Zippo said: No, of course not... we've been working hard on our testing DB to make it less and less representative of normal gameplay because at the end our goal is finding good tactics to play FIFA and PES and not FM.
Pfff... it would be ridiculous to assume that we've been updating our DB to make it better for finding good tactics to play FM! The less our DB represents normal gameplay, the better for us! Isn't it obvious?
I can only take a wild guess here... it could be that we want to keep the result of our hard work on learning the FM privately for ourselves? Yes, I know it's unbelievable selfishness but unfortunately, we're that evil. Once more, it's just my wild guess. Expand
His question may have been worded rather rudely, but do you think it’s helpful to reply like that? Why not just give a straightforward answer and if you need to be vague to protect your work then say that.
Personally I remove get stuck in from all tactics since I don’t believe it does anything if the players already have tackle harder. Then like Mark I ease off tackles for anyone getting yellows. It’s pretty rare for me to get more than one or two red cards per season. Getting a lot of yellows doesn’t bother me; sure I get some suspensions for accumulated yellows but I rotate a lot anyway so it’s just rearranging which games the player gets rested. I do have a pretty strict code of conduct ( it’s either the one recommended by ZaZ or a slight modification).
Not sure what that even means, but changing the number of tests based on the result just introduces bias. There’s now no way to know if the result is more or less accurate than before.
Any thoughts on the default selection of players for various set piece roles? I noticed that aerial threats seem to be selected based on an average of jumping reach, heading and strength, but this does not seem optimal. Jumping reach seems much more important than the other two; some level of heading is needed to get the ball on target, but I can’t tell if strength has any benefit at all.
pixar said: My friend, with your method, my injuries have decreased significantly and my players continue their physical-mental-technical development. I do not give my players individual training. I only provide weak foot strengthening training for my eligible players. They usually solve it in 1 season. I do not give my players manual holidays. But I play in rotation. I play with my main squad in Europe, with my reserve squad in the matches I play at home before or after the European matches in the league, with my main squad in the remaining matches and finally with my reserve squad in the local cups.
The key to this program is to give players full rest (for 3 sessions) the day after match days. If there is no match on Wednesday-Saturday, it is just a corner routine or shot stopping training for goalkeepers. Additionally, if there is no match on Wednesday, I added team bonding training to the evening session to boost team harmony and motivation.
I played 6 seasons with 24.3.0 using this method. I play my matches on Wednesday-Saturday.
Below is a preview of the training program and download links.
Cheers.
Expand
Are you actually doing what ZaZ said, because you also say you don’t manually rest players? If you just have 3 rests in training and the players are “in training” then they recover substantially less than 3 rests plus manual rest for the day. It’s still better than scattering the 3 rests on different days but not a “super rest”. The strangest thing in the testing done by EBFM and ZaZ is that players on manual rest, no pitch or gym work, or even holiday, are affected by the training that the rest of the team is doing. This has to be a bug, but doing even one training session on the day makes all the players, including resting and on holiday, recover less. I suspect this bug is also part of what makes recovery sessions useless.
I always train 3 tactics with different shapes but I only change shape if the games is going badly (to me this means losing on the scoreboard and on xG during the second half). Otherwise I adjust the intensity and maybe a few specific instructions. Your question is more about automating the game with a match plan so you are going to be very limited with only the starting tactic plus two modifications. The most common change I make is when winning easily and it’s just designed to save fitness and reduce yellow cards. The next most common would be either defending a narrow lead (reduce mentality, have the GK slow things down, maybe time waste in the last 10 minutes), or trying to get a goal (increase mentality, maybe work the ball into the box, maybe move some players from support to attack). One thing you just can’t do with match plans is identify who is playing really badly and sub them out; this will be as or more effective than any tactical change.
The stamina increase at half time isn’t relevant since the replacement will always have more stamina. You also can’t start your best 11 every single game, so if you are replacing selectively (not substituting your star player when he’s playing well) then its quite possible the substitute will be better. You may also be winning comfortably at half time and bring on players to grow attributes. I would agree with ZaZ, and I’m sure he wasn’t saying to substitute early if it’s the CL final and the sub is clearly worse.
