Mark said: As I said, I only used the first 6 tests, so is your data showing the average from all tests where you have tested 4000 matches or only the first 2400 matches? Expand
The thing is, the order in which the result screenshots are displayed isn't the order in which the tests happened. The result screenshot are randomly shuffled.
Mark said: I also only looked at the first 6 tests (ie 2400 matches) for each HOF tactic to ensure they were also all comparable. Expand
As you can see we've got 1,200 matches, 2,400 matches and 4,000 matches in our test result tables and we compare all them just fine.
You just need to calculate P.P.M.(Points Per Match) and then translate it for 38 matches.
Total Points / Total Matches Played = P.P.M.(Points Per Match)
P.P.M.(Points Per Match) x 38 = expected points after 38 matches
Mark said: I accept the statistical noise for less than 2 points. I was just trying to look at how I can use a combination of tactics through the season to my advantage. Expand
One important thing to note, we are talking about 1-3 points RNG when there's 1,200 matches and we are talking about 0.5 - 1.5 points RNG when there's 4,000 matches.
But if we take 2,400 matches then in this case a tactic is tested against a specific formation only for 480 matches.... could you imagine the level of the RNG at such small distance.
For example, look at your result for "424 deformation II" tactic vs "5-2-3" AI formation, it's "42.5" points after 6 runs, but after 10 runs it drops to "36" points so the RNG of 6 runs is about "6.5" points.
As I said when the RNG could be as high as "6.5" points, an attempt to rank tactics in 0.5-2.5 points range doesn't make sense.
dzek said: How did you test all the formations? AI against AI? Expand
dzek said: If we had a H2H tactic testing league with all the existing formations that the game has for AI managers then we would have a more accurate answer I believe on this case. Expand
Hi,
If you take the best 200 AI managers in the game then:
- 35% of them have 4-3-3 Preferred Formation - 35% of them have 4-2-3-1 Preferred Formation - 20% of them have "different variations of 5 at the back formation" Preferred Formation - 10% of them have 4-4-2 Preferred Formation
So 70% of 200 best AI managers in the game play with 4-3-3 or 4-2-3-1
Also, look at the performance of AI's 5-2-3, it's significantly worse than AI's 4-2-3-1 or AI's 4-2-3-1.
We tested the rest of AI formations and they did even worse than AI's 5-2-3 so what's the point in having in the testing league?
According to our data If one tactic does better than other tactic vs AI's 4-2-3-1 or AI's 4-2-3-1 then it also will do better against other AI formations which are even weaker.
So when the overall difference between tactics is just 1-2 points then when it comes to a specific formation then the difference is also 1-2 points and such small difference is just "statistical noise".
Because I've checked our DB and here's what I got:
Red General 4231 v1.0
424 CF Cabuloso
433 Tsukyomi I
According to your table, "Red General 4231 v1.0" tactic is ranked "8th" and "424 CF Cabuloso" tactic is ranked "1st" when it come to 4-3-3 formation but according to our numbers the difference is about 1.5 points which could be just "statistical noise".
If you ask me then I think it makes sense to "rank" tactics only when there's a meaningful difference between them, I'd say it should be at least 10-15 points difference but when the difference is just 1-2 points then it can be easily "statistical noise".
Also, according to your table "433 Tsukyomi I" tactic is ranked "1st" when it comes to 5-3-2 formation but that's not true.
Here're some conclusions that we can draw from the result above:
- obviously, "meta" attributes is the real thing
- it doesn't seem there're minimal/optimal levels for other non-meta attributes that must be reached
- "non-meta" attributes still also matter because increasing 4 meta attributes to 20 gave less boost to the score than increased only 1 meta attribute but having other "non-meta" attributes at a much higher level - https://fm-arena.com/tactic/7528-5-points-to-pace-for-all-positions/. Yes, "non-meta" attributes also matter but their impact is much less than the impact of "meta" attributes.
Also, there's one important thing to understand when in our tactic testing league we set Acceleration attribute of a player to "20" then his Accelerating attribute will be 5-6 points higher than the Acceleration attribute any other player in the league.
