, with most players managing elite squads or near-peers, at least from what I’ve seen in the video content and forums. This would, in turn, make your valuable work even more useful to the community. Otherwise, many might draw the wrong conclusions from it: simply because they don’t read the full accompanying text, for whatever reason. Expand
This issue should be somehow compensate with data being standardised which means attributes for every player are 'relative' to their league average. The question with this method is - are attributes relation is linear - example if league average is 10 pace, and the player has 15 will the effect be the same as for the league having average pace 15 and player 20.
If we wanted to tackle this issue my guess would be - make a graph of league 'rank' and players average CA and based on the results filter out players playing in leagues with high enough 'rank' (just reminder - rank is a league place in continental ranking).
It's doable but would require additional work. Especially exporting again data for all the players for all the seasons for all the engines since I did not include CA in their export data.
DoubleR said: So it looks like pace and acceleration are still king? Expand
Yep, and Jumping Reach. Difference is that sometimes some technical attributes like Passing or Technique are showing. Previously it was mostly just Dribbling.
They are custom schedules made by author of the test. You have literally indicated what, and how many, exercises there are. Their order doesn't matter. Just make yourself custom schedules. Keep in mind that for schedules that have more exercises do two versions of them. One for 1 game week and the other for 2 games week.
Great work. 1) That would explain why Work Rate was so crucial attribute in your previous test 2) This somehow force to diverse the players attributes and roles in a way that the player won't benefit from 10 playmakers 3) This could be the reason why games promotes physical attributes - the don't introduce this 'negative' effect 4) It's interesting that this mostly affects decision - so mostly for other attributes still the higher - the better, in general case
JustinCredible said: Thank you for your reply, took some messing about (including removing the favoured position column from the view as that's not in the excel sheet) but just about got there in the end!
I guess the thing that stands out to me upon first inspection is how closely rated many of the players in the same position are (for example trying to pick the best two wingers). Not sure if this is because I'm a top 6 Prem team with a strong squad but I'm not sure I've learnt anything from this sheet.
That said I can see how it would help in trying to work out who to buy to improve the squad (if I can work out how to get potential signings data into the Squad_Data_N tab) Expand
That looks really good. @DeFlow If I may suggest one thing - I'd use conditional formatting to highlight player's best position/highest score on the table. I know it shows his best position on the right but highlighting it with very different colour on the table itself would be in my opinion very helpful.
Mrjoser said: Althoug possible, I think there might be another explanation. In your experiment, you are comparing players with respect to the league average. I think its safe to assume that in any given league, majority of players are close to some "average" CA value for that league. Lets take 2 strikers from one league with the same CA. For Strikers Finishing has PA weight of 8; Acceleration has PA weight of 10, Pace has PA weight of 7, Jumping reach 5 and Balance 2!!! Striker A has 15 finishing. Striker B has only 10 finishing. But this will free 40 PA that can be redistributed to other attributes. For example: 4 ACC; 5 PAC + 1 JUM; 10 BAL + 4 JUM; and so on.
So I think in the end, Finishing isnt useless or negative. But there are simply better attributes to have for striker. Therefore striker with high FIN can suffer from lack of other essential attributes if the CA should be on similar level.
I have sometimes problems to expres my thoughts, but I home this makes some sence. Expand
I have to admit I might've made a small mistake. The coefficients in this model where I showed all the features in FM23 was probably for all outfield players not just forwards. My mistake was due to having too much of those models made and I just didn't check which one was which - since I have main program that does the model and just change the input data.
Jolt said: Yes, the underlying assessment that for strikers (the one position where the attribute would be essential to successfully accomplish the main task of the position), having above average finishing is does not make any meaningful difference in average rating (knowing that scoring goals is practically the one certain and main way of substantially increasing the average rating for strikers) is mind boggling.
This one experiment was also conducted with the same underlying parameters (A very significant amount of leagues and divisions, running for 11 seasons)? Expand
Sorry if it's misleading but I was doing so many of those test in different settings that I can't really remember if those coefficients that I provided (with all 51 features) was for Strikers or for just overall outfield players. I'll check that tomorrow. Regarding this test it was performed on the data extracted from the same saves as the FM 24 goalkeepers test so you have all the leagues listed and it was 10 seasons there as far as I remember.
