@olimorris On the ykykyky balanced, I think someone playing as D/WB (L) should not have FB rating of zero. When I see the json file, maybe adding "D/" can fix that.
HVS said: @olimorris On the ykykyky balanced, I think someone playing as D/WB (L) should not have FB rating of zero. When I see the json file, maybe adding "D/" can fix that. Expand
Mark said: I did. Still testing the different ratings files for FM24 Expand
I'm always one FM behind, so have just started on FM23. I read in this thread you flittered between yk balanced ratings and the MDW22. Do you prefer one?
Also I had a quick look at some players using the MDW22. The highest rated player interested in joining me had 9/10 Pace and Acceleration. Would you still sign this player? It didnt sit right with me that I'd sign players with such low speed, as we are always pushing that physicals are the most important, but this might be subjective bias. What are your thoughts?
Middleweight165 said: I'm always one FM behind, so have just started on FM23. I read in this thread you flittered between yk balanced ratings and the MDW22. Do you prefer one?
Also I had a quick look at some players using the MDW22. The highest rated player interested in joining me had 9/10 Pace and Acceleration. Would you still sign this player? It didnt sit right with me that I'd sign players with such low speed, as we are always pushing that physicals are the most important, but this might be subjective bias. What are your thoughts? Expand
FM23 I used MDW and trusted the ratings. Seemed to work
I'm toying with a new rating system, and would love some feedback. It ignores GK - but bear with me on that. I think a different approach is warranted for that special position.
First question is - how much do we believe the underlying simulation algorithms change from year to year? Specifically,
Do we suppose that these would be meaningful in 24?
If so, then I like these more than weights, because it shows how the simulator responds to inputs - it's more "evidence-based."
So I did a little analysis last night, and compared the Attribute Testing results to ykykyk, in an effort to create a new roster prioritization / scoring approach that relies as much as possible on the Attribute Testing results.
Methodology -
1) Looked at the measured attributes and segmented them into whether there was a defensive impact, an offensive impact, or both.
Looking specfically at Goals Allowed (Defense) vs. Goals Scored (Attacking).
2) Next, created a basic Attack template for attacking heavy roles, and a Defense template for Defensive heavy roles.
The different shades of green are used to help me identify if the impact of the attribute was pretty strong, medium, or weak.
I then reckoned that I would limit the number of attributes I measured based on roughly 50% of total attributes available (so 18 attributes to be considered). That meant that for attackers - there were only 14 relatively meaningful attributes, and for defenders - 16.
4) I decided to "fill in the missing attributes" by comparing the Attribute Testing with ykykyk weights for each position. So, Finishing isn't represented in the Attribute testing, but sure seems important to attackers in terms of weights, so I added that.
5) This all resulted in the following:
So you see the attributes, the numbers in the first row represent the ykykyk weights, the second row numbers are the impact on Goals Allowed difference or Goals Scored difference.
The red numbers represent extrapolation (complete hogwash, just interested in looking at unmeasured attributes i added, and if the attribute wasn't measured, what is the possible impact largest impact based on attributes with similar ykykyk weights). I pretty much know the red numbers are hogwash because Off The Ball is weighted highly in some positions, but with no demonstrable impact in the attribute testing results.
From here, I would simply replace my current approach (weights-only used in ranking, with this new idea.
As I've already said on a different post, I am sceptical of ykykyky's ratings for each role. However I concur that we should try to put more effort into finding the most relevant attributes for each role, as you've done here.
I'm confident the team at SI rarely changes existing simulation algorithms and just tries to add more 'situations' every year. However this opinion is based on nothing, other than the relevant attributes from the attribute testing table don't seem to shift that much every year.
I do have to say I am a bit confused by your last graph with the final results, in particular with the Goals Allowed difference/Goals Scored difference. What testing did you do to get these differences?
Is it possible to reduce the attacking and defensive template into weights, for each attribute with these findings?
I appreciate the effort put into this, so I feel bad about giving negative feedback. The problem of trying to work from the attribute testing is that these are single variable tests; any positional differences are averaged out. Also when you try to combine the attribute tests you are not accounting for the correlations between the different attributes that exist in the player database; a player with 20 pace is going to be very likely to have high acceleration and very unlikely to have extremely low acceleration. So if you treat the effects of pace and acceleration as additive you will be overestimating their combined effect. The value of the machine learning approach is that it tests all variables simultaneously and accounting for position.
