New findings: Why is it that the higher some attributes are, the worse the team performs, because the more players with Specific higher attributes, the more negative the effect
1. The standard group consists of 11 players who are all decision 10, with a goal difference of 8.3
Next, each test independently improved one position, full-back, forward, back, centre-back, wingers, decision increased from 10 to 20, goal difference increased (13.5,16.3,11.6,19.1,13.7). Only the striker is slightly down(6.3).
2. If all non-goalkeeper 10 player's decision is increased from 10 to 20 , the goal difference is reduced to 2.8, which is less than the standard value (8.3)
That is, the stats have increased, but the goal difference has decreased The better the stats , the worse the Team goal difference
3. If all non-goalkeeper 10 player's decision is reduced from 10 to 1, the goal difference is reduced to 7.0, less than the standard value (8.3) So the stats are down, the goal difference is down
4. Select 3 players( Attacking Midfielder + Centre Back + Winger ), the decision is increased from 10 to 20, the goal difference is increased from 8.3 to15.8
5. Select 5 players( Attacking Midfielder + Fullback + Defensive midfielder + Centre Back + Winger ) to increase decision from 10 to 20 , goal difference increased from 8.3 to8.5
6. Select 7 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to5.9
7. Select 11 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to 2.8
That is to say, after the number of players with a high value of this attribute increase, it does not produce the effect of "1+1>2", but becomes a negative effect, and becomes a negative attribute.
Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling all seem to have this kind of tendency This happens not only when it go from 10 to the highest 20, but also happen in other like from all-10 to 15 , all-5 to 10, ect
Another finding was that a player's rating did not fully reflect his impact on the team. In individual cases, by improving certain attribute, the player's rating goes up, but the team's score goes down
Other attributes are "1+1>2" mutual cooperation attributes, such as Pace, the higher the Pace of the whole team, the better the team's total result
I haven't finish tested it yet, but I've seen it happen with Technique and Flair Performance like this in different tactics.
From 红骑士Sakura https://www.playgm.cc/thread-971631-1-1.html While modifying the game engine (previous post), he discovered some logically related Stats of the engine's behavior
As you can see, Stats not only affect whether the player can complete better, but also affect how much the player likes to do this action.
For example, Decision , there might be a situation where the team "too likes to do this action Led to negative results " ? We don't know how the engine could have caused this, But from this result, decision of a team can not have too many people at the same time have high attributes,
for example, there are 2-3 players with high decision attributes, the other 7-8 people maintain low attributes, which is good for the team. If this logic is correct (I haven't completely tested it yet), Some player are low skilled in Specific attribute and a few player are high skilled in these, which is good for the team.
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
So basically this means that we have to have players with a spread of attributes to have the best result? Or is just for the attributes (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) you mentioned?
CBP87 said: They are using a different match engine mate, @harvestgreen22 are these tests being done on the normal ME or the modified one? Expand
Are you referring to the "simate.fmf" file (new engine recently)? I used the original, unmodified
Alonso said: So basically this means that we have to have players with a spread of attributes to have the best result? Or is just for the attributes (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) you mentioned? Expand
Only (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) , Only they have the effect of "competing and hurting each other" attribute. (I didn't complete the test completely, so some of the list may be inaccurate or missing)
such as Decision , whole team player Decision = 1 (very low) , only 1 Attacking Midfielder Decision = 20 (only 1 man keep high), goal difference = 12.3 that's much better than whole team player Decision = 20 , goal difference = 2.8 (in the picture above)
Other attributes, usually show as " the more the better", such as Pace, the relationship between attributes is "cooperation", the more the performance is better, that is the so-called "1+1>2" attribute.
I must admit to being confused LOL, it seems attributes needed are changing every few days. What are the main attributes needed for the positions , i know Acceleration , pace, dribbling and Jumping reach seem to be the main ones, but im confused about this thread , what are the main attributes are we looking for , this feels like we need more and more attributes. Is there a list other than the FM Arena one that shows key attributes for GK , fullbacks , Centre backs, DM's, wingers, and Strikers , it could just be me , i havent been able to put as much time into the game or looking on here so im a bit lost tbh
CBP87 said: Thank you for clarifying, I have to admit, I'm not following this at all in regards to the attributes, can you dumb down please? Expand
Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense.