But when you play the game in normal game and you try to outsmart the AI mangers by focusing on the meta attributes, you really shouldn't expect "miracle results", especially, when you play in the top leagues such as English Premier League and other because just look at the players in the top clubs of such leagues:
As you can see there's a lot of players in the top clubs that have the meta attributes at 17-20 level so when you play in the top leagues and even if you manage to bring in your club players that have the meta attributes at 17-18 level then still you won't be dominate the league because there're many clubs that also have a lot of player with maximized meta attributes.
letsgo9 said: @Zippo@Droid Would make any difference test all the Hall of Fame tactics till 4000 matches? Expand
Hi,
On average, the change in score between 2,400 matches and 4,000 matches is just -/+ 1 point.
Bear in mind, it would take quite a lot computing power to test every tactic that gets in HoF for 4,000 matches and what for? Just to correct its score by -/+ 1 point?
Taking into consideration, the quite low attention that HoF tactics get in general, I'd say the idea to spend a low of computing power on testing them for 4,000 matches to correct their score just -/+ 1 point isn't a good one.
The score change:
1,200 matches -> 2,400 matches = -/+ 1-3 points
2,400 matches -> 4,000 matches = -/+ 0-1 points
So if a tactic got "58" points after 2,400 matches then it almost has no chances to get higher than "59" if it gets tested up to 4,000 matches.
I just want to let you know that we've decided if after 1200 matches a tactic gets a score that is 2 points away from the top tactic then it'll get an additional testing up to 2400 matches.
At the moment there're about 12 tactics that scored at least 61 points(2 points away from the top) after 1200 matches in the queue to get tested up to 2400 matches.
pixar said: wow. Sharp decline after the storm. I think you'll need to change the amount of matches made in tests. A 5 point drop is a serious amount.
I think the amount of 4000 matches in each test may tire the system. I'm not too sure about that, but it can be considered to be at least 3000-2500 matches.
sponsorkindest said: Sorry if this has been already answered, I thought the decision was to do attribute testing after December release, did something change? Just curious question, I don't think we need a re-run given the results are in line with FM23 but just wanted to understand the reasoning Expand
Hi,
We thought we hadn't time to do the attribute testing that early but we managed to do that before the patch.
nios said: Do we have any news about when the attribute test is going to take place? Is it scheduled after the update that SI announced in early December? Expand
Yes, we're planning to do attribute testing after that update.
dzek said: I said that those instructions (Get Stuck In + Tackle Harder) which theoretically do the same thing MAY improve the score. Of course there is RNG on the results of each test. This is clear Expand
It's very important to understand that the RNG factor is always present and there're many tactical instructions that make such a small difference that can't even beat the RNG factor and the result might vary not because of those tactical changes but due to the RNG factor.
The thing is, the order in which the result screenshots are displayed isn't the order in which the tests happened. The result screenshot are randomly shuffled.
Mark said: I also only looked at the first 6 tests (ie 2400 matches) for each HOF tactic to ensure they were also all comparable.
As you can see we've got 1,200 matches, 2,400 matches and 4,000 matches in our test result tables and we compare all them just fine.
You just need to calculate P.P.M.(Points Per Match) and then translate it for 38 matches.
Total Points / Total Matches Played = P.P.M.(Points Per Match)
P.P.M.(Points Per Match) x 38 = expected points after 38 matches
Mark said: I accept the statistical noise for less than 2 points. I was just trying to look at how I can use a combination of tactics through the season to my advantage.
One important thing to note, we are talking about 1-3 points RNG when there's 1,200 matches and we are talking about 0.5 - 1.5 points RNG when there's 4,000 matches.
But if we take 2,400 matches then in this case a tactic is tested against a specific formation only for 480 matches.... could you imagine the level of the RNG at such small distance.
For example, look at your result for "424 deformation II" tactic vs "5-2-3" AI formation, it's "42.5" points after 6 runs, but after 10 runs it drops to "36" points so the RNG of 6 runs is about "6.5" points.
