Mark said: Testing wasn't a huge sample. I think if you believe in the machine learning work then you go with their total results. I had a lot of faith in my original ratings and these seem to be slightly better. So I continue to use the ykykyky one successfully. Expand
Thanks for the reply, I was just curious from reading back through the thread
Mark said: You have 4 strikers in your side and Huanca would be your 5th. I have run the ratings against all 5 strikers using each of the ratings files. I would be tempted to buy him. He doesn't rate well in my MDW ratings file but goes well in the Y50 and even the ykykyk files.
I think from the analysis I have, done that the machine learning works. I personally will be using the full ykykyk ratings file as I think it will be more stable longer term. I will now balance that ratings file and derive the unrated positions as best I can.
@ZaZ prefers the chopped down versions because you get better value players. I prefer to try and get players you can keep as you progress. And I certainly move between the approaches and value what ZaZ does immensely.
My focus beyond changing my ratings file is to upgrade my approach to individual training focus. There were some notes in the machine learning example relating to that. I am exploring how I improve on that aspect in my games. At some stage I will provide my learnings on that front. Expand
@Mark Why did you come to the choice of using the ykykyk file? Didn't the ZaZ under minus 25 score higher when you did your test? I understand that ZaZs gives more value but why do you think ykykyk is more stable longer term?
@ZaZ How do you test your tactics? Just reading the parapgraph in the opening post. You use an edited database. Do you set your tactic for all 8 teams you listed and then holiday the 1st season 3 times?
I've read it all, many times, here and on other sites, I'm passionate about the work done by the members here, but I have a simple request. My way of looking for players on Genie Scout is as follows: I use the 50 minus 25 zaz balanced by Mark, as I much prefer a filter that maximizes important attributes rather than a balance that can lead to misinformation. Now I encounter a problem with this magnificent filter: I would like to have the raw rating (general rating) which does not modify the rating according to the affinity with the position. Because I want to be able to manage this problem on my side and decide to teach the player the new position. But I would like the general note to come out directly, I find it easier to find players who can stick to several positions. I hope my request is clear, and if this filter already exists, I apologize, I'm not English and I may have missed a subtlety.
thank you again for the exceptional and exciting work done here. Expand
Why do you think the balanced filter leads to misinformation?
Mark said: @Middleweight165 and @Enok sorry for the delay I have been away. These are calculated using the attribute costs as well as most important. Expand
No need to apologise Mark Is there an attribute cost list anywhere?
Mark said: I have also redone my assessment of attributes that can be be improved for each position through training for this year. And the winners are (drum roll please):
These are in order. You should check each of your players to see where the best improvements can be made and then set their individual training focus. These attributes are not always easily identifiable. In individual training focus they are sometimes hidden eg Decisions for a DC is Defensive Positioning. Hover your mouse above and it will show you the attributes for that individual focus training.
This is based on the cost and importance of each attribute. I am not convinced about the Acc for GKs but time will tell. Expand
Is this still applicable @Mark? I know it was published before the chinese results were published
Mark said: If you want to compare your team with other teams in your league you can open the dialog box for your team and go to Tactics (as per the method in my last post), you can see your teams rating. by pressing the left or right arrow on your keyboard you will move through the teams that are on the list page and be able to see their ratings for comparison.
I did a thread on training a while back and tested a few well renowned training routines.
@Mark Do you know how Team Rating is calculated in GS? It doesn't seem to change if I load a different ratings file
Mark said: I have always had a problem with the Club ratings. I calculate my own. If you select the club and bring up the dialog box and go to Tactics it will show you the team rating. You can select different formations or select the best 5 formations for your side. These ratings are based on the ratings file. I do disagree with their calculations where there is such a minor adjustment for not being Natural for a position. This is why I calculate my own. Expand
Thanks, I only wanted to compare my team with others but its just a curiosity. If it requires work, its not that important
Whats your thoughts on training? Ive always subscribed to the quickness and reflexes approach, but i think there might be a better way now we have this info
Mark said: There is a way to do this. First a bit of background. This used to work but a few years back FM changed their file format from .slf files to .fmf files. GS has never caught up to that, I suspect there is a fair bit of programming work involved.
Firstly check you know where the shortlists are stored, default directory in Windows is C:\Users\<<Username>>\Documents\Sports Interactive\Football Manager 2022\shortlists.