Rhumble said: Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense. Expand
Having high amount of attributes like Decisions across the board makes a teams perform worse but having high on certain positions is an increase in team performance. They think this is due players being more likely to take an action if certain attributes are high and it becomes counterproductive. (Decisions is good on DCs but not good when it’s high everywhere.)
The 2nd point: Higher Player performance Ratings also don’t necessarily lead to more wins. Increasing attributes will increase the individual player rating but team performance will go down. I think the example given is passing on AMC made his rating go up but the gd went down and the ratings were in Chinese.
The meta attributes are ones that don’t cause players to counter act each other and are additive.
TBH, I’m fairly convinced we are just slowly recreating the AI weightings from a couple years ago.
CBP87 said: Thank you for clarifying, I have to admit, I'm not following this at all in regards to the attributes, can you dumb down please? Expand
Average goal difference: It shows how good the team played , each season have 30 match , test for several seasons , Record the goal difference
Decision = 10 ： All attribute is 10 ,including Decision. In the case of nothing changed , the goal difference is 8.3
Decision = 20 (1 player)： 1 of the players Decision +10 : In most cases ,goal difference increased . goal difference higher than 8.3 . This is logical , As attribute increase , goal difference increased .
Decision = 20 (10 player)： All of the players (expect goalkeeper) Decision +10 : the goal difference is 2.8 This is not-logical , As attribute increase , goal difference Decreased .
Decision = 1 (10 player)： All of the players Decision -10 : the goal difference is 7.0 This is not-logical , This means "Player's Decision =10 " > "Player's Decision =1 " > "Player's Decision =20 ".
Decision = 20 (3 player)： Choose 3 of the 10 players , Decision +10 : the goal difference is 15.8 Decision = 20 (5 player)： Choose 5 of the 10 players , Decision +10 : the goal difference is 8.5 Decision = 20 (7 player)： Choose 7 of the 10 players , Decision +10 : the goal difference is 5.9 This means that the more people with "Decision = 20", the worse the team performs
Decision = 20 (11 player)： Choose 10 of the 10 players , Decision +10 : the goal difference is 2.8 When everyone's Decision are high , the team perform the most worse
Summary: ( only refers to specific partial attributes )
No more than 3 players have high Decision. Decision have a positive effect If more than 5 players have high Decision at the same time, Decision have a negative effect . The more people with high Decision , the worse the team does . The worst team you can have is whole team with Decision 20 .
This is completely unsimulated realistic ——It's just the mechanics of the game
harvestgreen22 said: Average goal difference: It shows how good the team played , each season have 30 match , test for several seasons , Record the goal difference
Decision = 10 ： All attribute is 10 ,including Decision. In the case of nothing changed , the goal difference is 8.3
Decision = 20 (1 player)： 1 of the players Decision +10 : In most cases ,goal difference increased . goal difference higher than 8.3 . This is logical , As attribute increase , goal difference increased .
Decision = 20 (10 player)： All of the players (expect goalkeeper) Decision +10 : the goal difference is 2.8 This is not-logical , As attribute increase , goal difference Decreased .
Decision = 1 (10 player)： All of the players Decision -10 : the goal difference is 7.0 This is not-logical , This means "Player's Decision =10 " > "Player's Decision =1 " > "Player's Decision =20 ".
Decision = 20 (3 player)： Choose 3 of the 10 players , Decision +10 : the goal difference is 15.8 Decision = 20 (5 player)： Choose 5 of the 10 players , Decision +10 : the goal difference is 8.5 Decision = 20 (7 player)： Choose 7 of the 10 players , Decision +10 : the goal difference is 5.9 This means that the more people with "Decision = 20", the worse the team performs
Decision = 20 (11 player)： Choose 10 of the 10 players , Decision +10 : the goal difference is 2.8 When everyone's Decision are high , the team perform the most worse
Summary: ( only refers to specific partial attributes )
No more than 3 players have high Decision. Decision have a positive effect If more than 5 players have high Decision at the same time, Decision have a negative effect . The more people with high Decision , the worse the team does . The worst team you can have is whole team with Decision 20 .
This is completely unsimulated realistic ——It's just the mechanics of the game Expand
Thank you for explaining. I understand it now. Great work. May I ask, have you got an opinion on the positions that should have high decision making please?
Hello @harvestgreen22 , thanks again for your work! I have some questions about your test setup. I've been using the same base-file that you seem to be using (from EBFM), frozen with FMRTE, etc.
