let's say you upload a 4-2-4(DMs) tactic and after only 1 run(400 matches) it scores 50 pts, obviously, after such low score for 4-2-4(DMs) formation our tactic testing algorithm would stop testing it further.
Let's assume that the hitting 50 pts after the first 400 matches was very unlucky event and if we continue testing the tactic further up to 4,800 matches it would ended with a higher score about 60 pts but our tactic testing algorithm won't be testing it further because he knows(based on the statistic) that yes, probably, if it were tested further up 4,800 matches it could improved the score from 50pts to 60pts but anyway the score still would be lower than the highest score for the formation which is 63.2 pts at the current moment.
The purpose of fm-arena tactic testing is to find the most effective tactics to play FM and IS NOT to compete with other tacticians for the 1st place in the table or IS NOT determine the score of every uploaded tactic with a precise accuracy like 0.5 pts or 1 pts or 2 pts because something like that would required testing EVERY uploaded tactic for more than 30,000 matches which is no feasible task I'd say.
We have a tactic testing algorithm that is responsible for determining how many matches to test every tactic and it constantly evolving.
I'll try to explain how it works:
Your and Gerrard's tactics both use 4-4-2(DMs) formation and that means your tactics compete with the highest ranked 4-4-2(DMs) tactic so far which is this - https://fm-arena.com/thread/12717-ef-424-if-hp-v2-p101-ac/. As you can see it scored 63.2 pts.
During the test of your tactic after 2400 matches our testing algorithm looked at the current score and concluded then there's no point in any further testing your tactic because with such low score after 2400 matches it has no chances to beat the highest score for that formation which is 63.2 pts.
During the test of Gerrad's tactic after 3600 matches our testing algorithm looked at the current score and concluded then there's no point in any further testing your tactic because with such low score after 3600 matches it has no chances to beat the highest score for that formation which is 63.2 pts.
Obviously, during the test of Gerrard's tactic there was more positive RNG then in the test of your tactic that's why Gerrard's tactic hold testing for longer than your tactic but anyway as you can see none of the tactics reaches 4,800 matches because our tactic testing algorithm stopped the testing earlier.
Our tactic testing algorithm doesn't care about the accuracy of the test of your and Gerrard's tactics, because it concluded that none of the tactics can't beat the highest score for the formation which is 63.2 pts.
Are we allowed to post mirror versions of tactics? Expand
Hi,
No, uploading mirror versions of OWN tactics or OTHER people tactics is NOT ALLOWED on here because it would be waste of our testing server's computing power.
If we notice that some person uploads mirror versions of HIS OWN or OTHER people tactics then we remove his ability to upload tactics on our site.
Please note, I won't be deleting this thread/tactic because the computing power of our testing server was used to testing this tactic and we don't want it to be wasted.
BUT I'll rename this thread to "EF 424 IF HP V2 P101 AC ( Mirror Version )" and also edit the opening post and put a link to the original tactic.
CBP87 said: I don't mean no disrespect to you or the creators of the tactics but I don't see how that is a fair comparison on the tactics around these 2. All other tactics around these have had 9600 matches and has more than likely had a drop in points over them additional 4800 matches but yet are showing further down the table against a tactic that has been tested for half of the games they have.
Thanks Expand
The purpose of fm-arena tactic testing is not to create competition between tactic creators, the purpose is to help people to find the most effective to play FM and try to reveal how FM work under the hood.
I also want to add that at the moment we find the accuracy of fm-arena tactics is already excessively high and something much more less accurate would work just fine to play FM.
I mean an accuracy of +/- 5 pts or even -/+ 10 pts would be enough already and gave people a massive edge over AI managers in the game. We really don't need an accuracy like +/- 1 or 2 pts or even 0.1, 0.2, 0.3 pts.
At this moment the highest score is 63.2 pts(the 1st place) so to be tested for more than 4800 matches (up to 9600 matches) a tactic must score higher than 63.2 pts after 4800 matches and keep getting a score above 63.2 after every additional test until it reaches 9600 matches.
dzek said: I just think that some TIs can show some signs of difference from their opposites without necessarily being affected by other combinations. Expand
@dzek, even putting aside the fact that TIs and PIs interact with each other and that make a difference then still testing TIs and PIs would be just a waste of time because it isn't possible to "break" 1 points RNG(at least in our testing), even after 100,000 matches tested we still find there's "1" point RNG but many TIs/PIs make less than "1" point difference.