@skawkclsrn - GA/GS difference testing comes from the attribute testing on this site. It’s specious for some of the reasons @MeanOnSunday mentions. But I think it’s somewhat meaningful in that we’re looking not just at the inputs into the simulator (as ykykyk or Mark’s or SIs or whomever) but shows us how the simulator responds. It allows us to ask ourselves questions like, “if ‘Off the Ball’ isn’t meaningful in a controlled environment - how could it be meaningful in my environment - (a 4-2-4 in my case)?”
First, let me be the first to say I’m not good at analytics, soccer (I’m just a Yank who became interested in it at the tail end of COVID), nor FootballManager (I’m only on my 5th career here).
So negative feedback is EXACTLY what I’m looking for. I hadn’t thought about the correlation and additive nature of calculating.
RE: positional differences being averaged out - wouldn’t that also be a potential benefit? Going into this analysis, I thought about:
1) the fact that it’s not my tactic, and 2) it’s coarse on positional differences, and 3) it’s older (FM22), and 4) it doesn’t test all attributes, and 5) some of the differences may not be statistically significant (+2 goals over 913 simulations? Meh?).
So the compromise I’m offering is not weighting every attribute, but simply trying to combine those that the simulation seems to respond to, and then adding enough attributes that are NOT measured in the attribute testing with ykykyk weights.
Finally, I’m starting to think that any of these approaches are “good enough to be good enough.” Using @Mark weights - I’m dominating MLS on the Touch version. Like no scum save just went a season without losing. Would it even matter if I switched to ykykyk? Is my success because I “cheated” came onto this site, and emulated 4-2-4 Bombyte Tweak more than finding really great players?
I mean, I had success in Ligue 1 with Lorient (!) just measuring how many attributes weighted above 50 were above 16 for each player. Not nearly as granular as any of the above - but still successful.
As long as I know that P/A rule, and the stars are a lie, and I’m playing a proven tactic - is the whole XLS exercise just there for me to not ever overpay for Xavi Simmons again?
I don’t think you can go wrong with either the machine learning weights, Mark’s, or the more recent weights from squirrel_says. The AI in the game is really not very smart and seems to value players based on CA and reputation. So if you play a half decent tactic with players selected based on what matters in the ME, and managing fitness and morale properly, you can win everything with an average CA substantially less than the top AI teams. My point was more a theoretical one that a “better” rating system should consider how the attributes are correlated since its easy to over- or under-weight attributes based on tests done one variable at a time.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook. Expand
No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features?
Gracolas said: No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features? Expand
I find the GS workflow to be really slow. Loading the DB takes ages. If I wish to change the ratings I have to go into so many different windows + tabs and I have to remember to apply them each time I load the tool up .
I can't speak for other projects but I plan on including a MoneyBall element soon and combining that with the machine learning ratings.
olimorris said: I find the GS workflow to be really slow. Loading the DB takes ages. If I wish to change the ratings I have to go into so many different windows + tabs and I have to remember to apply them each time I load the tool up .
I can't speak for other projects but I plan on including a MoneyBall element soon and combining that with the machine learning ratings. Expand
Ok, makes sense.. I dont mess up with the ykykykyk ratings, so i just leave them unchanged. In this case i find GS really smooth and quick.
Gracolas said: Ok, makes sense.. I dont mess up with the ykykykyk ratings, so i just leave them unchanged. In this case i find GS really smooth and quick. Expand
It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook. Expand
Hi, thank you for your work with pyscoutfm. I tried to use your tool by following your guide on GitHub, but I run across a ExPathEvalError on the report generation step. How do I fix this error?
I've just added a section around how you can install this in the cloud with minimal effort - here. I recommend doing that and then following the guide on how to Generate a Scout Report.
Appreciate not many people have experience working with Python and it can messy to install if you're on Windows.
I've added an FAQ section comparing PyScoutFM to Squirrel's, here. Hard not to make it look like a d*ck measuring contest but I think it's the best way to show how the tools differ. I'll stand by the outputs from PyScoutFM compared to Squirrel's as it uses the ykykykyky_balanced ratings that @Mark put together in the Genie Scout filter file all those months ago.
