Project4704 took up a lot of processing time, and it turned out there is a much less labour-intensive way of going about the same task.
To recap, what I had done with Project4704, is to repeat the same friendly match between two edited teams, 4704 times. Some macro scheduling software would load a saved game, play the match, dump the results into a folder as Google Chrome documents, and repeat that six times.
On the seventh repetition – I rested, but the macro scheduler did not – the away team’s formation was changed to a different one of the 28 available AI manager formations, and the six-match block was done again, and then again, until all 28 formations had played against all 28 formations, including itself.
But, wait! There is another way.
What if I did something like this?
That’s to say, what if I ran tests for an entire season, instead of a single game at a time? Of course, it takes far, far less time, and so it’s much easier to create large amounts of data (and, thus, it strengthens the conclusions which you draw.)
So, I went into the pre-game editor, and butchered the whole of the Brazilian top division. All the players have attributes of 10, and are managed by – identical – unsackable human managers. The 30 clubs I created have reputations of 10000, so the players won’t have any desire to leave, and have been placed under a transfer embargo so that they can’t buy anyone. They also have huge squads, full of identical wing-backs, defensive midfielders, attacking midfielders, etc., to take injuries into account.
The division’s 30 clubs are named after the 28 default formations which AI managers can use, and there are also a couple of test formations each time, so I can see how the earlier ideas I have had held up. There are no player instructions or team instructions – everything has been left to the default. The human managers have then been sent on holiday, with ‘Use current match tactics’ ticked.
This is how a single completed season looks:
And this is how 21 seasons of it looks, with the test formations having played a fraction of that number of seasons, and the 28 default teams having played the lot:
Formations with a flat back four, and lots of defensive/central midfielders work, but there needs to be some width there, too. This is essentially what I learnt before, but the conviction has been strengthened due to the amount of data created. (I expect the league-topping 4330 DM Wide to drop down the standings slightly over time. It just happens to be the last one I tried, and then I decided to write something about it.)
What a boost for a small club it would be, to not have to bear the expense (in terms of transfer fees, and wages) of strikers. If a 4330, or 4240, is the way to go, then I’d imagine I can recruit three attacking midfielders or wingers, for the price of two forwards, and the extra body in the squad may be invaluable over the course of a season. It looks as though strikerless.com was always ahead of its time…