When doing some tactics testing, I've noticed that even when simulating 10-20 seasons, I couldn't reproduce the same results on average in many cases, because some variables can't be frozen (like player links). Have you been able to reproduce your findings here?
I'm also curious why the "baseline" has a positive goal difference, shouldn't a baseline roughly have +0?
Padovan said: Did I see that right? Having 1 in first touch improves 4 goals compared to having 20? Damn, the ME is completely broken. Expand There is some error, I wouldn't make huge conclusions with 4 difference, but we can probably say that difference between 1 to 20 first touch isn't much.
Rhumble said: Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense. Expand It's not that complicated. Check the table posted. Perhaps easiest is to look at the difference between 1 and 20 for a certain attribute. The bigger the difference more important it is. Couple of examples: - acceleration: at 1 is -103,1 goal difference and at 20 it's +88,2. So that is roughly 190 difference between 1 and 20, a huge effect - off the ball: at 1 is 8,3; at 20 is 10,4. A difference of only 2 goals which is well within the margin of error meaning on a team level it's not doing much.
Seems to support the hypothesis that some of the attributes are good when on a limited number of players and/or role specific. Back half he starts editing attributes.
Doesn’t really change the fact that signing Fast players is best but I did find it interesting.
svonn said: Hello @harvestgreen22 , thanks again for your work! I have some questions about your test setup. I've been using the same base-file that you seem to be using (from EBFM), frozen with FMRTE, etc.
When doing some tactics testing, I've noticed that even when simulating 10-20 seasons, I couldn't reproduce the same results on average in many cases, because some variables can't be frozen (like player links). Have you been able to reproduce your findings here?
I'm also curious why the "baseline" has a positive goal difference, shouldn't a baseline roughly have +0? Expand
This question has been discussed in the Chinese forum community, and it goes something like this: 1.FMRTE takes a few seconds to lock. Because the way it works is it checks every few seconds. The software is not designed to be used for testing, it is designed to meet the needs of popular players, so it may be checked every few seconds to reduce memory /CPU usage. 2. If the Test league's match schedule is too dense, there is a chance to lock in failures (number of occurrences: normal) 3. When the amount of player data is too large, some players have the opportunity to lock failure (number of occurrences: rare) 4. For AI-controlled teams, lock often fails (frequency of occurrence: normal) . This failure locking Creat the most the Biggest Error in testing . 5. Locking methods include FMRTE locking and in-game editor locking
Some of these conclusions may be wrong or right, but to be on the safe side, we treat them all as if they need to be addressed
For the above 5 points, the tentative conclusion obtained after discussion is as follows: 1 and 2: Test the leagues used, the matches are not too dense, leaving enough time for FMRTE to re-lock 3. don't have too many teams and players 4. It is best not to have any AI teams, all teams used for testing should create a "player manager" to take over. If there must be an AI team, it is best not to change any the attributes of the players in the AI team after creat the league (because after modifying the attributes, I do not know why the attributes can easily get out of control and become completely different , When you change the AI's player's "current CA" to be the same as the" recommended CA", it sometimes randomizes the attributes immediately, even if you increase the PA to "PA = recommended CA +10" , or adjust the PA to 200.). 5. Use both types of locks
In the end, we think we need to sacrifice a little "universality" and choose the EBFM league with only 4 teams, It has fewer players, less intensive games (30 match per team and 120 per season), and all teams are operated by humans, which basically avoids the problems mentioned above The downside is that it's less race-intensive, so you need a lot more time to test the same sample size
baseline : It is the goal difference of team D at full attribute 10 (unmodified), the translation may be different, you can understand its role. If you need to be more "normalized," you can subtract all goal difference results from this standard value
Choosing different initial conditions will vary this criterion. For example, in the test, we chose to let team A and Team C get the better attributes (all 11), Team B get the best attributes (all 12), and Team D get the best tactics (all 10 attributes).
Yarema said: There is some error, I wouldn't make huge conclusions with 4 difference, but we can probably say that difference between 1 to 20 first touch isn't much. Expand
Yes, as you can see from the excel, The statistical standard deviation (" =STDEV() "in excel) is generally around 15, And the "standard error" varies depending on the sample size, A float range of +-4 is normal (error due to insufficient sample)
Thank you for that detailed response! Very interesting findings regarding the locking mechanism. I've also been using the same test setup, how many games / seasons is your sample size?