P.S. I predict, some people will rush to remove "Overlaps" TIs from the top tactics in hope to see their scores improves but I can assure your that highly likely it won't happen because almost all tactics at the top already hit a quite high positive RNG which is about +2 points so imagine you remove the Overlaps TI from the top tactics which would give +1 points but you struck -2 points negative RNG after 4,000 matches and as result you end up with a worse score. Expand
Your general point is correct that making this change is unlikely to lead to a higher score in the test league, but the logic is wrong since what you refer to is actually ranking bias which is not the same as the random error in estimating the score. The size of the ranking bias is not possible to calculate given the nature of the submissions to the ranking table, but it would only be +2 by an unlikely coincidence.
I think more relevant is that for use in the game, removing overlaps would be expected to be favorable (at least for the role combinations used). Anyway thanks for the useful experiment.
Meriten said: It's really strange when you say "Only 9 attributes matter but I set other attributes to 13 because these attributes matter too".
I've been testing a lot lately and have regained the joy of FM (I haven't bought an FM every year in the last few years because it's become too easy for me). But the results of tests should always be presented objectively and the Reddit user didn't do that. All attributes matter and it is interesting to find out how important each attribute is for each position. I would beat e.g. this reddit users team with my 19 acceleration, 19 pace, 19 anticipation, 19 jumping reach, 20 finishing, 20 technique, 20 first touch, 20 heading striker. This problem is more complex than reddit users posts. Yes, we found meta-attributes but this isn't the end of the research. Expand
Tbf one experiment involved only the 9 attributes, the other did involve 4 others but they didn’t matter in the sense that a mediocre value was sufficient. I think you are interpreting the word “matter” different to me. This doesn’t mean they do absolutely nothing, only that you really don’t have to pay attention to those attributes. If I sign a PL quality player then he probably will have at least 10 passing but I am never looking for the guy with 20 passing because there is always someone else who has better physicals and that’s what matters in the match engine.
Yarema said: Dribbling is one of the 9 at 20, super important. In fact dribbling attribute isn't even dribbling in the classical sense, but more like ball control.
Anyway, I think anyone visiting this site is hardly surprised by the results. Expand
Thanks for pointing that out, I was writing from memory. The other one set to 13 was work rate. I’ll fix my post.
Since you can’t see the original post now, one point that isn’t clear in the text but was shown in images is that in test number 2 there were 4 attributes being set to 13 (this is why the word “almost” is used). These were work rate, determination, natural fitness and stamina. I think it’s obvious that with natural fitness 1 and stamina 1 your players will be exhausted most of the time and either injured or playing badly. There was a big firestorm on Reddit but the main points were:
- despite several claims to the contrary the experiments were done correctly and verified by multiple people, so the 9 attributes are the important ones (as this site has shown repeatedly)
- you need certain other attributes to be able to use the 9 effectively. Fitness and stamina are obvious ; work rate and determination less so but you don’t want a really low value.
- Dribbling is interesting; if you dropped this down to 1 and set one of the other technicals to 20 performance crashed. So your super fast players do need to be able to run with the ball.
- Of course the profile of player used in experiment 2 doesn’t really exist in the game since the initial attributes are not completely random, and can only grow 5-6 points at most (and with limited user input). So you are not going to find some 80 CA player that can rip up the PL while on £50 a week. In some searching on a large database I did find a couple of 120-130 CA players that were so high on the 9 attributes that they could probably be world class. But the main conclusion is that when selecting players you can just focus on 9 attributes being as high as possible and it’s ok if some of the other attributes look really bad with the only red flags being low stamina, fitness, determination or work rate.
Edit: made some corrections based on Yarema’s comment.
Hope it’s ok to post a link to Reddit, since this experiment is based off FM Arena research. CA 180+ version of Man City gets totally destroyed by CA 90 version of Man City. I am not the OP on Reddit.
Meriten said: I don't agree. You have to vary one attribute for one position to know its value for this position (in a particular tactic). If you vary multiple attributes than you don't know which attribute is important for the performance increase.
Spoiler
Why do scientists change only one variable in a controlled experiment?