For example, for 400 matches tested the RNG can be as high as about "7" points and for 800 matches tested the RNG can be as high as "5" points and if you want to bring the RNG to the minimal number then you have to test no less than 50,000 matches and even after 50,000 there's still 1 point RNG but many TIs and PIs make less than 1 point difference.
I've really had enough experience with testing "TIs/PIs", I came to a conclusion that testing "TIs/PIs" is a stupid thing to do. Why that? Because at some point it became obvious that there's an interaction between "TIs/PIs", in other words, for example, "Be More Disciplined" TI might work "good" in 4-3-3 shape with many "Attacking" duties and "Attacking" mentality but it might work "poor" in 4-2-4 shape with many "Support" duties and "Balanced" mentality. So you have to test all possible combinations(shapes/TIs/PIs/) to find out what every tactical instructions do and that's is madness I'd say.
Also, it turned out the RNG in the testing is much higher than I thought it was. For example, even a test of 9,600 matches still has +/- 1.5 points RNG, which mean the same tactic after 9,600 matches still can produce 60 points, 61.5 points, 63 points results but many TIs/PIs do like 0.1 points difference and testing them would be just playing with the RNG.
dzek said: I apologize but I don't know him either and I don't have enough time anymore to follow this situation.
EDIT: I think the only way you can contact the author of FMRTE is through their Discord channel or through their forum. In the past I suggested an improved formula regarding the coach rating system and they implemented my suggestion into their program. Expand
FMRTE is capable of "freezing" many things in the game
but many of them aren't really important, I mean, of course, it's great if they are "frozen" but their values can be checked with much lower frequency than the values of really important things such as the values Conditions and Morale, which must be checked and set to 100% every 2-3 seconds.
It would be great if you made a guide that shows how to "freeze" the morale and condition of all players in a specific league/club using Cheat Engine.
Thank you for your involvement.
EDIT: @pichxyaponn, ok... I've figured out how to "freeze" with Cheat Engine using the tables you posted but is there a way to "save" the freezing setup once you made it in order not to do that all over aging every time you load your testing game save?
Long story short: in our tactic testing we use FMRTE ( https://www.fmrte.com/ ) to "freeze" the Morale and Conditions(and other attributes) of all the players in the testing league.
FMRTE is a great tool but it's a bit "slow", by calling it "slow" I mean that it fails to set the Morale and Conditions of all the players in the league to 100% between matches during a simulation in our tactic testing league.
When we're simulation a season there's like 2-3 seconds between matches and there's about only 200 players in our tactic testing league, I really think that 2-3 seconds should be enough to set the Morale and Conditions for all of that players.
I'm not sure and it might be that the author of FMRTE intentionally made it that "slow" in order to not overheat PC of regular players but we aren't "regular" players.
Anyway, if someone can offer a solution to this problem that definitely reduces the RNG in our testing league and increases the accuracy of the results.
Germaniac said: will the results still be representative of a normal gameplay? Expand 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?
Germaniac said: Why can't you be more specific? Expand 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.
Nah, I wouldn't call it "a drastic swing". It's just about 6 - 9 points drop across the board, which translates into 8%-12% points drop in points. No way 8%-12% points drop can be considered as "a drastic swing", to define it as "a drastic drop" it should not be no less than 50% or more.
In general, if we take a regular season that consists of 38 matches then 6 - 9 points drop means that you get just 2 or 3 wins less than before.
So in the new DB the AI managers show a bit more resistance which translates into 9%-12% points drops for the human managers and that's all.
More importantly, in the new DB the balance between tactics has stayed 99% the same as it was before.
CBP87 said: What did you change that resulted in such a drastic swing in the previous top tactics? Expand
Some settings that are responsible for the way AI managers play. Sorry, I can't be more specific here.
We've updated the testing league and made changes that increase the difficulty in the league, now tactics should be more efficient to come on top and also, the changes reduce the RNG which improves the accuracy of the results.
Our dedicated algorithm determines how many matches to test a tactic and it's constantly adjusting.
CBP87 said: I've noticed over the last few days that tactics that have scored relatively high, have only been tested for a few thousand matches. Expand
It's just the scores that were considered by the algorithm as "relatively high scores" some time ago, now aren't considered as "relatively high scores" due to the increase of the top score.
I just want to let you know that we've fixed a small bug with notifications.