1. The standard group consists of 11 players who are all decision 10, with a goal difference of 8.3
Next, each test independently improved one position, full-back, forward, back, centre-back, wingers, decision increased from 10 to 20, goal difference increased (13.5,16.3,11.6,19.1,13.7). Only the striker is slightly down(6.3).
2. If all non-goalkeeper 10 player's decision is increased from 10 to 20 , the goal difference is reduced to 2.8, which is less than the standard value (8.3)
That is, the stats have increased, but the goal difference has decreased The better the stats , the worse the Team goal difference
3. If all non-goalkeeper 10 player's decision is reduced from 10 to 1, the goal difference is reduced to 7.0, less than the standard value (8.3) So the stats are down, the goal difference is down
4. Select 3 players( Attacking Midfielder + Centre Back + Winger ), the decision is increased from 10 to 20, the goal difference is increased from 8.3 to15.8
5. Select 5 players( Attacking Midfielder + Fullback + Defensive midfielder + Centre Back + Winger ) to increase decision from 10 to 20 , goal difference increased from 8.3 to8.5
6. Select 7 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to5.9
7. Select 11 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to 2.8
That is to say, after the number of players with a high value of this attribute increase, it does not produce the effect of "1+1>2", but becomes a negative effect, and becomes a negative attribute.
Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling all seem to have this kind of tendency This happens not only when it go from 10 to the highest 20, but also happen in other like from all-10 to 15 , all-5 to 10, ect
Another finding was that a player's rating did not fully reflect his impact on the team. In individual cases, by improving certain attribute, the player's rating goes up, but the team's score goes down
Other attributes are "1+1>2" mutual cooperation attributes, such as Pace, the higher the Pace of the whole team, the better the team's total result
I haven't finish tested it yet, but I've seen it happen with Technique and Flair Performance like this in different tactics.
From 红骑士Sakura https://www.playgm.cc/thread-971631-1-1.html While modifying the game engine (previous post), he discovered some logically related Stats of the engine's behavior
As you can see, Stats not only affect whether the player can complete better, but also affect how much the player likes to do this action.
For example, Decision , there might be a situation where the team "too likes to do this action Led to negative results " ? We don't know how the engine could have caused this, But from this result, decision of a team can not have too many people at the same time have high attributes,
for example, there are 2-3 players with high decision attributes, the other 7-8 people maintain low attributes, which is good for the team. If this logic is correct (I haven't completely tested it yet), Some player are low skilled in Specific attribute and a few player are high skilled in these, which is good for the team. Expand
great work @harvestgreen22 , your tests are a great help to the community. i think the results from decisions would also apply to attributes like Flair ( for amcs), Teamwork (for dms) aggression (for dms) and technique (for dms, mcs,and amcs)
svonn said: Thank you for that detailed response! Very interesting findings regarding the locking mechanism. I've also been using the same test setup, how many games / seasons is your sample size? Expand
In terms of the size of the standard deviation, about 10-15 seasons (30 games per season) can get a sufficiently accurate value , You can compare the accuracy of different sample sizes directly in that excel
Middleweight165 said: Does this information change anything about the important attributes for each position from the opening paragraph in the following post?
https://pixeldrain.com/u/m1QY3CTr
Test league
https://pixeldrain.com/u/Lf1YNXaC
Test data (I didn't translate it)
1. The standard group consists of 11 players who are all decision 10, with a goal difference of 8.3
Next, each test independently improved one position, full-back, forward, back, centre-back, wingers, decision increased from 10 to 20, goal difference increased (13.5,16.3,11.6,19.1,13.7).
Only the striker is slightly down(6.3).
2. If all non-goalkeeper 10 player's decision is increased from 10 to 20 , the goal difference is reduced to 2.8, which is less than the standard value (8.3)
That is, the stats have increased, but the goal difference has decreased
The better the stats , the worse the Team goal difference
3. If all non-goalkeeper 10 player's decision is reduced from 10 to 1, the goal difference is reduced to 7.0, less than the standard value (8.3)
So the stats are down, the goal difference is down
4. Select 3 players( Attacking Midfielder + Centre Back + Winger ), the decision is increased from 10 to 20, the goal difference is increased from 8.3 to15.8
5. Select 5 players( Attacking Midfielder + Fullback + Defensive midfielder + Centre Back + Winger ) to increase decision from 10 to 20 , goal difference increased from 8.3 to8.5
6. Select 7 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to5.9
7. Select 11 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to 2.8
That is to say, after the number of players with a high value of this attribute increase,
it does not produce the effect of "1+1>2", but becomes a negative effect, and becomes a negative attribute.
Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling all seem to have this kind of tendency
This happens not only when it go from 10 to the highest 20, but also happen in other like from all-10 to 15 , all-5 to 10, ect
Another finding was that a player's rating did not fully reflect his impact on the team. In individual cases, by improving certain attribute, the player's rating goes up, but the team's score goes down
Other attributes are "1+1>2" mutual cooperation attributes, such as Pace, the higher the Pace of the whole team, the better the team's total result
I haven't finish tested it yet,
but I've seen it happen with Technique and Flair Performance like this in different tactics.
From 红骑士Sakura
https://www.playgm.cc/thread-971631-1-1.html
While modifying the game engine (previous post), he discovered some logically related Stats of the engine's behavior
As you can see, Stats not only affect whether the player can complete better, but also affect how much the player likes to do this action.
For example, Decision , there might be a situation where the team "too likes to do this action Led to negative results " ?
We don't know how the engine could have caused this,
But from this result, decision of a team can not have too many people at the same time have high attributes,
for example, there are 2-3 players with high decision attributes, the other 7-8 people maintain low attributes, which is good for the team.
If this logic is correct (I haven't completely tested it yet), Some player are low skilled in Specific attribute and a few player are high skilled in these, which is good for the team.
.
This is great lol. Now we know best attributes AND tactic. RIP my casual friend I play FM with.
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
@Bafici do your thing please mate for the GS weightings
KingChazza said: This is great lol. Now we know best attributes AND tactic. RIP my casual friend I play FM with.
Hey mate. Which tactic do you suggest for new match engines?
Teremin said: Hey mate. Which tactic do you suggest for new match engines?
Hey, I don't think the match engine has changed recently. I use this one
KingChazza said: Hey, I don't think the match engine has changed recently. I use this one
They are using a different match engine mate, @harvestgreen22 are these tests being done on the normal ME or the modified one?
So basically this means that we have to have players with a spread of attributes to have the best result? Or is just for the attributes (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) you mentioned?
CBP87 said: They are using a different match engine mate, @harvestgreen22 are these tests being done on the normal ME or the modified one?
Are you referring to the "simate.fmf" file (new engine recently)? I used the original, unmodified
Alonso said: So basically this means that we have to have players with a spread of attributes to have the best result? Or is just for the attributes (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) you mentioned?
Only (Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling) , Only they have the effect of "competing and hurting each other" attribute. (I didn't complete the test completely, so some of the list may be inaccurate or missing)
such as Decision , whole team player Decision = 1 (very low) , only 1 Attacking Midfielder Decision = 20 (only 1 man keep high), goal difference = 12.3
that's much better than whole team player Decision = 20 , goal difference = 2.8 (in the picture above)
Other attributes, usually show as " the more the better", such as Pace, the relationship between attributes is "cooperation", the more the performance is better, that is the so-called "1+1>2" attribute.
Thank you for clarifying, I have to admit, I'm not following this at all in regards to the attributes, can you dumb down please?
I must admit to being confused LOL, it seems attributes needed are changing every few days.
What are the main attributes needed for the positions , i know Acceleration , pace, dribbling and Jumping reach seem to be the main ones, but im confused about this thread , what are the main attributes are we looking for , this feels like we need more and more attributes.
Is there a list other than the FM Arena one that shows key attributes for GK , fullbacks , Centre backs, DM's, wingers, and Strikers , it could just be me , i havent been able to put as much time into the game or looking on here so im a bit lost tbh
CBP87 said: Thank you for clarifying, I have to admit, I'm not following this at all in regards to the attributes, can you dumb down please?
Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense.
Rhumble said: Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense.
Having high amount of attributes like Decisions across the board makes a teams perform worse but having high on certain positions is an increase in team performance. They think this is due players being more likely to take an action if certain attributes are high and it becomes counterproductive. (Decisions is good on DCs but not good when it’s high everywhere.)
The 2nd point:
Higher Player performance Ratings also don’t necessarily lead to more wins. Increasing attributes will increase the individual player rating but team performance will go down. I think the example given is passing on AMC made his rating go up but the gd went down and the ratings were in Chinese.
