Introduction
I will be conducting a controlled experiment to see how much a team talk effects your team in Football Manager 2008.

I had suspected in previous versions of FM that the team talks had too large of an impact on performance. Almost like the game in reality is a multiple choice drop down list, and if you choose the right things to say, your tactics, training, and signings don't matter. Now i will actually put this hypothesis to the test.

VM Ware
Unlike most AI experiments i have seen, this one will be a controlled experiment, using VM Ware Workstation. For those who don't know what VM Ware is, i'll give you the basics.

VM Ware is basically a "virtual" computer that can be installed within your own computer. It acts just like a real PC, so it won't work unless you install Windows and give it software.

The good thing about VM Ware when conducting experiments is, the "state" of the computer can be saved at any time to be restored later. That means that, as long as you keep the same tactics and talks, the match will play out THE EXACT SAME WAY every time you reload. If you change one thing (half time sub), the match will be different, but if you reload, and make the same half time sub, the match will play out EXACTLY like it did the first time you made the sub. If you've never used VM Ware before, it's quite scary to see!

Procedure
I haven't fully decided how i will conduct the experiment. At the moment i am thinking that i will take control of a few different "types" of teams. The following 6 clubs are being considered:

Chelsea - Massive club with superstar players
Newcastle - Thinks they're a massive club, notorious in recent years for their players perhaps not listening to the manager.
Cheltenham - mid-sized club in terms of England. I just needed a team "in the middle", and Cheltenham is right next to Chelsea in the list of clubs
Braintree - Amongst the smallest clubs that can be controlled in England. Managers seemingly have the greatest impact at this level.
Seongnam - Big Korean club. No one in the team speaks English. Will my team talk even register?
Changwon City - Small Korean club (second division). Again, no English speakers on the team.

I wanted to choose a team with all youngsters, but the only clubs like that are in foreign countries, so i would be combining two variables and it would not be controlled. Perhaps i will do this in the future if i can conclusively determine the effect of team talks on foreign players.

I will create 12 Managers and give them control of the 6 teams in two different save games. One save will consist of managers with no pre-set experience. The other game will have 6 managers who are former International Footballers.

For each manager, i will use nothing but a default 4-4-2, choosing the best 11 that they can play in a league match. At the start of the match, i will try one of the available team talks, and play until half time. Then i will reload VM Ware, and play the first again with a different team talk. I will repeat until all talks are used.

Then i will choose the best and worst half time results, and run each of those games from half time onward, using each of the available team talks. At this time i will not be looking at the effect of individual player talks. This will result in approximately 48 iterations of the same match using the same random seed on every matchday.

I will post only my data without analysis, until there is sufficient data (or lack thereof) to draw a conclusion.

I will not begin this experiment until later this evening, so if anyone has suggestions, feel free to contribute.


-Chas

Sounds intresting, I wholly agree with you about the team talks though. They do seem a bit too important, well thats my take on them anyway. I'll be keeping an intrest on here

Sounds like hard work.

Good luck!

Good luck mate. Seems like a well-thought experiment. Most of us, players, I guess always wondered how team talks effects the team.

Sounds really well thought out!

Will check back to see the results!

Sounds good. Good Luck

ill be watching. great idea. itll help deffo

clap clap clap, nice idea!!!

im unsure as to whether the game will play out the same way. isnt the match engine a load of formulas deciding what a playe will do and what the chances are he will manage it etc. (obviously a bit more complicated than that tho). If the chances a player will complete it ae 50/50 then in one game he may manage it but on reload he may not. im not entirely sure how all this works tho

If you don't have VM Ware Workstation, i suggest you try a demo sometime.

You get the SAME EXACT result, down to the slightest kick of a ball, every time you reload from the same point.

I've tried it. Suppose you save right at the kick off. The game goes 5-0 at half time. You reload again, and it will always end 5-0 at half time, and the goals will be exactly the same.

Now suppose you make all three subs at half time, and the game ends up 5-3. Now if you reload from the very beginning of the match, and you make the same three subs at half time, the match will still end up 5-3, and the second half goals will be identical.

VM Ware is scary that way!

To be clear, i'm not reloading the save game. I'm reloading the state of the machine via VM Ware. The random seed therefore stays the same throughout, so any 50/50 situations will always come out the same.

To give another sillier example, suppose you're flipping a coin. You flip it 5 times and get heads every time. After the 5th one, i slap you in the face. From that point on, you get 5 tails in a row.