As I said when the RNG could be as high as "6.5" points, an attempt to rank tactics in 0.5-2.5 points range doesn't make sense.
dzek said: How did you test all the formations?
AI against AI?
We use a much more sophisticate method for that.
Hi,
If you take the best 200 AI managers in the game then:
- 35% of them have 4-3-3 Preferred Formation
- 35% of them have 4-2-3-1 Preferred Formation
- 20% of them have "different variations of 5 at the back formation" Preferred Formation
- 10% of them have 4-4-2 Preferred Formation
So 70% of 200 best AI managers in the game play with 4-3-3 or 4-2-3-1
Also, look at the performance of AI's 5-2-3, it's significantly worse than AI's 4-2-3-1 or AI's 4-2-3-1.
We tested the rest of AI formations and they did even worse than AI's 5-2-3 so what's the point in having in the testing league?
According to our data If one tactic does better than other tactic vs AI's 4-2-3-1 or AI's 4-2-3-1 then it also will do better against other AI formations which are even weaker.
Best Of Both Worlds v4
424 deformation II
So when the overall difference between tactics is just 1-2 points then when it comes to a specific formation then the difference is also 1-2 points and such small difference is just "statistical noise".
Hi,
I have no idea what testing methodology was used to get those numbers so I can't say anything about it.
Xeno94 said: or the weighting on the editor highly value "decisions" for multiple positions.
If your are talking abut the official attribute weighting then it's been proven many times by our tests that they are far away from being "correct".
Are you sure your numbers are correct?
Because I've checked our DB and here's what I got:
Red General 4231 v1.0
424 CF Cabuloso
433 Tsukyomi I
According to your table, "Red General 4231 v1.0" tactic is ranked "8th" and "424 CF Cabuloso" tactic is ranked "1st" when it come to 4-3-3 formation but according to our numbers the difference is about 1.5 points which could be just "statistical noise".
If you ask me then I think it makes sense to "rank" tactics only when there's a meaningful difference between them, I'd say it should be at least 10-15 points difference but when the difference is just 1-2 points then it can be easily "statistical noise".
Also, according to your table "433 Tsukyomi I" tactic is ranked "1st" when it comes to 5-3-2 formation but that's not true.
Just wanted to wish you, your families and the team a very merry Christmas and a happy new year.
Thank you for all the hard work you've put in this year delivering outstanding testing
All the best
Chris
Thank you, @CBP87.
I wishing everyone hope, peace and lots of Christmas cookies this holiday season!
I'm sure many of you guys have seen the results of our Player Attributes Testing for FM24 - https://fm-arena.com/table/26-player-attributes-testing/
The results of the testing have raised many important questions and there're a few of them:
- Should you pay attention only to meta attributes such as Pace, Acceleration and some other?
- What about other attributes? Do they matter at all?
- Are there any "optimal" levels for non-meta attributes that must be reach before you can ignore them?
So we've done another testing in the hope that it can answers the questions above.
On the screenshots below you can find the default attributes of the players in our tactic testing league.
FORWARDS
MIDFIELDERS
CENTRAL DEFENDERS
FULL BACKS
Ok, now let's edit every player in the human controlled teams and set their non-meta attributes to "7" and their meta attributes to "20".
So the attributes of every player in the human controlled teams look like this:
IMPORTANT NOTICE: The CA of every player has decreased by 40-50 points.
Now, let's test re-test "Red General 4231 v1.0" tactic with the adjusted players - https://fm-arena.com/thread/8026-red-general-4231-v1-0/
Here're some conclusions that we can draw from the result above:
- obviously, "meta" attributes is the real thing
- it doesn't seem there're minimal/optimal levels for other non-meta attributes that must be reached
- "non-meta" attributes still also matter because increasing 4 meta attributes to 20 gave less boost to the score than increased only 1 meta attribute but having other "non-meta" attributes at a much higher level - https://fm-arena.com/tactic/7528-5-points-to-pace-for-all-positions/. Yes, "non-meta" attributes also matter but their impact is much less than the impact of "meta" attributes.