Watching how Squirrel explains his rating system (I've also linked to this in the FAQ section), they're mostly based on how FM colours various attributes for a given role. With the exception of Acceleration, Pace and Finishing for an F9 which Squirrel marks as "Key" and multiplies by 5. He also grades other attributes as "Green" and "Blue", which he multiplies by 3 and 1, respectively. Unless I've missed something in his previous videos, I don't understand the delta of 2 between the Key and Green and the Green and Blue. In other words, why is it 2 and not 3.7 or some other made up number?!
Anyway, I've created this tool to scratch my own itch as I prepare to do an English Level 10 LLM adventure in FM24. I wanted to share it with this thread as thanks for all of the amazing effort and content that so many people have put in. I'm likely to expand it in the coming weeks with some Moneyball stats and ratings.
If you're hitting any weird issues or can't get it loaded, open up a discussion in the repo and I'll be happy to help.
olimorris said: I've just added a section around how you can install this in the cloud with minimal effort - here. I recommend doing that and then following the guide on how to Generate a Scout Report. Expand
@Floppyaams this is probably going to be the easier way of using PyScoutFM. Reach out to me if you need any help.
MeanOnSunday said: It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script. Expand
I mean, opening the correct view in FM, copying and pasting, opening the right file and all the other minor things you have to do also take like a minute or two in total. Plus you actually need to be doing stuff in that minute instead of 1 click to load database and do other things while it finishes.
Gracolas said: No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features? Expand
I want to add my view here, I have been using python scripts for more than 8 months to play football manager and I have my code base than what @olimorris or Squirrel's has created. So the reason why I like it is
1. Mac: I play in Mac M1, so Genie scout is not an option and FMRTE requires me to by-pass security in place. So the only way I can do some analytics is to use excel file or Python and Python is much more comfortable.
Since I haven't used Genie Scout or FMRTE, take the below comments with a grain of salt and I might be wrong
2. Own Squad Evaluation: I use a script that quickly gives me the Min, Max, Mean and Median score for each positions in my squad with a click of button. This is what I use in order to plan my future recruits. Not sure if this is possible in other tools.
3. Moneyball: I know some might think it's cheating / making the game easy playing the game with attributes unmasked (Hey, its a single player game do what's fun for you right :P). For those, the only way you can do it is using python script. FM Stag released his latest stats for different positions and what those means. I have used in a python script to come up with a score for each player I can sort based on it. @olimorris : I know you said you have plans to go here, so maybe this can help you with it right (again this is just one way to do it). See the below query will provide a score of 0 if the Hdrs W/90 <=2.76 and 1 if its between 2.76 and 5.08 and if its more than 5.08 a score of 2. I do the same for all the different stats and then add em up to come up with a single score for my striker which then I will use for recruitment.
MeanOnSunday said: It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script. Expand
Personally, i dont think that makes much difference.. i regularly load GS after a couple of in-game days, so ratings dont change that much.
I appreciate the work in Python. For those asking "why don't" - well, some of us do a bit more. I have my own ratings (customized to my passing-heavy style of play), my own metrics (ratings against "average" or "best in the world" ) and my own views (what is my best squad for national competitions?).
Diversity of thought and effort is pretty good. But I'm sure if you just use Squirrel Plays or whatever, you'll be fine too
Yarema said: I mean, opening the correct view in FM, copying and pasting, opening the right file and all the other minor things you have to do also take like a minute or two in total. Plus you actually need to be doing stuff in that minute instead of 1 click to load database and do other things while it finishes. Expand
I was only commenting on GS, not making a comparison. The point is that something that can be coded to run almost instantly (and is in other tools) takes minutes in GS. The rating part of it is (to me) easier in GS than changing a python script, and can get the same result.
AlexH said: I appreciate the work in Python. For those asking "why don't" - well, some of us do a bit more. I have my own ratings (customized to my passing-heavy style of play), my own metrics (ratings against "average" or "best in the world" ) and my own views (what is my best squad for national competitions?).
Diversity of thought and effort is pretty good. But I'm sure if you just use Squirrel Plays or whatever, you'll be fine too Expand
Can I just say, totally off topic, that this post is in the top half a dozen viewed posts of all time on FM Arena with over 130k views, and the highest of any non tactic post on the site. So clearly a subject that is important to the FM Community.
