bnveg

3 ms·

1. 1.▲
   by bnveg 6y ago

   The model shown in the article is cross-validated.
2. 2.▲
   by bnveg 6y ago

   Now do that with cross validation and get a better result than the reported
3. 3.▲
   by bnveg 6y ago

   >Since you have a literal dependency between one data point and the next, you can't train your model using randomized data This is why the train/test split was not randomized, but sequential.
4. 4.▲
   by bnveg 6y ago

   This is an out of sample result. 4 formulas were compared. Sure it may be an overfit, as any machine learning model can be, but the procedure that was used to generate this formula cannot be so easily dismissed.
5. 5.▲
   by bnveg 6y ago

   As a rule of thumb, you cannot. I have tried to fit the Mona Lisa with the software (brighness as a function of x and y) and could not find anything. It performs poorly when no structure exists.
6. 6.▲
   by bnveg 6y ago

   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.
7. 7.▲
   by bnveg 6y ago

   This is out of sample performance. Only 4 models were compared outside the train domain. Also I don't see where you see 22 parameters, the complexity of a model is not defined as its number of parameters.
8. 8.▲
   A mathematical formula that predicts US elections with 87.5% accuracy(turingbotsoftware.com)

   8 points by bnveg 6y ago|27 comments