So according to your results, STC having higher Finishing equals having worse average rating? Expand
Keep in mind that this is for difference between actual attribute value and the average for the league. So for example if average finishing was 10, the higher than that would result in -0.000456 per every finishing point. So if the average was 10 and players value was 20 it still results in lowering average rating by 0,00456. So for features with coefficients this small you can basically say that they just don't make any difference. Finishing having negative coefficient this low could be possibly just 0. Simply for such numerical models it's very unlikely to have coefficient equal exactly 0. Tl;dr coefficients this low are basically noise and they don't contribute either way into target variable.
Who needs strikers in such scenario?
You mean for the team that plays 'hidden games' in nonplayable league or the one with no friendlies on default?
The issue with B teams is that they are mostly semipro which is kind of awful for development. Different countries do B teams differently though. For example in France a B team is part of the main club, sharing professional status. In most other countries B team is an affiliate.
I am ware of this feature that the team can play games that are no in the schedule. I'm talking here about a case where B team have no games at all.
As always thank you very much for providing crucial evidence!
Anyway, I understand I'm not the target audience of your work, and I thank you for taking the time to elaborate.
Cheers
Then you'd noticed for example the part with 1000 minutes required to be part of the training data set and combining league rank into players data.
The goal was to make rather universal model that will be a more of a guidance for the player what attributes look for than some absolute solution to players selection. I did it mostly for myself out of curiosity and since I already did the work I've decided to share it with the community so others can benefit from it. And then I've decided to update with it attributes for each position (because first model was for all outfield players) and then expanded it with custom match engines.
I'm currently not planning on doing another models dedicated for top leagues because I believe they won't be that different from general models.
ClaudeJ said: My main concern is that the average includes non-playing players, possibly those not even on the squad roster.
It definitely does not because:
1) players that don't play won't have average rating at all - so the target value
2) part of players filtering contains minimum of 1000 minutes played in the league in the season
ClaudeJ said: PS: ingame, each competition has a reputation value, regardless of its continent, and that could serve as a global relative ranking.
Each player has added a custom feature that represents their respective league position (that is based on the league reputation) in Europe. Every player attributes are 'transformed' to not use 'absolute' value of the attribute but it shows difference between players actual value and the average value for the league his playing in, for every single attribute considered.
ClaudeJ said: I understand that refining the dataset would require a lot of additional work, but I truly believe it would better serve its intended purpose by providing a clearer and more accurate representation of attribute importance at a competitive level.
Not everyone plays with top leagues - I do not. I never in FM era played with a team in top 5 leagues.
I'd also make a bet that among people who lurk on this forum there is higher share of players playing in lower leagues than in general population of FM players.
Standardisation should compensate for the 'relativeness' of attributes. Please read those topics mentioned before. It would answer a lot of your questions and concerns.
By the way, I noticed that for the DMC position in the vanilla Match Engine, only ACC has a coefficient, while Pace is 0. However, in all other MEs, these two seem to have the highest coefficients. Compared to all other positions, DMC appears to be quite an outlier in the vanilla ME.
Could you maybe check again to see if there's a mistake in there?
This is only an anecdotal evidence but I had in my playing experience cases where DMC was this particular position that did not required great physicals from the player (maybe beside balance - but this is not that affected by ageing). I had a 36yo veteran playing DLP in DMC slot with far below league and even team average physicals while maintaining one of the best average rating - even thou he had almost no goals and assist so his rating was not 'inflated' with g/a.
I had a plan on doing another 10 years simulation for 'vanilla ME' because now in this testing save it's so far in the game that there are almost no real players, only the newgens. So the results would be not skewed towards real players attributes - and we know that for certain positions real player have different attribute distribution than newgens with Fullbacks/Wingbacks being good example since the newgens there have usually relatively low crossing for example.
By focusing on top teams and excluding young, developing players, the analysis might better capture the attributes that truly influence performance at an elite level. Given that, what do you think about refining the sample to focus on actual top-tier teams and filtering out younger, still-developing players?
I feel this would make the findings more representative of how FM is actually played: primarily in the top five leagues, with most players managing elite squads or near-peers, at least from what I’ve seen in the video content and forums. This would, in turn, make your valuable work even more useful to the community. Otherwise, many might draw the wrong conclusions from it: simply because they don’t read the full accompanying text, for whatever reason.