Out of interest, the top 2 by a massive margin, are both from the great @ZaZ with ZaZ Blue 4.0 in FM22 tactics with over 630k views, and ZaZ Blue 3.0 in FM21 tactics with just under 580K views. In third is Phoenix v5.0 in FM21 tactics with over 230k views.
Can I just say, totally off topic, that this post is in the top half a dozen viewed posts of all time on FM Arena with over 130k views, and the highest of any non tactic post on the site. So clearly a subject that is important to the FM Community.
Out of interest, the top 2 by a massive margin, are both from the great @ZaZ with ZaZ Blue 4.0 in FM22 tactics with over 630k views, and ZaZ Blue 3.0 in FM21 tactics with just under 580K views. In third is Phoenix v5.0 in FM21 tactics with over 230k views. Expand
That is actually nice to know! That means people need you for positional filters in FM24.
@olimorris On the ykykyky balanced, I think someone playing as D/WB (L) should not have FB rating of zero. When I see the json file, maybe adding "D/" can fix that.
HVS said: @olimorris On the ykykyky balanced, I think someone playing as D/WB (L) should not have FB rating of zero. When I see the json file, maybe adding "D/" can fix that.
Great spot and completely agree. Will fix later!
@Mark Did you use the MDW22 GS rating for FM23?
Middleweight165 said: @Mark Did you use the MDW22 GS rating for FM23?
I did. Still testing the different ratings files for FM24
Mark said: I did. Still testing the different ratings files for FM24
I'm always one FM behind, so have just started on FM23. I read in this thread you flittered between yk balanced ratings and the MDW22. Do you prefer one?
Also I had a quick look at some players using the MDW22. The highest rated player interested in joining me had 9/10 Pace and Acceleration. Would you still sign this player? It didnt sit right with me that I'd sign players with such low speed, as we are always pushing that physicals are the most important, but this might be subjective bias. What are your thoughts?
Middleweight165 said: I'm always one FM behind, so have just started on FM23. I read in this thread you flittered between yk balanced ratings and the MDW22. Do you prefer one?
Also I had a quick look at some players using the MDW22. The highest rated player interested in joining me had 9/10 Pace and Acceleration. Would you still sign this player? It didnt sit right with me that I'd sign players with such low speed, as we are always pushing that physicals are the most important, but this might be subjective bias. What are your thoughts?
FM23 I used MDW and trusted the ratings. Seemed to work
I'm toying with a new rating system, and would love some feedback. It ignores GK - but bear with me on that. I think a different approach is warranted for that special position.
First question is - how much do we believe the underlying simulation algorithms change from year to year? Specifically,
https://fm-arena.com/table/13-attribute-testing/#options
Do we suppose that these would be meaningful in 24?
If so, then I like these more than weights, because it shows how the simulator responds to inputs - it's more "evidence-based."
So I did a little analysis last night, and compared the Attribute Testing results to ykykyk, in an effort to create a new roster prioritization / scoring approach that relies as much as possible on the Attribute Testing results.
Methodology -
1) Looked at the measured attributes and segmented them into whether there was a defensive impact, an offensive impact, or both.
Looking specfically at Goals Allowed (Defense) vs. Goals Scored (Attacking).
2) Next, created a basic Attack template for attacking heavy roles, and a Defense template for Defensive heavy roles.
The different shades of green are used to help me identify if the impact of the attribute was pretty strong, medium, or weak.
I then reckoned that I would limit the number of attributes I measured based on roughly 50% of total attributes available (so 18 attributes to be considered). That meant that for attackers - there were only 14 relatively meaningful attributes, and for defenders - 16.
4) I decided to "fill in the missing attributes" by comparing the Attribute Testing with ykykyk weights for each position. So, Finishing isn't represented in the Attribute testing, but sure seems important to attackers in terms of weights, so I added that.
5) This all resulted in the following:
So you see the attributes, the numbers in the first row represent the ykykyk weights, the second row numbers are the impact on Goals Allowed difference or Goals Scored difference.
The red numbers represent extrapolation (complete hogwash, just interested in looking at unmeasured attributes i added, and if the attribute wasn't measured, what is the possible impact largest impact based on attributes with similar ykykyk weights). I pretty much know the red numbers are hogwash because Off The Ball is weighted highly in some positions, but with no demonstrable impact in the attribute testing results.