Also, there's one important thing to understand when in our tactic testing league we set Acceleration attribute of a player to "20" then his Accelerating attribute will be 5-6 points higher than the Acceleration attribute any other player in the league.
But when you play the game in normal game and you try to outsmart the AI mangers by focusing on the meta attributes, you really shouldn't expect "miracle results", especially, when you play in the top leagues such as English Premier League and other because just look at the players in the top clubs of such leagues:
As you can see there's a lot of players in the top clubs that have the meta attributes at 17-20 level so when you play in the top leagues and even if you manage to bring in your club players that have the meta attributes at 17-18 level then still you won't be dominate the league because there're many clubs that also have a lot of player with maximized meta attributes.
Hi,
On average, the change in score between 2,400 matches and 4,000 matches is just -/+ 1 point.
Bear in mind, it would take quite a lot computing power to test every tactic that gets in HoF for 4,000 matches and what for? Just to correct its score by -/+ 1 point?
Taking into consideration, the quite low attention that HoF tactics get in general, I'd say the idea to spend a low of computing power on testing them for 4,000 matches to correct their score just -/+ 1 point isn't a good one.
The score change:
1,200 matches -> 2,400 matches = -/+ 1-3 points
2,400 matches -> 4,000 matches = -/+ 0-1 points
So if a tactic got "58" points after 2,400 matches then it almost has no chances to get higher than "59" if it gets tested up to 4,000 matches.
I just want to let you know that we've decided if after 1200 matches a tactic gets a score that is 2 points away from the top tactic then it'll get an additional testing up to 2400 matches.
At the moment there're about 12 tactics that scored at least 61 points(2 points away from the top) after 1200 matches in the queue to get tested up to 2400 matches.
Cheers.
Hi,
As you might noticed every tactic has a star rating: "Excellent", "Good", "Ok" and so on...
If some tactics have the same rating then it means that our tactic testing considers them to be equally good in real game.
So you can pick any tactic you like with "Excellent" rating and you really won't notice any difference in real game.
Also, tactics with "Good" rating can be almost as good as "Excellent" tactic.
I think the amount of 4000 matches in each test may tire the system. I'm not too sure about that, but it can be considered to be at least 3000-2500 matches.
@Zippo
Here's our testing algorithm for the current patch(24.2.0):
- the top tactic gets 4,000 matches
- all tactics in Hall Of Fame get 2,400 matches
- all other tactics get 1,200 matches
Would be nice just to understand the baseline everything else is working off of. Thanks
FORWARDS
MIDFIELDERS
CENTRAL DEFENDERS
FULL BACKS
Hi,
We thought we hadn't time to do the attribute testing that early but we managed to do that before the patch.
Cheers.
https://fm-arena.com/table/26-attribute-testing/?mark=7544
take a look
It's been fixed.
Thanks.
We took this tactic as the base tactic - https://fm-arena.com/thread/7174-cameron-4-2-3-1-bal/
Is it scheduled after the update that SI announced in early December?
Yes, we're planning to do attribute testing after that update.
Yes, but we're considering to start attribute testing after the next M.E. update, which should happen soon according to SI.
it's done - https://fm-arena.com/board/13-set-piece-sharing-section/
Cheers.
It's very important to understand that the RNG factor is always present and there're many tactical instructions that make such a small difference that can't even beat the RNG factor and the result might vary not because of those tactical changes but due to the RNG factor.
So don't be tricked by the RNG factor.
That's what I tried to say.
dzek said: It boosts “Get Stuck In” much more.
I's been said a thousand times already... ok, I'll say it once more.
The RNG of 1,200 tests is 1-3 points.
The RNG of 4,000 tests is 0.5-1.5 points.
At least that RNG numbers were true for FM23.
We haven't measured the RNG for FM24 yet. It might be higher or lower.
What does all that above mean?
It means that let's say if you test some tactic for 100,000 matches and you get 60 pts then you can call it "the true score" of this tactic.
But if you test only for 1,200 matches then you can get 57,58,59,60,61,62,63 pts.
So adding "Tackle Harder" PI might don't do anything and this time it just hit the highest RNG for this tactic.