How the AI thinks

Groundhog Day Experiments, First Article

Introduction

Have you kids watched Groundhog Day, starring Bill Murray?

It's a really good movie! The plot revolves around this guy, who starts re-living the same day, over and over again. You should definitely watch it if you can!

This concept can be applied to FMM18 as well. Normally, it's called cheating - reloading a save, specially when things don't go your way

However, it is a very powerful tool to use when trying to understand the game. To answer many what if questions:

• What if I had used a Control Mentality instead of Attacking, would I have won the game?
• What if my player was not sent-off?
• What if I had picked Agüero instead of Gabriel Jesus?

You get the drill.

Still, replaying the game one time will not necessarily give you all the answers you want. There's a lot of random outcomes in each game, and even if you play the same tactic, with the same players, you will get a different outcome most of the time.

How do I know that? Well, By playing the same game ten times over!

Why to play the same game over and over again?

My initial idea was testing a specific tactical formation against another specific tactical formation. The question I wanted to answer was "If I play this opponent who uses an Attacking 4-2-3-1, is there an optimal tactic to counter it? However, I never got to answer this question...

But keep reading - I discovered other things! After playing a few games, I got answers to questions I was not asking. For example:

• How the AI behaves depending on the match circumstances?
• How predictable the AI is?
• Is there any logic for the AI tactical decision in game?
• How much variation in the game stats are there after playing the same game using the very same tactic?

Sounds quite interesting, right? (I hope so!)

The setup

• Manager: Unknown reputation, no badges
• Loaded leagues: English Premier only
• The team: Almighty Bournemouth
• Other: Holiday through pre-season, until the 10th of August.

The Groundhog Day

Actually, 3 days!

• Groundhog start: 10th of May
• Groundhog end: 12th of May, after match
• Opponent: Leicester, away. First game of the league season.
• Odds: Leicester 1.18, Bournemouth 4.0 - "A difficult test"

Tactics

Bournemouth

• Always the same tactic - a flat 4-4-2 (G, FB, CD, CD, FB, W, CM, CM, W, TM, P)
• Balanced/Normal Mentality, Width, Tempo and Creative Freedom
• Balanced/Normal Defence, Pressing, Tackling - No time wasting nor offside trap.
• No special options, Mixed passing, distribution and focus.
• Mostly the same players (selected by the assistant)

 

Starting 11 - Bournemouth

GK	G	Federici (10 games)
DR	FB	Francis (9 games) / Smith (1 game)
DL	FB	Daniels (10 games)
DC (R)	CD	Aké (9 games) / Cook (1 game)
DC (L)	CD	Cook (9 games) / Francis (1 game)
MR	W	Fraser (10 games)
ML	W	Pugh (10 games)
MC (R)	CM	Gosling (10 games)
MC (L)	CM	Surman (10 games)
FC (R)	P	Defoe (7 games) / King (3 games)
FC (L)	TM	Wilson (10 games)

 

 

 

 

 

 

 

Leicester

The AI was quite consistent - always the same tactic, always the same players.

• Always the same tactic - a 4-2-3-1 (G, FB, LD, CD, FB, BWM, BBM, IF, SS, IF, P)
• Always the same players
• Always started with Attacking Mentality, Balanced Width and Balanced Creative Freedom.
• Started 2 games using Slow Tempo, 8 games Normal Tempo.
• Not possible to see the other options (Defence, Attack) 

 

Starting 11 - Leicester

GK	G	Schmeichel (10 games)
DR	FB	Amartey (10 games) - Accomplished Aptitude
DL	FB	Fuchs (10 games)
DC (R)	CD	Morgan (10 games)
DC (L)	LD	Huth (10 games)
MC (R)	BBM	King (10 games)
MC (L)	BWM	Silva (10 games)
AMR	IF	Mahrez (10 games)
AML	IF	Gray (10 games)
AMC	SS	Okazaki (10 games)
FC	P	Vardy (10 games)

 

Results (Finally...)

