- guessing all the results doesn't guarantee you a win, but they decide the winner at their discretion.
- you have to give them all the source code etc, under generous MIT licence.
- they reserve the right to modify or withdraw the competition at any time.
Hey lazyant, I'm one of the engineers at Football Radar.
Thanks for the comment - we're happy to listen to community feedback, especially if it looks like we're doing something dubious - and I promise, we're not trying to!
I'll try to address your concerns:
* Maybe we could have explained it better, but the competition is about writing the best predictive model, not correctly guessing the results of the tournament. Football results are not deterministic - the best anyone can do is write a thorough probabilistic model that is correct more often than not. We can only judge this competition based on the source code, because the methodology is more important than the outcome.
* We have modified our terms to try and address some of your other points. We ask for submissions to be under the MIT licence so that we can promote the winning entry and share some of the best ideas with the community. We're not interested in using solutions in our own products.
Suggestion: grab the betting odds from Betfair and use those to derive the percentage chances. You'll be piggybacking off of the work done by people and companies who do this for a living and are good at what they do!
Alternatively, if you think that you can do better, then there's no need to enter the competition, as your insights would make you more money through gambling :)
So they're asking for the best algorithm (and the source implementation also) to guess football and other results, which is worth millions, and all you get in return is a crummy laptop? Not much incentive to participate
The contest is for the best predictive model for ... an undefined objective function! Unless I missed it, I didn't see a formula of how the models will be scored. For example, you can have an objective function where predicting the winner gives you a lot more points than predicting which team will not make it out of the group stage. Or you can have one where all the different stages are equally important.
Depending on the scoring method, it might be advantageous to generate solutions that are not mathematically correct, that is, solutions where each row sums to 1, but the columns don't sum to 16, 8, 4, 2, 1, and 1, respectively.
You really need to define what the objective function is here.
Any soccer/football fans will tell you that it's a tough sport to predict for a variety of reasons: low scoring games, lots of potential for human error in officiating, high levels of competence of even weak teams, team dynamic vs individual skill, etc.
I'd guess that it's one of the harder sports to build a prediction model for. Good luck!
But in many ways it is actually a sport that can be modelled fairly accurately. The betting markets have remarkable consistency, in that they 'agree' on what the odds should be for any particular match. Compare this with horse racing, where the odds on any horse vary wildly before the start. For example, a horse might be 10/1 in the morning, 3/1 half an hour before the race and 6/1 at the off. Whereas the odds on a football team will stay fairly constant.
So the matches may be difficult to predict the winner but the markets price up all non-random factors very well.
Yes, this is something every soccer fan knows. Also, the world cup is every 4 years and it's difficult, if not impossible, to predict how a renewed team will work. You can make a cluster of few teams that will win in 99% of the cases but not optimize much more from that.