The meta attributes are ones that don’t cause players to counter act each other and are additive.
TBH, I’m fairly convinced we are just slowly recreating the AI weightings from a couple years ago.
CBP87 said: Thank you for clarifying, I have to admit, I'm not following this at all in regards to the attributes, can you dumb down please?
Average goal difference:
It shows how good the team played , each season have 30 match , test for several seasons , Record the goal difference
Decision = 10 ：
All attribute is 10 ,including Decision.
In the case of nothing changed , the goal difference is 8.3
Decision = 20 (1 player)：
1 of the players Decision +10 : In most cases ,goal difference increased .
goal difference higher than 8.3 .
This is logical , As attribute increase , goal difference increased .
Decision = 20 (10 player)：
All of the players (expect goalkeeper) Decision +10 : the goal difference is 2.8
This is not-logical , As attribute increase , goal difference Decreased .
Decision = 1 (10 player)：
All of the players Decision -10 : the goal difference is 7.0
This is not-logical , This means "Player's Decision =10 " > "Player's Decision =1 " > "Player's Decision =20 ".
Decision = 20 (3 player)：
Choose 3 of the 10 players , Decision +10 : the goal difference is 15.8
Decision = 20 (5 player)：
Choose 5 of the 10 players , Decision +10 : the goal difference is 8.5
Decision = 20 (7 player)：
Choose 7 of the 10 players , Decision +10 : the goal difference is 5.9
This means that the more people with "Decision = 20", the worse the team performs
Decision = 20 (11 player)：
Choose 10 of the 10 players , Decision +10 : the goal difference is 2.8
When everyone's Decision are high , the team perform the most worse
Summary: ( only refers to specific partial attributes )
No more than 3 players have high Decision. Decision have a positive effect
If more than 5 players have high Decision at the same time, Decision have a negative effect .
The more people with high Decision , the worse the team does .
The worst team you can have is whole team with Decision 20 .
This is completely unsimulated realistic ——It's just the mechanics of the game
Did I see that right? Having 1 in first touch improves 4 goals compared to having 20? Damn, the ME is completely broken.
harvestgreen22 said: Average goal difference:
It shows how good the team played , each season have 30 match , test for several seasons , Record the goal difference
Decision = 10 ：
All attribute is 10 ,including Decision.
In the case of nothing changed , the goal difference is 8.3
Decision = 20 (1 player)：
1 of the players Decision +10 : In most cases ,goal difference increased .
goal difference higher than 8.3 .
This is logical , As attribute increase , goal difference increased .
Decision = 20 (10 player)：
All of the players (expect goalkeeper) Decision +10 : the goal difference is 2.8
This is not-logical , As attribute increase , goal difference Decreased .
Decision = 1 (10 player)：
All of the players Decision -10 : the goal difference is 7.0
This is not-logical , This means "Player's Decision =10 " > "Player's Decision =1 " > "Player's Decision =20 ".
Decision = 20 (3 player)：
Choose 3 of the 10 players , Decision +10 : the goal difference is 15.8
Decision = 20 (5 player)：
Choose 5 of the 10 players , Decision +10 : the goal difference is 8.5
Decision = 20 (7 player)：
Choose 7 of the 10 players , Decision +10 : the goal difference is 5.9
This means that the more people with "Decision = 20", the worse the team performs
Decision = 20 (11 player)：
Choose 10 of the 10 players , Decision +10 : the goal difference is 2.8
When everyone's Decision are high , the team perform the most worse
Summary: ( only refers to specific partial attributes )
No more than 3 players have high Decision. Decision have a positive effect
If more than 5 players have high Decision at the same time, Decision have a negative effect .
The more people with high Decision , the worse the team does .
The worst team you can have is whole team with Decision 20 .
This is completely unsimulated realistic ——It's just the mechanics of the game
Thank you for explaining. I understand it now. Great work. May I ask, have you got an opinion on the positions that should have high decision making please?
Hello @harvestgreen22 , thanks again for your work!
I have some questions about your test setup. I've been using the same base-file that you seem to be using (from EBFM), frozen with FMRTE, etc.
When doing some tactics testing, I've noticed that even when simulating 10-20 seasons, I couldn't reproduce the same results on average in many cases, because some variables can't be frozen (like player links).
Have you been able to reproduce your findings here?