All that said, I have not looked into individual player's performance - sorry! That would be soooo much work. Read: maybe another time

Overall winner:

• Leicester: 4 games
• Draw: 1 game
• Bournemouth: 5 games

Analysis: Overall, not a bad showing for Bournemouth. Given the odds of a Bournemouth win were 4.00 - which converts to a 25% probability of winning - a 40% winning ratio would mean that the bookies would lose some money overall

Average goals:

• Leicester: 2,2
• Bournemouth: 2,3

Analysis: Quite a high-scoring affair in general, to be understood in more detail soon!

Possession:

• Bournemouth:
  • Min: 45%
  • Max: 59%
  • Avg: 49%

Analysis: A 14% variation from the worst to best, highly dependent on the game circumstances. Overall, a 49% average.

Shots/Shots on Target:

• Leicester:
  • Min: 8 / 4
  • Max: 19 / 9
  • Avg: 11,7 / 5,6
• Bournemouth
  • Min: 5 / 2
  • Max: 18 / 8
  • Avg: 10,8 / 5,5

Analysis: Also, a large variation depending on each game, even though the same tactic was used by each team every time. Points to highly circumstantial in-game events.

Chances:

• Leicester:
  • Min: 0
  • Max: 3
  • Avg: 1,2
• Bournemouth
  • Min: 0
  • Max: 2
  • Avg: 1

Analysis: I never quite understood how chances are calculated in game, so I can't read much into this.

Successful Passes (%)

• Leicester:
  • Min: 66%
  • Max: 79%
  • Avg: 73,5%
• Bournemouth
  • Min: 64%
  • Max: 77%
  • Avg: 73%

Analysis: A 13% variation from worst to best in both teams, with roughly the same players. Is that a lot? Leave your opinion below

Successful Crosses (%)

• Leicester:
  • Min: 11%
  • Max: 31%
  • Avg: 20,4%
• Bournemouth
  • Min: 7%
  • Max: 37%
  • Avg: 22,3%

Analysis: The crossing data is a bit more interesting than the passes. It makes you believe that Bournemouth is slightly better at crossing, but it also shows way more variation game-by-game. It may vary more than passes due to the fact that there are fewer crosses each game.

Successful Tackles (%)

• Leicester:
  • Min: 78%
  • Max: 91%
  • Avg: 84,4%
• Bournemouth
  • Min: 80%
  • Max: 100%
  • Avg: 92,4%

Analysis: Anyone know how this correlates with real-world data? It seems to be very high in my opinion! Actually, Bournemouth went to games without missing a single tackle! Impressive!

Successful Headers (%)

• Leicester:
  • Min: 46%
  • Max: 76%
  • Avg: 64,8%
• Bournemouth
  • Min: 33%
  • Max: 63%
  • Avg: 44,2%

Analysis: Now it is obvious that Leicester is much better on the high balls. The game analysis duly noted that Bournemouth strikers struggled to win aerial duels in 6 of the 10 games. Big variation from game to game, but overall consistent with the players' heading ability.

 

Offsides

• Leicester:
  • Min: 0
  • Max: 6
  • Avg: 3,3
• Bournemouth
  • Min: 0
  • Max: 5
  • Avg: 2,1

Analysis: Leicester tends to be offside more often. Can't pinpoint a definite reason: it could be the style of play or the attacking mentality.

 

Committed Fouls

• Leicester:
  • Min: 4
  • Max: 13
  • Avg: 10
• Bournemouth
  • Min: 8
  • Max: 16
  • Avg: 11,7

Analysis: Bournemouth plays rougher - they have to. May indicate worse players at tackling.

 

Yellow / Red Cards

• Leicester:
  • Min: 0 / 0
  • Max: 4 / 1
  • Avg: 1,9 / 0,1
• Bournemouth
  • Min: 0 / 0
  • Max: 4 / 1
  • Avg: 2,0 / 0,1

Analysis: Very balanced between teams, not much to read here.

