I have data collected from a computer simulation of football games which seem to have recurring patterns of the following form.

if madrid plays arsernal and the match ends under 3 goal, then on their next match against each others, madrid will win. if madrid happens to loose and then plays against chelsea next, they will win 90% of the time.

how do I find such inferences from simulation generated data like this. There are other forms of hidden patterns that I believe exists in the dataset.

  • $\begingroup$ You really can't draw inferences from simulated data, and, in most cases, not even from real data. Since you have so many parameters to play with, you will always (through sheer chance) find some "pattern" that doesn't really exist. $\endgroup$
    – user4710
    Oct 27, 2014 at 21:10
  • $\begingroup$ @user4710 - i disagree. You can't draw inferences about real football, but you can draw plenty of inferences about video-game football. This sounds like they want a random-forest fit, then they want to make a single-cart approximation of the random forest, and look at the splits and decision points for probability of win. $\endgroup$ May 5, 2021 at 2:49
  • $\begingroup$ Is this a reasonable dataset for an example solution? github.com/jmerullo/football $\endgroup$ May 5, 2021 at 14:49

1 Answer 1


Define a specific environmental state (e.g., madrid plays arsernal and the match ends under 3 goal), then run the simulation many times each time with a random seed. The result will be a distribution of simulated outcomes. Summary statistics can be computed on the distribution of outcomes.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.