Hi everyone,
I am new here and wanted to reach out for any advice/resources around pymc modelling for football/soccer.
I am currently learning about Bayesian process and pymc via this project and have managed to get a decent base model adapting the Dixon Coles model structure but have hit a roadblock when expanding it.
The block is more about expanding the model network itself rather than code implementation. For example, I am unsure how to include league effects: if I want to include Championship data to capture promoted teams strength from previous seasons etc. I understand this is quite vague but I wanted to kick off basic discussions first before getting into the weeds!
Thanks,
Hi there,
I see that you haven’t received an answer yet so I’ll have a go in case you’re still interested!
The problem about including Championship data is a bit tricky indeed. The issue is that Championship teams will generally not play against Premier League teams, except if they’re promoted or relegated, right? That’s not great since there won’t be a lot of mixing between the levels. Still, I think you could just give it a go and see if you get a reasonable result.
Beyond that I’m also not super sure. One thing you could try is to modify the prior. I’m guessing each team’s strengths is drawn from some distribution, probably with the same mean and variance? If so, you could draw the Premier League initial skills from a different distribution, maybe with a mean that is strictly larger to indicate that they’re better.
Anyway, hope this is helpful – curious to hear what you think if you’re still working on this!
Martin
Hi Martin, that sounds like an interesting approach regarding the promotions.
I had parked this project for the meantime, and have been exploring machine learning methods to predict football matches. However, I am sure I will come back to the Bayesian model in the future.
Louis
1 Like