From here, I would simply replace my current approach (weights-only used in ranking, with this new idea.
Thoughts?
Amazing work again AlexH.
As I've already said on a different post, I am sceptical of ykykyky's ratings for each role. However I concur that we should try to put more effort into finding the most relevant attributes for each role, as you've done here.
I'm confident the team at SI rarely changes existing simulation algorithms and just tries to add more 'situations' every year. However this opinion is based on nothing, other than the relevant attributes from the attribute testing table don't seem to shift that much every year.
I do have to say I am a bit confused by your last graph with the final results, in particular with the Goals Allowed difference/Goals Scored difference. What testing did you do to get these differences?
Is it possible to reduce the attacking and defensive template into weights, for each attribute with these findings?
I appreciate the effort put into this, so I feel bad about giving negative feedback. The problem of trying to work from the attribute testing is that these are single variable tests; any positional differences are averaged out. Also when you try to combine the attribute tests you are not accounting for the correlations between the different attributes that exist in the player database; a player with 20 pace is going to be very likely to have high acceleration and very unlikely to have extremely low acceleration. So if you treat the effects of pace and acceleration as additive you will be overestimating their combined effect. The value of the machine learning approach is that it tests all variables simultaneously and accounting for position.
Hi,
@skawkclsrn - GA/GS difference testing comes from the attribute testing on this site. It’s specious for some of the reasons @MeanOnSunday mentions. But I think it’s somewhat meaningful in that we’re looking not just at the inputs into the simulator (as ykykyk or Mark’s or SIs or whomever) but shows us how the simulator responds. It allows us to ask ourselves questions like, “if ‘Off the Ball’ isn’t meaningful in a controlled environment - how could it be meaningful in my environment - (a 4-2-4 in my case)?”
@MeanOnSunday
First, let me be the first to say I’m not good at analytics, soccer (I’m just a Yank who became interested in it at the tail end of COVID), nor FootballManager (I’m only on my 5th career here).
So negative feedback is EXACTLY what I’m looking for. I hadn’t thought about the correlation and additive nature of calculating.
RE: positional differences being averaged out - wouldn’t that also be a potential benefit? Going into this analysis, I thought about:
1) the fact that it’s not my tactic, and
2) it’s coarse on positional differences, and
3) it’s older (FM22), and
4) it doesn’t test all attributes, and
5) some of the differences may not be statistically significant (+2 goals over 913 simulations? Meh?).
So the compromise I’m offering is not weighting every attribute, but simply trying to combine those that the simulation seems to respond to, and then adding enough attributes that are NOT measured in the attribute testing with ykykyk weights.
Finally, I’m starting to think that any of these approaches are “good enough to be good enough.” Using @Mark weights - I’m dominating MLS on the Touch version. Like no scum save just went a season without losing. Would it even matter if I switched to ykykyk? Is my success because I “cheated” came onto this site, and emulated 4-2-4 Bombyte Tweak more than finding really great players?
I mean, I had success in Ligue 1 with Lorient (!) just measuring how many attributes weighted above 50 were above 16 for each player. Not nearly as granular as any of the above - but still successful.
As long as I know that P/A rule, and the stars are a lie, and I’m playing a proven tactic - is the whole XLS exercise just there for me to not ever overpay for Xavi Simmons again?
@AlexH
I don’t think you can go wrong with either the machine learning weights, Mark’s, or the more recent weights from squirrel_says. The AI in the game is really not very smart and seems to value players based on CA and reputation. So if you play a half decent tactic with players selected based on what matters in the ME, and managing fitness and morale properly, you can win everything with an average CA substantially less than the top AI teams. My point was more a theoretical one that a “better” rating system should consider how the attributes are correlated since its easy to over- or under-weight attributes based on tests done one variable at a time.
Hi Everyone.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
You can access it here.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook.
olimorris said: Hi Everyone.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
You can access it here.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook.
No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features?
Gracolas said: No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features?
I find the GS workflow to be really slow. Loading the DB takes ages. If I wish to change the ratings I have to go into so many different windows + tabs and I have to remember to apply them each time I load the tool up .