 

What have we learned so far?

Well, the game seems pretty random, no? 4 wins from home, 5 from away in 10 games where the setup was almost identical. A lot of the stats had quite big differences from game to game, and both teams seemed quite equivalent in a lot of things (except tackling and heading). So does it mean that whatever we do, it will not really affect the outcome? Is there any point in playing this?

Calm down José, the most interesting stuff comes soon!

We are definitely missing something - that's why I give you the Detailed game-to-game Analysis!

 

Game to game Analysis

If you're still here, you are in for a treat! I have followed closely the tactical changes in the AI throughout every game, so let's check it out!

Groundhog day 1: Leicester 0 - 3 Bournemouth

• After Bournemouth scored the first goal on the first minutes of the game, Leicester mentality changed to Overload
• After Bournemouth scored the 3rd goal, Leicester changed mentality to Contain and width to Narrow in order to avoid further embarrassment

Groundhog day 2: Leicester 1 - 2 Bournemouth

• Leicester started the game with a Slow tempo. Rather odd.
• After Bournemouth scored the first goal on the first minutes of the game, Leicester changed mentality to Overload and the tempo to Normal
• During the game, the inside forwards (Mahrez and Gray) swapped positions for roughly 10-15 minutes. Fancy strategy there.

Groundhog day 3: Leicester 4 - 1 Bournemouth

• Again, after Bournemouth scored the first goal on the first minutes of the game, Leicester mentality changed to Overload
• Bournemouth was down to 10 men (CM was sent off, no tactical changes)
• After scoring the third goal, Leicester changed mentality to Control, width to Narrow and tempo to Slow in order to administer the advantage.
• After the minute 65, the mentality was further decreased to Balanced.

Groundhog day 4: Leicester 3 - 1 Bournemouth

• No goal from Bournemouth in the first minutes, but Leicester changed the mentality to overload in minute 10. Strongest team on the paper, needs to score.
• After scoring 1-0, Leicester changed mentality to Balanced and width to Narrow to administer the advantage.
• Bournemouth scored, 1-1. Leicester returned to an Overload mentality and Balanced width in order to try to score.
• It worked, game is now 2-1. Balanced mentality and Narrow width.
• In the end of the game, Leicester scored again. 3-1, Contain, Narrow and Slow to slowly wait for the final whistle.

Groundhog day 5: Leicester 2 - 4 Bournemouth

• Again, Overload after 10 minutes (0-0)
• Bournemouth scored twice (0-2), no changes in Leicester mentality.
• After scoring twice (2-2) in minute 70, Leicester decreases mentality to Control - still attacking, but with more caution
• Bournemouth scored a third 2-3, Leicester went Overload again.
• Game over: 2-4.

Groundhog day 6: Leicester 3 - 5 Bournemouth

• Again, Overload after 10 minutes (0-0)
• Bournemouth scored twice, Leicester pegged one back, Bournemouth scored again. (2-3)
• After scoring again (3-3) in minute 65, Leicester decreases mentality to Attacking - still attacking, but with a tad bit more caution
• Bournemouth scored a fourth 3-4, Leicester went Overload again.
• Game over: 3-5.

Groundhog day 7: Leicester 3 - 4 Bournemouth

• Again, Overload after 10 minutes (0-0)
• Mahrez-Gray swap strategy used again! Yeah!
• Balanced mentality, Narrow width after scoring (1-0) - minute 44.
• Overload mentality, Balanced width after Bournemouth scored (1-1)
• Balanced mentality, Narrow width after scoring again (2-1) - administer advantage.
• Overload mentality, Balanced width after Bournemouth scored another (2-2) - minute 54.
• That backfired, as Bournemouth scored twice and won the game 3-4.