In a controlled experiment, scientists change only one variable at a time in order to accurately determine the effect of that specific variable on the outcome. By isolating and manipulating one variable while keeping all others constant, scientists can establish a cause-and-effect relationship between the variable they are testing and the results they observe. This helps in drawing accurate conclusions and making reliable predictions based on the experiment's findings. If multiple variables were changed at the same time, it would be difficult to attribute any observed effects to a specific variable, leading to unreliable or inconclusive results.
Neither you or the quote are correct in the context we are discussing, which is optimizing performance of a team with many positions and attributes. If you think of performance as a multi-dimensional surface, changing one variable only tells you that performance is sloping upwards in that region of the surface. But this can be towards only a local maximum, not the global maximum. Think of climbing a mountain; if the steepest path from where I am standing is North then will going North always lead me to the summit? No, if I looked around me I might see the summit is to the east and that what is North is just in a smaller peak below the main one. Or suppose I test the strength of a bond by varying the amount of glue; what will happen is that strength will increase up to a certain point, but beyond that more glue will actually weaken the bond. This is because I failed to consider another variable, the size of the surface to be bonded; this and the amount of glue interact and the optimum is actually a particular thickness of the the glue applied. I could go on about chemotherapy’s that are effective when used on their own but are useless for practical purposes because they add nothing but side effects when used with other more effective treatments. But you hopefully get the point. There are simplistic problems where it’s possible to isolate a single variable, but FM is really not such a case.
Lapidus said: No way that Stamina can be impactful as Acceleration and Pace for any positions.
In the FM-Arena test Stamina attribute was increased by 5 points for every positions including the DMs and WBs and that improved the result just by 2 points from 60 pts to 62 pts, which honestly can be just the RNG but when Acceleration or Pace attributes were increased by 5 then it improved the result by 21 points from 60 pts to 81 pts.
So no way Stamina attribute or any other attribute can be as impactful as Pace and Acceleration for any position.
I don’t think your logic is correct. When you test attributes one by one and for all positions it can only tell you if an attribute is important for the majority of positions. The only way to know at a single position level is vary multiple attributes in the same experiment and vary by position as well (this is in effect what the machine learning data is reporting). The best you can say about stamina is that it is not important for the majority of positions, and in a very controlled setting where players are always starting games at peak performance and have equally capable replacements available on the bench.
I think it’s important to remember that players 21+ are not improving with training, and if they are playing regularly you only need enough training to avoid an attribute loss. So you need to be careful about using heavier training sessions in your main squad that will benefit only a minority of players while increasing injury risk for all players.
An unanswered question is whether the attribute increases that are directly caused by playing time occur in areas that are influenced by what the player is training? So for example if I have a 23 year old that hasn’t maxed his CA I know that training quickness does not itself increase his CA, but does it direct the CA gain from playing into pace? I would tend to say yes, but I haven’t seen any objective evidence.
Yarema said: What are you talking about? I know how consistency improves with age.
I'm questioning that X consistency means a player will play to his full ability in X games out of 25, both the linear mechanism and the 25 number. And also that the ability drop is randomly determined by 1-20 CA drop, why not percentage based, why not influenced by consistency attribute and so on.
Would very much like to see some data on that if anyone has any. Expand
Sorry, I completely misread what you said. The tests I’ve seen show that the effects of consistency aren’t very large anyway if players have high morale.
Yarema said: Yes I get it that someone said it somewhere. There is zero backing for the numbers though. Or if there is I would very much like to see it. Expand
Just take a save you’ve played a decent number of years and calculate the average consistency for each age in the division you’ve been playing (including reserves and youth). You will see the increase. Or google it ; there are definitely some videos from either FM22 orFM23.
The only reason to train one tactic is when you first join a team and you have to get the team up to speed quickly. Otherwise I always train 3 tactics, the main one, one that concedes less goals, and one that’s attacking but a bit of a different look. Of course you want to keep positions/roles that still fit your team and that you can switch to mid-game without using a lot of subs.
Pfff... it would be ridiculous to assume that we've been updating our DB to make it better for finding good tactics to play FM! The less our DB represents normal gameplay, the better for us! Isn't it obvious?
I can only take a wild guess here... it could be that we want to keep the result of our hard work on learning the FM privately for ourselves? Yes, I know it's unbelievable selfishness but unfortunately, we're that evil. Once more, it's just my wild guess.