If you only want to bring in a handful of players or less there is some very easy functionality. After you have found a player you are interested in double click on the line the player is on in GS. A pop up box will be brought up for that player with more detail.
If you hover your mouse over the name in the pop up box it will say "Click to copy name to clipboard". Click the name. Now go back to FM. In middle top of most skins it will tell you what page you are on - Home, Inbox, Squad etc. If you click there you can paste the player name. FM will search for the player. Select the player if it brings back more than one, make sure the players age, team and position are the same. You can now make an offer for the player or save them in shortlists if you want.
If you have a heap of players you want to bring across to FM, firstly create the shortlist you want to make, and ensure it is in the correct directory.
Second, download FMRTE (the free version works you don't need to pay for it) and load your game.
1. At the top of FMRTE it says "shortlists" click on it 2. Click on load shortlist 3. Load the shortlist from FM genie scout into FMRTE 4. Highlight a player in the shortlist and press ctrl+A (to select every player in your shortlist) 5. In the bottom right there is an option to click which allows you to save the shortlist 6. Save as a new shortlist.
Now you can import the new shortlist you created into FM.
Summary - make the shortlist in FMGS, import shortlist into FMRTE, save shortlist in FMRTE then import into FM via FM itself. Expand
Life Saver! Thanks mate. I thought I was going to have to manually type each player into a shortlist
The standard place for Genie Scout ratings files is C:\FM Genie Scout 22g\Ratings. If you did a custom install it will be in a folder called Ratings under the FMGS install directory. Once you have copied the file to the correct location you should be able to select my ratings file by clicking on the GS symbol in the top left of the screen, and then clicking Ratings in the drop down menu, and then ykykyky balanced from the sub menu.
Below is the link to the place in this thread where I shared the FM22 default ratings file - post #33.
The problem with trying to produce a ratings file based on the attributes testing from FM Arena is that the for each player will be the same for each position. The only thing that will then impact is the adjustment for their positional rating out of 20 ie you need to multiply their GS Rating by 100% minus the difference between 20 and their positional rating multiplied by 2.5%. For example if their GS rating is 50% and their positional rating is 19, it is 20 - 19= 1 multiplied by 2.5% is 2.5%. Take this off 100% = 97.5% multiplied by 50.00% is 48.75%.
I will give it a go though for a comparison against ykykyky and the default. Expand
Sorted now thanks!
I've run into another problem though. I can't export shortlists from GS. DO you have this issue? I had no problem export a .slf file in FM21 but now they dont appear.
Mark said: @Middleweight165 I have a few comments around your observations above. The FM Arena attribute tests were against all positions in the side so not by position. The machine learning tests were against each of the positions in the ZaZ Blue tactic, GK, DR, DC, DL, DM, MR, MC, ML and ST.
I have tested using the Machine learning findings in positional filters o Genie Scout and find them to be very good. The best players from my team at the end of the season are the ones that were found and are at the top of the list when I use the ratings file.
I also know that if you just use acceleration to pick strikers and wingers you will generally get decent players as I have tried this before. Your observation was that Work Rate was higher for the DM. This is correct and it is worth noting that this is the only position where Work Rate is higher than Acceleration for any of the Machine learning positional analysis. My DMs don't tend to have as high Acceleration numbers as my attacking players.
If you look at the Machine learning numbers for DM, there are values against 36 attributes, so the 90 rating for Work Rate is around 6.2% of the overall rating for that position. You can see from that, if you just pick a DM on the highest rated attribute for the position it might not work too well and it certainly wouldn't be in keeping with what the Machine learning analysis was trying to get across to us. Also, if you chose players using Acceleration and Pace (which tend to be reasonably close numbers for most players), combined they total around 9.3% of the overall rating for the DM position.
I would suggest you use the Genie Scout ratings file it makes it easy and you can compare how all players rate for particular positions you are interested in. I also posted the ratings file on FM Scout. Here is an extract of the response I received:
"I must say that this filter gives way better ratings than the default ratings which comes with Genie Scout. The reason is, I see way more players which perform in-game better than the rest with your filters.