Groundhog day 8: Leicester 3 - 1 Bournemouth

• Again, Overload after Bournemouth scored on minute 6 (0-1)
• Balanced mentality, Narrow width after turning the score around (2-1) - minute 44.
• Contain after minute 85, game was already 3-1.

Groundhog day 9: Leicester 2 - 2 Bournemouth

• Again, Overload after 10 minutes (0-0)
• Mahrez-Gray swap strategy used again! Loving it!
• Bournemouth took the lead twice (0-1, 1-1, 1-2)
• Leicester got the draw on minute 92 (2-2), changed mentality to Contain to guarantee the late draw.

Groundhog day 10: Leicester 1 - 0 Bournemouth

• Again, Overload after 10 minutes (0-0)
• Balanced mentality, Narrow width after scoring (1-0) - minute 44.
• Control, Balanced width on minute 58. Perhaps chasing a second goal?
• Balanced mentality after minute 69 - maintain advantage
• Contain, Narrow after receiving a red card. 10 men Leicester.
• Played a strikerless formation after left back got sent out. Kept the victory.

Wow, that was a lot to take in, wasn't it. So, what have we learnt:

How the AI thinks (TL/DR):

• The AI is more clever than I expected. It changes it's playing style to respond to in-game events.
• Leicester had the advantage on paper, much stronger team. Always started Attacking.
• Leicester always changed mentality to Overload if suffering a goal, or whenever the clock hit the 10 minute mark.
• Leicester often changed the playing style to Balanced if leading, sometimes accompanied with a Narrow width.
• When losing badly (first game), AI changes mentality to Contain in order to avoid embarrassment - if the clock is too close to 90 minutes.
• Twice the AI started playing with a Slow tempo. It can also change the tempo to slow in the end of a winning game.
• AI also uses Control and Attacking when the game is a draw or it has a one goal lead if it thinks there is enough time to get another goal.
• AI got very nervous after the red card - strikerless, Contain, riding the lead to the end
• AI also uses Contain in end-games when winning.

End notes

That was much longer than I planned

I hope this can help you all to realise that the AI is not a static machine. It reacts to the game, and so should you. Don't be afraid to use Counter when the AI get desperate, or to even resort to Contain to maintain a lead.

The AI is not as predictable as I thought, the Mahrez-Gray swap was quite cool to witness! And it has nuances to its decisions, just like a real footie manager.

I hope the read was enjoyable and that it gives whoever is reading this many ideas and a new view of the game we all love

Thanks and it is nice to be back!

 

[FuddledFox]

This is a great insight into how the AI thinks and shows that a little more attention on our part when we play matches could really help boost the results we get.

Those tackling % seem a little far fetched to me but I don't know how they would relate to real world %. I find that even my defensive players don't make a lot of tackles any way compared with what I would expect in real life.

Great work.

[mcandrew003]

And he's back! Really interesting article mate, top job. 

[Dar J]

AI switching to overload too early in the game has been fixed with the update. They still change mentality though during games

[PriZe]

Some cracking work mate! Great read  

[Rob]

Bloody heck. Wow.

[Real_Random]

Impressive article, very informative, inntereeting to see how the AI works. KIU

[rafa]

3 hours ago, FuddledFox said:

This is a great insight into how the AI thinks and shows that a little more attention on our part when we play matches could really help boost the results we get.

Those tackling % seem a little far fetched to me but I don't know how they would relate to real world %. I find that even my defensive players don't make a lot of tackles any way compared with what I would expect in real life.

Great work.

Thanks mate! I agree, the tackles are way out there! 

2 hours ago, mcandrew003 said:

And he's back! Really interesting article mate, top job. 

Not back like in the old days, but maybe one article here and there every month or so

Thanks

2 hours ago, Dar J said:

AI switching to overload too early in the game has been fixed with the update. They still change mentality though during games

Hah I need to check this out! I thought my game was updated, but it turns out that it wasn't! Thanks for the heads up! 

