A mathematical formula that predicts US elections with 87.5% accuracy

Re:A mathematical formula that predicts US elections with 87... by Imnimo on Thursday June 25, 2020 @01:39PM (#23643262)

I would expect to be able to get 100% (on the training data) on this task. Shouldn't any sequences of election victories be easily represented with multiple sin functions?

I feel like the message I'm supposed to take from this demo is "Look how good TuringBot is! They can automatically find a function to match this data!" But the actual message I'm getting is "Symbolic regression is too hard for TuringBot!"

[ Parent | on HN ]

Re:A mathematical formula that predicts US elections with 87... by frabert on Thursday June 25, 2020 @01:39PM (#23643279)

Indeed. Just perform an FFT on the data and encode the resulting coefficients in a closed formula.

[ Parent | on HN ]

Re:A mathematical formula that predicts US elections with 87... by bnveg on Thursday June 25, 2020 @08:17PM (#23647353)

This is a Pareto optimization with a limit on the size of the formulas. Sure, with a formula of arbitrary size you can fit anything with 100% accuracy, but the task is much harder if what you are looking for are short models.

[ Parent | on HN ]