This issue should be somehow compensate with data being standardised which means attributes for every player are 'relative' to their league average. The question with this method is - are attributes relation is linear - example if league average is 10 pace, and the player has 15 will the effect be the same as for the league having average pace 15 and player 20.
If we wanted to tackle this issue my guess would be - make a graph of league 'rank' and players average CA and based on the results filter out players playing in leagues with high enough 'rank' (just reminder - rank is a league place in continental ranking).
It's doable but would require additional work. Especially exporting again data for all the players for all the seasons for all the engines since I did not include CA in their export data.
Yep, and Jumping Reach. Difference is that sometimes some technical attributes like Passing or Technique are showing. Previously it was mostly just Dribbling.
I knew that teams like that have some 'hidden' fixtures since they don't have them in official schedule
but when you check their players profile page they have some competitive games
So recently I got monthly report about my loan players and I noticed something very weird. They are playing against some random teams around the world
One team is from Burkina Faso
And second one is from Spain
Did anyone had/noticed similarly weird behaviour?
You can just use
GK:
Agility: 100%
Reflex: 87,7%
Aerial Reach: 80,6%
1) That would explain why Work Rate was so crucial attribute in your previous test
2) This somehow force to diverse the players attributes and roles in a way that the player won't benefit from 10 playmakers
3) This could be the reason why games promotes physical attributes - the don't introduce this 'negative' effect
4) It's interesting that this mostly affects decision - so mostly for other attributes still the higher - the better, in general case
I guess the thing that stands out to me upon first inspection is how closely rated many of the players in the same position are (for example trying to pick the best two wingers). Not sure if this is because I'm a top 6 Prem team with a strong squad but I'm not sure I've learnt anything from this sheet.
That said I can see how it would help in trying to work out who to buy to improve the squad (if I can work out how to get potential signings data into the Squad_Data_N tab)
That looks really good.
@DeFlow
If I may suggest one thing - I'd use conditional formatting to highlight player's best position/highest score on the table. I know it shows his best position on the right but highlighting it with very different colour on the table itself would be in my opinion very helpful.
The main post is ok. I'm referring to this reply.
In your experiment, you are comparing players with respect to the league average. I think its safe to assume that in any given league, majority of players are close to some "average" CA value for that league. Lets take 2 strikers from one league with the same CA.
For Strikers Finishing has PA weight of 8; Acceleration has PA weight of 10, Pace has PA weight of 7, Jumping reach 5 and Balance 2!!!
Striker A has 15 finishing.
Striker B has only 10 finishing. But this will free 40 PA that can be redistributed to other attributes. For example:
4 ACC; 5 PAC + 1 JUM; 10 BAL + 4 JUM; and so on.
So I think in the end, Finishing isnt useless or negative. But there are simply better attributes to have for striker. Therefore striker with high FIN can suffer from lack of other essential attributes if the CA should be on similar level.
I have sometimes problems to expres my thoughts, but I home this makes some sence.
I have to admit I might've made a small mistake. The coefficients in this model where I showed all the features in FM23 was probably for all outfield players not just forwards.
My mistake was due to having too much of those models made and I just didn't check which one was which - since I have main program that does the model and just change the input data.
This one experiment was also conducted with the same underlying parameters (A very significant amount of leagues and divisions, running for 11 seasons)?
Sorry if it's misleading but I was doing so many of those test in different settings that I can't really remember if those coefficients that I provided (with all 51 features) was for Strikers or for just overall outfield players. I'll check that tomorrow.
Regarding this test it was performed on the data extracted from the same saves as the FM 24 goalkeepers test so you have all the leagues listed and it was 10 seasons there as far as I remember.
So according to your results, STC having higher Finishing equals having worse average rating?
Keep in mind that this is for difference between actual attribute value and the average for the league. So for example if average finishing was 10, the higher than that would result in -0.000456 per every finishing point. So if the average was 10 and players value was 20 it still results in lowering average rating by 0,00456. So for features with coefficients this small you can basically say that they just don't make any difference. Finishing having negative coefficient this low could be possibly just 0. Simply for such numerical models it's very unlikely to have coefficient equal exactly 0.
Tl;dr coefficients this low are basically noise and they don't contribute either way into target variable.
Yes. A little more details are described in the topics linked in the first post.