His question may have been worded rather rudely, but do you think it’s helpful to reply like that? Why not just give a straightforward answer and if you need to be vague to protect your work then say that.
Not sure what that even means, but changing the number of tests based on the result just introduces bias. There’s now no way to know if the result is more or less accurate than before.
I do not give my players manual holidays. But I play in rotation. I play with my main squad in Europe, with my reserve squad in the matches I play at home before or after the European matches in the league, with my main squad in the remaining matches and finally with my reserve squad in the local cups.
The key to this program is to give players full rest (for 3 sessions) the day after match days. If there is no match on Wednesday-Saturday, it is just a corner routine or shot stopping training for goalkeepers. Additionally, if there is no match on Wednesday, I added team bonding training to the evening session to boost team harmony and motivation.
I played 6 seasons with 24.3.0 using this method.
I play my matches on Wednesday-Saturday.
Below is a preview of the training program and download links.
Cheers.
Are you actually doing what ZaZ said, because you also say you don’t manually rest players? If you just have 3 rests in training and the players are “in training” then they recover substantially less than 3 rests plus manual rest for the day. It’s still better than scattering the 3 rests on different days but not a “super rest”. The strangest thing in the testing done by EBFM and ZaZ is that players on manual rest, no pitch or gym work, or even holiday, are affected by the training that the rest of the team is doing. This has to be a bug, but doing even one training session on the day makes all the players, including resting and on holiday, recover less. I suspect this bug is also part of what makes recovery sessions useless.
P.S. I predict, some people will rush to remove "Overlaps" TIs from the top tactics in hope to see their scores improves but I can assure your that highly likely it won't happen because almost all tactics at the top already hit a quite high positive RNG which is about +2 points so imagine you remove the Overlaps TI from the top tactics which would give +1 points but you struck -2 points negative RNG after 4,000 matches and as result you end up with a worse score.
Your general point is correct that making this change is unlikely to lead to a higher score in the test league, but the logic is wrong since what you refer to is actually ranking bias which is not the same as the random error in estimating the score. The size of the ranking bias is not possible to calculate given the nature of the submissions to the ranking table, but it would only be +2 by an unlikely coincidence.
I think more relevant is that for use in the game, removing overlaps would be expected to be favorable (at least for the role combinations used). Anyway thanks for the useful experiment.
I've been testing a lot lately and have regained the joy of FM (I haven't bought an FM every year in the last few years because it's become too easy for me). But the results of tests should always be presented objectively and the Reddit user didn't do that. All attributes matter and it is interesting to find out how important each attribute is for each position. I would beat e.g. this reddit users team with my 19 acceleration, 19 pace, 19 anticipation, 19 jumping reach, 20 finishing, 20 technique, 20 first touch, 20 heading striker. This problem is more complex than reddit users posts. Yes, we found meta-attributes but this isn't the end of the research.
Tbf one experiment involved only the 9 attributes, the other did involve 4 others but they didn’t matter in the sense that a mediocre value was sufficient. I think you are interpreting the word “matter” different to me. This doesn’t mean they do absolutely nothing, only that you really don’t have to pay attention to those attributes. If I sign a PL quality player then he probably will have at least 10 passing but I am never looking for the guy with 20 passing because there is always someone else who has better physicals and that’s what matters in the match engine.
Anyway, I think anyone visiting this site is hardly surprised by the results.
Thanks for pointing that out, I was writing from memory. The other one set to 13 was work rate. I’ll fix my post.
- despite several claims to the contrary the experiments were done correctly and verified by multiple people, so the 9 attributes are the important ones (as this site has shown repeatedly)
- you need certain other attributes to be able to use the 9 effectively. Fitness and stamina are obvious ; work rate and determination less so but you don’t want a really low value.
- Dribbling is interesting; if you dropped this down to 1 and set one of the other technicals to 20 performance crashed. So your super fast players do need to be able to run with the ball.
- Of course the profile of player used in experiment 2 doesn’t really exist in the game since the initial attributes are not completely random, and can only grow 5-6 points at most (and with limited user input). So you are not going to find some 80 CA player that can rip up the PL while on £50 a week. In some searching on a large database I did find a couple of 120-130 CA players that were so high on the 9 attributes that they could probably be world class. But the main conclusion is that when selecting players you can just focus on 9 attributes being as high as possible and it’s ok if some of the other attributes look really bad with the only red flags being low stamina, fitness, determination or work rate.