I really thank you for your effort and sharing this! Now I can buy players without having to pray for them to perform in-game" Expand
Thanks for such a detailed reply. I have uploaded your balanced ratings file into Genie Scout. Can I ask does your file override the existing file or do i need to load it in when Im in GS? I copied the file into the ratings folder in the C drive. When clicking between your ratings file and the old one 'Football Manager 2016', the ratings dont seem to change, so I'm wondering if it has automatically updated to yours
ZaZ said: Testing the impact of attributes is something very hard to do, because important attributes for DC are different from important attributes for ST. Basically, you need to test each position/role individually. It's also not something linear, which means the effect from 10 > 12 can be very big, but from 12 > 14 can be negligible. That means there are too many combinations to test, and FM-Arena just doesn't have the resources to do so. What they did was an approximation that is pretty accurate for most cases. We should be thankful they did that when no one else wanted to do it, and appreciate their contribution to FM community.
About the study from the Chinese group, they had more resources since they are a company that works in a basketball game with focus on statistics. That includes machines (possibly a supercomputer), experts in the field of statistics / machine learning, and time. I believe their results are more precise, but you have to consider the difference in their methodology.
For example, the AI made variations in attributes but kept the same CA for players, while in FM-Arena the attributes were changed without considering CA. The result is that attribute weights have direct impact in the Chinese study, while it was not considered by FM-Arena since it wanted to isolate the impact of the attribute.
Again, you have to understand that even if FM-Arena wanted to do a similar methodology, that wouldn't be possible because they just don't have resources to do so. I think FM-Arena's approximation is pretty decent considering how hard it is to test attributes. The Chinese guy itself said they wouldn't be doing any more similar tests using machine learning since the cost was too high to repeat the experiment.
So, in short, FM-Arena's results are different mainly because they are an approximation, and also because they chose to not consider the weight of attributes to CA, which has the potential to counter-balance the impact and make the analysis even harder. Expand
Thanks for taking the time to write such a detailed reply. I understand completely what you're that the FM arena testing is limited because that's all they are capable of doing, but it still provides us with a huge insight into whats important and whats not.
What instigated my question was my assumption that the chinese research was more thorough, therefore a more accurate analysis of which attributes are most important for each position, but it felt to me, and this could just be a personal feeling is that members of this forum have't abandoned the FM Arena test results in favour of the Chinese results and I wanted to know why. Maybe paragraph 3 of your reply provides more info on this.
What is your personal opinion now? Do you follow the Chinese guidelines or the FM Arena ones? For the DM position would you prioritise Work Rate (Chinese) or Acceleration (FM Arena)?
ZaZ said: It can be several reasons. For example, it can be very important until certain value, then less important after some point. Expand
Can you give me a practical example? I don't understand. Looking at the chinese table, for out field players Work Rate is 5th most important for CB, 4th FB, 1st DM, 4th W, 4 AM, 7th ST, so on average should be around the 4th most important attribute. yet on the FM arena testing it is around 23rd
It's quite simple to me. For all positions except central defender you should look for Acceleration, Pace, Dribbling, Anticipation, Stamina and for central defenders you should look Jumping Reach and Strength instead of Dribbling. Expand
Doesn't the work of the chinese forum suggest different though
Thanks for the reply, I was just curious from reading back through the thread
I think from the analysis I have, done that the machine learning works. I personally will be using the full ykykyk ratings file as I think it will be more stable longer term. I will now balance that ratings file and derive the unrated positions as best I can.
@ZaZ prefers the chopped down versions because you get better value players. I prefer to try and get players you can keep as you progress. And I certainly move between the approaches and value what ZaZ does immensely.
My focus beyond changing my ratings file is to upgrade my approach to individual training focus. There were some notes in the machine learning example relating to that. I am exploring how I improve on that aspect in my games. At some stage I will provide my learnings on that front.
@Mark Why did you come to the choice of using the ykykyk file? Didn't the ZaZ under minus 25 score higher when you did your test? I understand that ZaZs gives more value but why do you think ykykyk is more stable longer term?
Where can I find this?
I've read it all, many times, here and on other sites, I'm passionate about the work done by the members here, but I have a simple request.
My way of looking for players on Genie Scout is as follows: I use the 50 minus 25 zaz balanced by Mark, as I much prefer a filter that maximizes important attributes rather than a balance that can lead to misinformation.
Now I encounter a problem with this magnificent filter: I would like to have the raw rating (general rating) which does not modify the rating according to the affinity with the position.