I'm also curious why the "baseline" has a positive goal difference, shouldn't a baseline roughly have +0?
Padovan said: Did I see that right? Having 1 in first touch improves 4 goals compared to having 20? Damn, the ME is completely broken.
There is some error, I wouldn't make huge conclusions with 4 difference, but we can probably say that difference between 1 to 20 first touch isn't much.
Rhumble said: Yep , im lost tbh , it could just be me or something is missing somewhere but i just dont see the information in laymans terms, i read and reread the opening post but whether the translation from Chinese to English isnt quite getting through i dont know, if we could know the attributes needed from important down for each position then it might start to make sense.
It's not that complicated. Check the table posted. Perhaps easiest is to look at the difference between 1 and 20 for a certain attribute. The bigger the difference more important it is. Couple of examples:
- acceleration: at 1 is -103,1 goal difference and at 20 it's +88,2. So that is roughly 190 difference between 1 and 20, a huge effect
- off the ball: at 1 is 8,3; at 20 is 10,4. A difference of only 2 goals which is well within the margin of error meaning on a team level it's not doing much.
Seems to support the hypothesis that some of the attributes are good when on a limited number of players and/or role specific. Back half he starts editing attributes.
Doesn’t really change the fact that signing Fast players is best but I did find it interesting.
svonn said: Hello @harvestgreen22 , thanks again for your work!
I have some questions about your test setup. I've been using the same base-file that you seem to be using (from EBFM), frozen with FMRTE, etc.
When doing some tactics testing, I've noticed that even when simulating 10-20 seasons, I couldn't reproduce the same results on average in many cases, because some variables can't be frozen (like player links).
Have you been able to reproduce your findings here?
I'm also curious why the "baseline" has a positive goal difference, shouldn't a baseline roughly have +0?
This question has been discussed in the Chinese forum community, and it goes something like this:
1.FMRTE takes a few seconds to lock. Because the way it works is it checks every few seconds.
The software is not designed to be used for testing, it is designed to meet the needs of popular players, so it may be checked every few seconds to reduce memory /CPU usage.
2. If the Test league's match schedule is too dense, there is a chance to lock in failures (number of occurrences: normal)
3. When the amount of player data is too large, some players have the opportunity to lock failure (number of occurrences: rare)
4. For AI-controlled teams, lock often fails (frequency of occurrence: normal) . This failure locking Creat the most the Biggest Error in testing .
5. Locking methods include FMRTE locking and in-game editor locking
Some of these conclusions may be wrong or right, but to be on the safe side, we treat them all as if they need to be addressed
For the above 5 points, the tentative conclusion obtained after discussion is as follows:
1 and 2: Test the leagues used, the matches are not too dense, leaving enough time for FMRTE to re-lock
3. don't have too many teams and players
4. It is best not to have any AI teams, all teams used for testing should create a "player manager" to take over.
If there must be an AI team, it is best not to change any the attributes of the players in the AI team after creat the league
(because after modifying the attributes, I do not know why the attributes can easily get out of control and become completely different ,
When you change the AI's player's "current CA" to be the same as the" recommended CA", it sometimes randomizes the attributes immediately, even if you increase the PA to "PA = recommended CA +10" , or adjust the PA to 200.).
5. Use both types of locks
In the end,
we think we need to sacrifice a little "universality" and choose the EBFM league with only 4 teams,
It has fewer players, less intensive games (30 match per team and 120 per season), and all teams are operated by humans, which basically avoids the problems mentioned above
The downside is that it's less race-intensive, so you need a lot more time to test the same sample size
baseline :
It is the goal difference of team D at full attribute 10 (unmodified), the translation may be different, you can understand its role. If you need to be more "normalized," you can subtract all goal difference results from this standard value
Choosing different initial conditions will vary this criterion. For example, in the test, we chose to let team A and Team C get the better attributes (all 11), Team B get the best attributes (all 12), and Team D get the best tactics (all 10 attributes).
Yarema said: There is some error, I wouldn't make huge conclusions with 4 difference, but we can probably say that difference between 1 to 20 first touch isn't much.
Yes, as you can see from the excel,
The statistical standard deviation (" =STDEV() "in excel) is generally around 15,
And the "standard error" varies depending on the sample size,
A float range of +-4 is normal (error due to insufficient sample)
Thank you for that detailed response! Very interesting findings regarding the locking mechanism. I've also been using the same test setup, how many games / seasons is your sample size?