[samhardy]

See knew it wasn't possible for you to just "snoop around"

Really interesting stuff mate, as always. Glad to have you back. Already looking forward to another holidaying manager spectacular!

[Woody]

I enjoyed this....cracking article. 

[Ashez]

Very interesting, nice work Rafa :).

[henda666]

interesting

[Titjes]

Fantastic write-up and very educational, thanks  

[rafa]

10 hours ago, samhardy said:

See knew it wasn't possible for you to just "snoop around"

Really interesting stuff mate, as always. Glad to have you back. Already looking forward to another holidaying manager spectacular!

Haha, well... I'm not sure if I will have time to post more. Lots of ideas, very few time!

The holidaying manager was fun, wasn't it? Quite a mediocre performance if I remember well

[Twister045]

Great write up mate, it’s articles like this that whet your FMM appetite!

I think to an extent there has to be a random element to matches as real life managers can only do so much. Ultimately it’s how the manager reacts to the flow of the game and then up to the players to carry out instructions – they shouldn’t be able to carry these out perfectly every time as there should always be a chance of individual mistakes or pieces of brilliance depending on their ability and traits

I suppose the only thing that would worry me slightly is a supposedly ‘weaker’ Bournemouth side managed to win more often without making any tactical changes..

[Buzz JR]

Nyc topic....have had difficult episodes managing Wigan from the Skybet league one esp in the Carabao cup so far.played liverpool 16 times until i won 2 nil. Kept on repeating da game coz i felt i had a chance....3 times the game went to penalties n i lost, lost by a 1 goal margin 10 times n was massacred  3 times.Used da same 3 5 2 formation ...

[rafa]

4 hours ago, Twister045 said:

Great write up mate, it’s articles like this that whet your FMM appetite!

I think to an extent there has to be a random element to matches as real life managers can only do so much. Ultimately it’s how the manager reacts to the flow of the game and then up to the players to carry out instructions – they shouldn’t be able to carry these out perfectly every time as there should always be a chance of individual mistakes or pieces of brilliance depending on their ability and traits

I suppose the only thing that would worry me slightly is a supposedly ‘weaker’ Bournemouth side managed to win more often without making any tactical changes..

Yeah, nice insight here! I guess the main issue for Leicester in this case was the constant switching to Overload mentality, which opened their defence up. This is supposedly fixed for the update, which I did not have installed when I run this experiment 

[rafa]

13 minutes ago, Buzz JR said:

Nyc topic....have had difficult episodes managing Wigan from the Skybet league one esp in the Carabao cup so far.played liverpool 16 times until i won 2 nil. Kept on repeating da game coz i felt i had a chance....3 times the game went to penalties n i lost, lost by a 1 goal margin 10 times n was massacred  3 times.Used da same 3 5 2 formation ...

Haha, I normally don't replay games when playing the game - but I'm happy you're having fun! Nice victory, nonetheless!

[Buzz JR]

Just now, rafa said:

Haha, I normally don't replay games when playing the game - but I'm happy you're having fun! Nice victory, nonetheless!

Thank yu Rafa.....im on a challenge to get to the First division bt im on a very low budget.winning such matches may hlp me out

[Twister045]

5 minutes ago, rafa said:

Yeah, nice insight here! I guess the main issue for Leicester in this case was the constant switching to Overload mentality, which opened their defence up. This is supposedly fixed for the update, which I did not have installed when I run this experiment 

Yeah the use of overload seemed excessive but you'd assume with that fixed it would reduce the amount of goals and make matchdays a little more challenging too which is always welcome

[samhardy]

6 hours ago, rafa said:

Haha, well... I'm not sure if I will have time to post more. Lots of ideas, very few time!

The holidaying manager was fun, wasn't it? Quite a mediocre performance if I remember well

I have fond memories of it. Think it was one trophy in 30 years but... not the point. Assman was the most famous member of vibe at that point