Because I want to be able to manage this problem on my side and decide to teach the player the new position. But I would like the general note to come out directly, I find it easier to find players who can stick to several positions.
I hope my request is clear, and if this filter already exists, I apologize, I'm not English and I may have missed a subtlety.
thank you again for the exceptional and exciting work done here.
Why do you think the balanced filter leads to misinformation?
No need to apologise Mark
Position Focus
GK GK Reactions
DC Quickness
DRL Endurance
DM Passing
MRL Quickness
AMRL Quickness
MC Passing
AMC Quickness
ST Ball Control
I think the quickness is obvious but can I ask why you came to the conclusion endurance for FBs, passing for DMs and ball control for STs?
Agreed! and yourself!
These are in order. You should check each of your players to see where the best improvements can be made and then set their individual training focus. These attributes are not always easily identifiable. In individual training focus they are sometimes hidden eg Decisions for a DC is Defensive Positioning. Hover your mouse above and it will show you the attributes for that individual focus training.
This is based on the cost and importance of each attribute. I am not convinced about the Acc for GKs but time will tell.
Is this still applicable @Mark? I know it was published before the chinese results were published
I did a thread on training a while back and tested a few well renowned training routines.
Training thread
Have a read, it might give you some ideas.
Thanks for the link, interesting read and thanks for all your efforts here. Really enhancing the experience
Mark said: I have always had a problem with the Club ratings. I calculate my own. If you select the club and bring up the dialog box and go to Tactics it will show you the team rating. You can select different formations or select the best 5 formations for your side. These ratings are based on the ratings file. I do disagree with their calculations where there is such a minor adjustment for not being Natural for a position. This is why I calculate my own.
Thanks, I only wanted to compare my team with others but its just a curiosity. If it requires work, its not that important
Whats your thoughts on training? Ive always subscribed to the quickness and reflexes approach, but i think there might be a better way now we have this info
Firstly check you know where the shortlists are stored, default directory in Windows is C:\Users\<<Username>>\Documents\Sports Interactive\Football Manager 2022\shortlists.
If you only want to bring in a handful of players or less there is some very easy functionality. After you have found a player you are interested in double click on the line the player is on in GS. A pop up box will be brought up for that player with more detail.
If you hover your mouse over the name in the pop up box it will say "Click to copy name to clipboard". Click the name. Now go back to FM. In middle top of most skins it will tell you what page you are on - Home, Inbox, Squad etc. If you click there you can paste the player name. FM will search for the player. Select the player if it brings back more than one, make sure the players age, team and position are the same. You can now make an offer for the player or save them in shortlists if you want.
If you have a heap of players you want to bring across to FM, firstly create the shortlist you want to make, and ensure it is in the correct directory.
Second, download FMRTE (the free version works you don't need to pay for it) and load your game.
1. At the top of FMRTE it says "shortlists" click on it
2. Click on load shortlist
3. Load the shortlist from FM genie scout into FMRTE
4. Highlight a player in the shortlist and press ctrl+A (to select every player in your shortlist)
5. In the bottom right there is an option to click which allows you to save the shortlist
6. Save as a new shortlist.
Now you can import the new shortlist you created into FM.
Summary - make the shortlist in FMGS, import shortlist into FMRTE, save shortlist in FMRTE then import into FM via FM itself.
Life Saver! Thanks mate. I thought I was going to have to manually type each player into a shortlist
Link to ykykyky balanced Ratings file
The standard place for Genie Scout ratings files is C:\FM Genie Scout 22g\Ratings. If you did a custom install it will be in a folder called Ratings under the FMGS install directory. Once you have copied the file to the correct location you should be able to select my ratings file by clicking on the GS symbol in the top left of the screen, and then clicking Ratings in the drop down menu, and then ykykyky balanced from the sub menu.
Below is the link to the place in this thread where I shared the FM22 default ratings file - post #33.
Link to default file
The problem with trying to produce a ratings file based on the attributes testing from FM Arena is that the for each player will be the same for each position. The only thing that will then impact is the adjustment for their positional rating out of 20 ie you need to multiply their GS Rating by 100% minus the difference between 20 and their positional rating multiplied by 2.5%. For example if their GS rating is 50% and their positional rating is 19, it is 20 - 19= 1 multiplied by 2.5% is 2.5%. Take this off 100% = 97.5% multiplied by 50.00% is 48.75%.