Edit: made some corrections based on Yarema’s comment.
https://www.reddit.com/r/footballmanagergames/comments/1ameaot/exposing_the_fm_match_engine_only_9_attributes/
Spoiler Why do scientists change only one variable in a controlled experiment?
In a controlled experiment, scientists change only one variable at a time in order to accurately determine the effect of that specific variable on the outcome. By isolating and manipulating one variable while keeping all others constant, scientists can establish a cause-and-effect relationship between the variable they are testing and the results they observe. This helps in drawing accurate conclusions and making reliable predictions based on the experiment's findings. If multiple variables were changed at the same time, it would be difficult to attribute any observed effects to a specific variable, leading to unreliable or inconclusive results.
https://www.quora.com/Why-do-scientists-change-only-one-variable-in-a-controlled-experiment
Neither you or the quote are correct in the context we are discussing, which is optimizing performance of a team with many positions and attributes. If you think of performance as a multi-dimensional surface, changing one variable only tells you that performance is sloping upwards in that region of the surface. But this can be towards only a local maximum, not the global maximum. Think of climbing a mountain; if the steepest path from where I am standing is North then will going North always lead me to the summit? No, if I looked around me I might see the summit is to the east and that what is North is just in a smaller peak below the main one. Or suppose I test the strength of a bond by varying the amount of glue; what will happen is that strength will increase up to a certain point, but beyond that more glue will actually weaken the bond. This is because I failed to consider another variable, the size of the surface to be bonded; this and the amount of glue interact and the optimum is actually a particular thickness of the the glue applied. I could go on about chemotherapy’s that are effective when used on their own but are useless for practical purposes because they add nothing but side effects when used with other more effective treatments. But you hopefully get the point. There are simplistic problems where it’s possible to isolate a single variable, but FM is really not such a case.
In the FM-Arena test Stamina attribute was increased by 5 points for every positions including the DMs and WBs and that improved the result just by 2 points from 60 pts to 62 pts, which honestly can be just the RNG but when Acceleration or Pace attributes were increased by 5 then it improved the result by 21 points from 60 pts to 81 pts.
So no way Stamina attribute or any other attribute can be as impactful as Pace and Acceleration for any position.
Pace - https://fm-arena.com/tactic/7528-5-points-to-pace-for-all-positions/
Acceleration - https://fm-arena.com/tactic/7527-5-points-to-acceleration-for-all-positions/
Stamina - https://fm-arena.com/tactic/7557-5-points-to-stamina-for-all-positions/
Here's another test when Acceleration and Pace was tested, specifically, for Central Midfielder position - https://fm-arena.com/thread/5351-should-you-follow-the-highlighted-attributes-of-the-roles/
I don’t think your logic is correct. When you test attributes one by one and for all positions it can only tell you if an attribute is important for the majority of positions. The only way to know at a single position level is vary multiple attributes in the same experiment and vary by position as well (this is in effect what the machine learning data is reporting). The best you can say about stamina is that it is not important for the majority of positions, and in a very controlled setting where players are always starting games at peak performance and have equally capable replacements available on the bench.
An unanswered question is whether the attribute increases that are directly caused by playing time occur in areas that are influenced by what the player is training? So for example if I have a 23 year old that hasn’t maxed his CA I know that training quickness does not itself increase his CA, but does it direct the CA gain from playing into pace? I would tend to say yes, but I haven’t seen any objective evidence.
I'm questioning that X consistency means a player will play to his full ability in X games out of 25, both the linear mechanism and the 25 number. And also that the ability drop is randomly determined by 1-20 CA drop, why not percentage based, why not influenced by consistency attribute and so on.
Would very much like to see some data on that if anyone has any.
Sorry, I completely misread what you said. The tests I’ve seen show that the effects of consistency aren’t very large anyway if players have high morale.
Just take a save you’ve played a decent number of years and calculate the average consistency for each age in the division you’ve been playing (including reserves and youth). You will see the increase. Or google it ; there are definitely some videos from either FM22 orFM23.