- Aug 18, 2009
- 4,230
- 17,507
- AFL Club
- Richmond
- Thread starter
- #701
The algorithm is explained in more detail than you could ever want here.I don't have any massive issues about where we are at, our defence looks strong on the chart and our lack of height is reflected that our offense is off the mark so far this year. I am just not sure if the algorithm is a true reflection based on the opposition, when and where you play them so not sure it is a good predictive algorithm or more a captain obvious position based more on basic metrics.
If it is just a basic metric algorithm then a small sample size can significantly vary, if it accounts for quality of opposition, conditions, days break between games vs the opposition and projects forward based on the remaining games then it shouldn't vary significantly if it is semi-accurately predictive.
I haven't tried to model days' break before. I really should.
There's nothing wrong with "basic" algorithms, though. There is a curse in modern statistics called over-fitting, where people throw every variable they have into a computer and have it spit out a model. And this seems like a good idea, because more variables must mean more accuracy, right? But no. This requires a proper post to explain, which I will do if I get distracted enough, but essentially with enough variables, you produce a model that fits past data very well but is terrible at predicting the future, because you're modelling random noise.
As the great mathematician John von Neumann once said, "With four parameters I can fit an elephant, and with five I can make him wiggle his trunk."