I will give it a go though for a comparison against ykykyky and the default.
Sorted now thanks!
I've run into another problem though. I can't export shortlists from GS. DO you have this issue? I had no problem export a .slf file in FM21 but now they dont appear.
I have tested using the Machine learning findings in positional filters o Genie Scout and find them to be very good. The best players from my team at the end of the season are the ones that were found and are at the top of the list when I use the ratings file.
I also know that if you just use acceleration to pick strikers and wingers you will generally get decent players as I have tried this before. Your observation was that Work Rate was higher for the DM. This is correct and it is worth noting that this is the only position where Work Rate is higher than Acceleration for any of the Machine learning positional analysis. My DMs don't tend to have as high Acceleration numbers as my attacking players.
If you look at the Machine learning numbers for DM, there are values against 36 attributes, so the 90 rating for Work Rate is around 6.2% of the overall rating for that position. You can see from that, if you just pick a DM on the highest rated attribute for the position it might not work too well and it certainly wouldn't be in keeping with what the Machine learning analysis was trying to get across to us. Also, if you chose players using Acceleration and Pace (which tend to be reasonably close numbers for most players), combined they total around 9.3% of the overall rating for the DM position.
I would suggest you use the Genie Scout ratings file it makes it easy and you can compare how all players rate for particular positions you are interested in. I also posted the ratings file on FM Scout. Here is an extract of the response I received:
"I must say that this filter gives way better ratings than the default ratings which comes with Genie Scout. The reason is, I see way more players which perform in-game better than the rest with your filters.
I really thank you for your effort and sharing this! Now I can buy players without having to pray for them to perform in-game"
Thanks for such a detailed reply. I have uploaded your balanced ratings file into Genie Scout. Can I ask does your file override the existing file or do i need to load it in when Im in GS? I copied the file into the ratings folder in the C drive. When clicking between your ratings file and the old one 'Football Manager 2016', the ratings dont seem to change, so I'm wondering if it has automatically updated to yours
About the study from the Chinese group, they had more resources since they are a company that works in a basketball game with focus on statistics. That includes machines (possibly a supercomputer), experts in the field of statistics / machine learning, and time. I believe their results are more precise, but you have to consider the difference in their methodology.
For example, the AI made variations in attributes but kept the same CA for players, while in FM-Arena the attributes were changed without considering CA. The result is that attribute weights have direct impact in the Chinese study, while it was not considered by FM-Arena since it wanted to isolate the impact of the attribute.
Again, you have to understand that even if FM-Arena wanted to do a similar methodology, that wouldn't be possible because they just don't have resources to do so. I think FM-Arena's approximation is pretty decent considering how hard it is to test attributes. The Chinese guy itself said they wouldn't be doing any more similar tests using machine learning since the cost was too high to repeat the experiment.
So, in short, FM-Arena's results are different mainly because they are an approximation, and also because they chose to not consider the weight of attributes to CA, which has the potential to counter-balance the impact and make the analysis even harder.
Thanks for taking the time to write such a detailed reply. I understand completely what you're that the FM arena testing is limited because that's all they are capable of doing, but it still provides us with a huge insight into whats important and whats not.
What instigated my question was my assumption that the chinese research was more thorough, therefore a more accurate analysis of which attributes are most important for each position, but it felt to me, and this could just be a personal feeling is that members of this forum have't abandoned the FM Arena test results in favour of the Chinese results and I wanted to know why. Maybe paragraph 3 of your reply provides more info on this.
What is your personal opinion now? Do you follow the Chinese guidelines or the FM Arena ones? For the DM position would you prioritise Work Rate (Chinese) or Acceleration (FM Arena)?
Can you give me a practical example? I don't understand. Looking at the chinese table, for out field players Work Rate is 5th most important for CB, 4th FB, 1st DM, 4th W, 4 AM, 7th ST, so on average should be around the 4th most important attribute. yet on the FM arena testing it is around 23rd
It's quite simple to me. For all positions except central defender you should look for Acceleration, Pace, Dribbling, Anticipation, Stamina and for central defenders you should look Jumping Reach and Strength instead of Dribbling.
Doesn't the work of the chinese forum suggest different though