Probably you might noticed that notifications about "approval your tactic for the testing" were appearing again after they were viewed or deleted. Now, it should be fixed and such notifications should be cleared correctly.
let's say you upload a 4-2-4(DMs) tactic and after only 1 run(400 matches) it scores 50 pts, obviously, after such low score for 4-2-4(DMs) formation our tactic testing algorithm would stop testing it further.
Let's assume that the hitting 50 pts after the first 400 matches was very unlucky event and if we continue testing the tactic further up to 4,800 matches it would ended with a higher score about 60 pts but our tactic testing algorithm won't be testing it further because he knows(based on the statistic) that yes, probably, if it were tested further up 4,800 matches it could improved the score from 50pts to 60pts but anyway the score still would be lower than the highest score for the formation which is 63.2 pts at the current moment.
I hope it helps.
Cheers.
The purpose of fm-arena tactic testing is to find the most effective tactics to play FM and IS NOT to compete with other tacticians for the 1st place in the table or IS NOT determine the score of every uploaded tactic with a precise accuracy like 0.5 pts or 1 pts or 2 pts because something like that would required testing EVERY uploaded tactic for more than 30,000 matches which is no feasible task I'd say.
We have a tactic testing algorithm that is responsible for determining how many matches to test every tactic and it constantly evolving.
I'll try to explain how it works:
Your and Gerrard's tactics both use 4-4-2(DMs) formation and that means your tactics compete with the highest ranked 4-4-2(DMs) tactic so far which is this - https://fm-arena.com/thread/12717-ef-424-if-hp-v2-p101-ac/. As you can see it scored 63.2 pts.
During the test of your tactic after 2400 matches our testing algorithm looked at the current score and concluded then there's no point in any further testing your tactic because with such low score after 2400 matches it has no chances to beat the highest score for that formation which is 63.2 pts.
During the test of Gerrad's tactic after 3600 matches our testing algorithm looked at the current score and concluded then there's no point in any further testing your tactic because with such low score after 3600 matches it has no chances to beat the highest score for that formation which is 63.2 pts.
Obviously, during the test of Gerrard's tactic there was more positive RNG then in the test of your tactic that's why Gerrard's tactic hold testing for longer than your tactic but anyway as you can see none of the tactics reaches 4,800 matches because our tactic testing algorithm stopped the testing earlier.
Our tactic testing algorithm doesn't care about the accuracy of the test of your and Gerrard's tactics, because it concluded that none of the tactics can't beat the highest score for the formation which is 63.2 pts.
I hope it helps.
Cheers.
Sandro said: @Zippo, I've got a question.
Are we allowed to post mirror versions of tactics?
Hi,
No, uploading mirror versions of OWN tactics or OTHER people tactics is NOT ALLOWED on here because it would be waste of our testing server's computing power.
If we notice that some person uploads mirror versions of HIS OWN or OTHER people tactics then we remove his ability to upload tactics on our site.
Please note, I won't be deleting this thread/tactic because the computing power of our testing server was used to testing this tactic and we don't want it to be wasted.
BUT I'll rename this thread to "EF 424 IF HP V2 P101 AC ( Mirror Version )" and also edit the opening post and put a link to the original tactic.
Cheers.
Thanks
The purpose of fm-arena tactic testing is not to create competition between tactic creators, the purpose is to help people to find the most effective to play FM and try to reveal how FM work under the hood.
I also want to add that at the moment we find the accuracy of fm-arena tactics is already excessively high and something much more less accurate would work just fine to play FM.
I mean an accuracy of +/- 5 pts or even -/+ 10 pts would be enough already and gave people a massive edge over AI managers in the game. We really don't need an accuracy like +/- 1 or 2 pts or even 0.1, 0.2, 0.3 pts.
Hi,
No, this tactic won't be tested further.
At this moment the highest score is 63.2 pts(the 1st place) so to be tested for more than 4800 matches (up to 9600 matches) a tactic must score higher than 63.2 pts after 4800 matches and keep getting a score above 63.2 after every additional test until it reaches 9600 matches.
I hope it helps.
Cheers.
@dzek, even putting aside the fact that TIs and PIs interact with each other and that make a difference then still testing TIs and PIs would be just a waste of time because it isn't possible to "break" 1 points RNG(at least in our testing), even after 100,000 matches tested we still find there's "1" point RNG but many TIs/PIs make less than "1" point difference.
For example, for 400 matches tested the RNG can be as high as about "7" points and for 800 matches tested the RNG can be as high as "5" points and if you want to bring the RNG to the minimal number then you have to test no less than 50,000 matches and even after 50,000 there's still 1 point RNG but many TIs and PIs make less than 1 point difference.