I can't speak for other projects but I plan on including a MoneyBall element soon and combining that with the machine learning ratings.
olimorris said: I find the GS workflow to be really slow. Loading the DB takes ages. If I wish to change the ratings I have to go into so many different windows + tabs and I have to remember to apply them each time I load the tool up .
I can't speak for other projects but I plan on including a MoneyBall element soon and combining that with the machine learning ratings.
Ok, makes sense.. I dont mess up with the ykykykyk ratings, so i just leave them unchanged. In this case i find GS really smooth and quick.
Gracolas said: Ok, makes sense.. I dont mess up with the ykykykyk ratings, so i just leave them unchanged. In this case i find GS really smooth and quick.
It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script.
olimorris said: Hi Everyone.
It's been a while and I took that time to turn the cruddy Jupyter Notebook into a command line app. It's wayyyy easier to use than a Jupyter Notebook and I've put a detailed usage guide over in the repository.
You can access it here.
It uses the machine learning weights by default and outputs to a HTML file that you can open and search in the browser...otherwise it's the exact same as the Jupyter Notebook.
Hi, thank you for your work with pyscoutfm. I tried to use your tool by following your guide on GitHub, but I run across a ExPathEvalError on the report generation step. How do I fix this error?
Honestly, for everyone having problems using the Jupyter Notebook or simply dont know how to. Squirrel released an app which is way simpler to use.
https://fm-recruitment-2ac38cc99aa0.herokuapp.com/
I've just added a section around how you can install this in the cloud with minimal effort - here. I recommend doing that and then following the guide on how to Generate a Scout Report.
Appreciate not many people have experience working with Python and it can messy to install if you're on Windows.
I've added an FAQ section comparing PyScoutFM to Squirrel's, here. Hard not to make it look like a d*ck measuring contest but I think it's the best way to show how the tools differ. I'll stand by the outputs from PyScoutFM compared to Squirrel's as it uses the ykykykyky_balanced ratings that @Mark put together in the Genie Scout filter file all those months ago.
Watching how Squirrel explains his rating system (I've also linked to this in the FAQ section), they're mostly based on how FM colours various attributes for a given role. With the exception of Acceleration, Pace and Finishing for an F9 which Squirrel marks as "Key" and multiplies by 5. He also grades other attributes as "Green" and "Blue", which he multiplies by 3 and 1, respectively. Unless I've missed something in his previous videos, I don't understand the delta of 2 between the Key and Green and the Green and Blue. In other words, why is it 2 and not 3.7 or some other made up number?!
Anyway, I've created this tool to scratch my own itch as I prepare to do an English Level 10 LLM adventure in FM24. I wanted to share it with this thread as thanks for all of the amazing effort and content that so many people have put in. I'm likely to expand it in the coming weeks with some Moneyball stats and ratings.
If you're hitting any weird issues or can't get it loaded, open up a discussion in the repo and I'll be happy to help.
olimorris said: I've just added a section around how you can install this in the cloud with minimal effort - here. I recommend doing that and then following the guide on how to Generate a Scout Report.
@Floppyaams this is probably going to be the easier way of using PyScoutFM. Reach out to me if you need any help.
MeanOnSunday said: It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script.
I mean, opening the correct view in FM, copying and pasting, opening the right file and all the other minor things you have to do also take like a minute or two in total. Plus you actually need to be doing stuff in that minute instead of 1 click to load database and do other things while it finishes.
Gracolas said: No offense to everyone putting the effort on building these pyton scripts. I can only imagine how many hours you've put into this, since im a dumbass in terms of programming.
But honest question: why would i use any of these methods, if i have genie scout doing exactly the same plus more features?
I want to add my view here, I have been using python scripts for more than 8 months to play football manager and I have my code base than what @olimorris or Squirrel's has created. So the reason why I like it is
1. Mac: I play in Mac M1, so Genie scout is not an option and FMRTE requires me to by-pass security in place. So the only way I can do some analytics is to use excel file or Python and Python is much more comfortable.
Since I haven't used Genie Scout or FMRTE, take the below comments with a grain of salt and I might be wrong
2. Own Squad Evaluation: I use a script that quickly gives me the Min, Max, Mean and Median score for each positions in my squad with a click of button. This is what I use in order to plan my future recruits. Not sure if this is possible in other tools.