Does this information change anything about the important attributes for each position from the opening paragraph in the following post?
https://fm-arena.com/thread/14201-fm24-experiment-most-important-attributes-for-each-respecitve-positions-with-their-coefficients/page-1/
harvestgreen22 said: https://pixeldrain.com/u/m1QY3CTr
Test league
https://pixeldrain.com/u/Lf1YNXaC
Test data (I didn't translate it)
1. The standard group consists of 11 players who are all decision 10, with a goal difference of 8.3
Next, each test independently improved one position, full-back, forward, back, centre-back, wingers, decision increased from 10 to 20, goal difference increased (13.5,16.3,11.6,19.1,13.7).
Only the striker is slightly down(6.3).
2. If all non-goalkeeper 10 player's decision is increased from 10 to 20 , the goal difference is reduced to 2.8, which is less than the standard value (8.3)
That is, the stats have increased, but the goal difference has decreased
The better the stats , the worse the Team goal difference
3. If all non-goalkeeper 10 player's decision is reduced from 10 to 1, the goal difference is reduced to 7.0, less than the standard value (8.3)
So the stats are down, the goal difference is down
4. Select 3 players( Attacking Midfielder + Centre Back + Winger ), the decision is increased from 10 to 20, the goal difference is increased from 8.3 to15.8
5. Select 5 players( Attacking Midfielder + Fullback + Defensive midfielder + Centre Back + Winger ) to increase decision from 10 to 20 , goal difference increased from 8.3 to8.5
6. Select 7 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to5.9
7. Select 11 players, increase decision from 10 to 20 , goal difference decreased from 8.3 to 2.8
That is to say, after the number of players with a high value of this attribute increase,
it does not produce the effect of "1+1>2", but becomes a negative effect, and becomes a negative attribute.
Decision, Positioning, Technique, Flair ,Off the ball , First touch, Tackling all seem to have this kind of tendency
This happens not only when it go from 10 to the highest 20, but also happen in other like from all-10 to 15 , all-5 to 10, ect
Another finding was that a player's rating did not fully reflect his impact on the team. In individual cases, by improving certain attribute, the player's rating goes up, but the team's score goes down
Other attributes are "1+1>2" mutual cooperation attributes, such as Pace, the higher the Pace of the whole team, the better the team's total result
I haven't finish tested it yet,
but I've seen it happen with Technique and Flair Performance like this in different tactics.
From 红骑士Sakura
https://www.playgm.cc/thread-971631-1-1.html
While modifying the game engine (previous post), he discovered some logically related Stats of the engine's behavior
As you can see, Stats not only affect whether the player can complete better, but also affect how much the player likes to do this action.
For example, Decision , there might be a situation where the team "too likes to do this action Led to negative results " ?
We don't know how the engine could have caused this,
But from this result, decision of a team can not have too many people at the same time have high attributes,
for example, there are 2-3 players with high decision attributes, the other 7-8 people maintain low attributes, which is good for the team.
If this logic is correct (I haven't completely tested it yet), Some player are low skilled in Specific attribute and a few player are high skilled in these, which is good for the team.
great work @harvestgreen22 , your tests are a great help to the community.
i think the results from decisions would also apply to attributes like Flair ( for amcs), Teamwork (for dms) aggression (for dms) and technique (for dms, mcs,and amcs)
svonn said: Thank you for that detailed response! Very interesting findings regarding the locking mechanism. I've also been using the same test setup, how many games / seasons is your sample size?
In terms of the size of the standard deviation, about 10-15 seasons (30 games per season) can get a sufficiently accurate value ,
You can compare the accuracy of different sample sizes directly in that excel
Middleweight165 said: Does this information change anything about the important attributes for each position from the opening paragraph in the following post?
https://fm-arena.com/thread/14201-fm24-experiment-most-important-attributes-for-each-respecitve-positions-with-their-coefficients/page-1/
good , I'll write it down and look at it later when i have time, I'm busy in things of life right now
harvestgreen22 said: (...)
From 红骑士Sakura
https://www.playgm.cc/thread-971631-1-1.html
While modifying the game engine (previous post), he discovered some logically related Stats of the engine's behavior
(...)
We know that the engine is only used in Full Details mode.
Are you test performed using Full Details mode too, or are simulated in another mode ?
PS: also, what tactic and roles are you using in your test ?