Could you do another test with the TI "Focus Play Through The Middle"?
Hi, @dzek.
I've really had enough experience with testing "TIs/PIs", I came to a conclusion that testing "TIs/PIs" is a stupid thing to do.
Also, it turned out the RNG in the testing is much higher than I thought it was. For example, even a test of 9,600 matches still has +/- 1.5 points RNG, which mean the same tactic after 9,600 matches still can produce 60 points, 61.5 points, 63 points results but many TIs/PIs do like 0.1 points difference and testing them would be just playing with the RNG.
I hope you got what I mean here.
EDIT: I think the only way you can contact the author of FMRTE is through their Discord channel or through their forum. In the past I suggested an improved formula regarding the coach rating system and they implemented my suggestion into their program.
Hi, @dzek.
I already sent him DM on Discord yesterday.
Haven't received any replay yet but let's wait.
Probably, if no luck with Discord then I'll try to reach him on their forum.
Thanks.
Honestly, no idea how to contact him and in general don't have much free time lately.
If you know him then I delegate you to speak on behalf of fm-arena
but many of them aren't really important, I mean, of course, it's great if they are "frozen" but their values can be checked with much lower frequency than the values of really important things such as the values Conditions and Morale, which must be checked and set to 100% every 2-3 seconds.
Hi, @pichxyaponn.
It would be great if you made a guide that shows how to "freeze" the morale and condition of all players in a specific league/club using Cheat Engine.
Thank you for your involvement.
EDIT: @pichxyaponn, ok... I've figured out how to "freeze" with Cheat Engine using the tables you posted but is there a way to "save" the freezing setup once you made it in order not to do that all over aging every time you load your testing game save?
Unforutanlty, the SI in-game editor can't "freeze" the morale and conditions. It would be great if it could do that.
Long story short: in our tactic testing we use FMRTE ( https://www.fmrte.com/ ) to "freeze" the Morale and Conditions(and other attributes) of all the players in the testing league.
FMRTE is a great tool but it's a bit "slow", by calling it "slow" I mean that it fails to set the Morale and Conditions of all the players in the league to 100% between matches during a simulation in our tactic testing league.
When we're simulation a season there's like 2-3 seconds between matches and there's about only 200 players in our tactic testing league, I really think that 2-3 seconds should be enough to set the Morale and Conditions for all of that players.
I'm not sure and it might be that the author of FMRTE intentionally made it that "slow" in order to not overheat PC of regular players but we aren't "regular" players.
Anyway, if someone can offer a solution to this problem that definitely reduces the RNG in our testing league and increases the accuracy of the results.
Thank you for your time.
Cheers.
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?
Germaniac said: Why can't you be more specific?
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.
Hi, @CBP87.
Nah, I wouldn't call it "a drastic swing". It's just about 6 - 9 points drop across the board, which translates into 8%-12% points drop in points. No way 8%-12% points drop can be considered as "a drastic swing", to define it as "a drastic drop" it should not be no less than 50% or more.
In general, if we take a regular season that consists of 38 matches then 6 - 9 points drop means that you get just 2 or 3 wins less than before.
So in the new DB the AI managers show a bit more resistance which translates into 9%-12% points drops for the human managers and that's all.
More importantly, in the new DB the balance between tactics has stayed 99% the same as it was before.
CBP87 said: What did you change that resulted in such a drastic swing in the previous top tactics?
Some settings that are responsible for the way AI managers play. Sorry, I can't be more specific here.
I hope that helps.
Cheers.
Hi,
We've updated the testing league and made changes that increase the difficulty in the league, now tactics should be more efficient to come on top and also, the changes reduce the RNG which improves the accuracy of the results.
Our dedicated algorithm determines how many matches to test a tactic and it's constantly adjusting.
CBP87 said: I've noticed over the last few days that tactics that have scored relatively high, have only been tested for a few thousand matches.
It's just the scores that were considered by the algorithm as "relatively high scores" some time ago, now aren't considered as "relatively high scores" due to the increase of the top score.
I hope that helps.
Cheers.
Thanks @smigler and Ill upload a positive version later
Fixed.
I just want to let you know that we've fixed a small bug with notifications.
Probably you might noticed that notifications about "approval your tactic for the testing" were appearing again after they were viewed or deleted. Now, it should be fixed and such notifications should be cleared correctly.
Thanks to @dzek for point it out.
Cheers.
Ohh... it's that thing. Ok, I'll check whether it can be tested or not.