If i reloaded that "state" with VM Ware, as long as i remembered to slap you in the face after the 5th flip, you will once again get 5 heads followed by 5 tails.

Get it?

oh right cool. sounds like a good idea too

I wonder about taking the best and worst results from each game. Testing only outliers strikes me as slightly flawed in that, although the most interesting scenarios, and, perhaps, the scenarios in which we would most like to know the effects of team talks, it leaves the most likely scenarios out of the experiment.

Sorry; that's a bit garbled. Essentially, I am curious about situations that we have all encountered, like:

-Outplaying the AI, but down at half-time only to go on to get trounced in the second half.

-Trying to hold on, up 1-0 at half against a superior side

I suppose my concern is that the extremely good and extremely bad halftime results are the situation in which the team talk seems to matter less. Up or down 3-0 at half, there is not much a manager can do. On the other hand, closer halftime results could be made or broken by team talks.

I also wonder if you have any plan to test the aggregating effect of team talks. The hints and tips suggest that repetition of team talks through multiple game stretches desensitizes your team to them. Is there a good way to test this?

Marc Vaughan has also suggested that the results of team talks are linked to morale (both team and individual?). If you can make some sense of the team talk AI, it would be interesting to see how one or two individual talks modify things.

This is already a laborious endeavor, but it makes curious. Although I have not had too many incidents in which I suspect my team talk was at fault, I wonder if I have missed some opportunities to turn things around.


Sorry. I'm a bit wordy, but I am quite interested in what you find.
Good luck. I hope you have a lot of coffee/beer/ganja or whatever it is that fuels your FM addiction.

I'm running the initial tests, and i see that running 6 (really 12) teams simultaneously is going to be very difficult in terms of record keeping (not to mention every save state is about 4 gigabytes).

I will conduct the test with one team at a time. When i've completed 30 matches (i don't want to get into details, but 30 is a magic number in statistics that makes experiments "valid"...anything more than 30 is considered overkill), i will move on to the next team.

I'll start with Chelsea.

You raise a great point, JR. I didn't think about that before.

I will test all half time scenarios. The description of the match situation (outplaying them, but trailing, leading a stronger team, etc) is going to be a bit subjective, but i will try and cope.

I could test the effect of desensitizing the team, but it would have to be in a different experiment.

Thanks!

I really like this.

Good luck with it, I will definitely read

quote:

Originally posted by Chas (Psyatika):
You raise a great point, JR. I didn't think about that before.

I will test all half time scenarios. The description of the match situation (outplaying them, but trailing, leading a stronger team, etc) is going to be a bit subjective, but i will try and cope.

I could test the effect of desensitizing the team, but it would have to be in a different experiment.

Thanks!

I definitely understand your reluctance to apply subjective labels to different scenarios. I think you understand the value of investigating the more nuanced situations, though.
Clearly, this is a critical experiment if any other questions about the team talk AI are to be answered. There is really no other place to start. I commend you. Hooray for systematic empirical analysis of trivial things!

And, WOW, 4 gigs per save state! You must have extensive hardware. In order to run a VM, you still need the 'real' memory in order to log the save states, right?

Correct.

I could make my save states smaller if i start with a fresh VM and install nothing but Windows and FM (that's really what i should do), but instead i'm using an existing Windows VM that i had already been using for other things. It has about 4GB of stuff on it (including FM 07), so it all gets saved when i save the state.

I'm currently holiday-ing the Chelsea game to October so the team will have match practice and be ready for guinea pig status. With only the EPL loaded and minimum detail, the game is just flying!

Looking forward to preliminary results!

As an update, i've come up with the idea to use the pre-match odds as a determinant of the level of difficulty of the match. So a 1.03 suggests a match i should easily win, whilst 50.00 is a match i shouldn't bother turning up for. Now the first half stats will actually mean something.

For the second half, i'm not sure how i'm going to measure the match situation. I could use average match rating of the players, or some kind of goal/possession difference factor...not sure.

I ran a few iterations. I will post my dataset whenever i take a break from the test. It is a MS Excel file, colour coded data

Current data - 6 Nov - 18:26

The best part about running an AI experiment with VM Ware, is that you can play FM on your actual computer while you wait for the experiment to play out on the VM!

Updated above file with an abbreviations and notes sheet.

Chas (Psyatika) you conducting this experiment, hope so, because wanted to see whether team talks actually do have an impact.

I'll keep an eye out anyway

Chas (Psyatika) is this still continuing???

is it dead ? seems really interesting, hope you continue