3. Moneyball: I know some might think it's cheating / making the game easy playing the game with attributes unmasked (Hey, its a single player game do what's fun for you right :P). For those, the only way you can do it is using python script. FM Stag released his latest stats for different positions and what those means. I have used in a python script to come up with a score for each player I can sort based on it. @olimorris : I know you said you have plans to go here, so maybe this can help you with it right (again this is just one way to do it). See the below query will provide a score of 0 if the Hdrs W/90 <=2.76 and 1 if its between 2.76 and 5.08 and if its more than 5.08 a score of 2. I do the same for all the different stats and then add em up to come up with a single score for my striker which then I will use for recruitment.
def calculate_stat_striker(row):
conditions = {
'Hdrs W/90': [(5.08, 2), (2.76, 1)],
'Drb/90': [(3.05, 2), (1.17, 1)],
'NP-xG': [(0.5, 2), (0.35, 1)],
'ShT/90': [(1.57, 2), (1.09, 1)],
'Conv %': [(0.3, 2), (0.21, 1)]
}
score = sum(value for col, ranges in conditions.items() for threshold, value in ranges if row[col] > threshold)
return score
MeanOnSunday said: It’s loading the db from the game that’s slow; it takes a couple of minutes for something that should take 1-2 seconds. Otherwise you can easily make your own ratings for GS to match any particular python script.
Personally, i dont think that makes much difference.. i regularly load GS after a couple of in-game days, so ratings dont change that much.
I appreciate the work in Python. For those asking "why don't" - well, some of us do a bit more. I have my own ratings (customized to my passing-heavy style of play), my own metrics (ratings against "average" or "best in the world" ) and my own views (what is my best squad for national competitions?).
Diversity of thought and effort is pretty good. But I'm sure if you just use Squirrel Plays or whatever, you'll be fine too
Yarema said: I mean, opening the correct view in FM, copying and pasting, opening the right file and all the other minor things you have to do also take like a minute or two in total. Plus you actually need to be doing stuff in that minute instead of 1 click to load database and do other things while it finishes.
I was only commenting on GS, not making a comparison. The point is that something that can be coded to run almost instantly (and is in other tools) takes minutes in GS. The rating part of it is (to me) easier in GS than changing a python script, and can get the same result.
AlexH said: I appreciate the work in Python. For those asking "why don't" - well, some of us do a bit more. I have my own ratings (customized to my passing-heavy style of play), my own metrics (ratings against "average" or "best in the world" ) and my own views (what is my best squad for national competitions?).
Diversity of thought and effort is pretty good. But I'm sure if you just use Squirrel Plays or whatever, you'll be fine too
still using mdw22 ratings in GS?
NandaldiaN said: still using mdw22 ratings in GS?
Yes I am.
Can I just say, totally off topic, that this post is in the top half a dozen viewed posts of all time on FM Arena with over 130k views, and the highest of any non tactic post on the site. So clearly a subject that is important to the FM Community.
Out of interest, the top 2 by a massive margin, are both from the great @ZaZ with ZaZ Blue 4.0 in FM22 tactics with over 630k views, and ZaZ Blue 3.0 in FM21 tactics with just under 580K views. In third is Phoenix v5.0 in FM21 tactics with over 230k views.
Mark said: Yes I am.
Can I just say, totally off topic, that this post is in the top half a dozen viewed posts of all time on FM Arena with over 130k views, and the highest of any non tactic post on the site. So clearly a subject that is important to the FM Community.
Out of interest, the top 2 by a massive margin, are both from the great @ZaZ with ZaZ Blue 4.0 in FM22 tactics with over 630k views, and ZaZ Blue 3.0 in FM21 tactics with just under 580K views. In third is Phoenix v5.0 in FM21 tactics with over 230k views.
That is actually nice to know! That means people need you for positional filters in FM24.
Hello, MDW22 is fantastic for me in FM22. are you using MDW22 for only positional rating? what do you think about role ratings?
Im using ZaZ - Blue 4.0. For AMR positon I have Damsgaard and Raphinha.
Positional Rating
Raphinha 84
Damsgaard 83,5
Role Rating For Winger Support Role
Raphinha 88
Damsgaard 91
Which player do you choose?
One more there are 3 different rating file
YKYKYKY
MDW22
Zaz
Which one is better for Zaz Blue?