Hans
Hans
See #131 and #130
0.001 is quite a high value for protected division. Good work on noticing this issue.
Evolutionary speaking it would make more sense to assign a penalty to those functions. Giving them the worst fitness is akin to simply deleting them from the population.
> @hwulfmeyer That is true. But removing numerically unstable individuals means the remaining ones will be stable. It is objectively much worse to be unstable, in a model. If you...
I have privately made my own additions to gplearn which includes [ParetoGP](https://doi.org/10.1007/0-387-23254-0_17) and EPLEX. See also #33 I would be more than happy to build a PR from my work....
Beginning this now. Not sure when I will be done.
So, as evident I did not get around to doing this. I do not know if I will ever have the motivation to do it at all as writing it...
@MilesCranmer What exactly do you mean with "how to use these techniques"? Programmatically, Theoretically? :D
I added additional hyperparameters/options: complexity => 'kommenda' (for the kommenda complexity) selection => 'eplex' paretogp => 'True' or 'False' paretogp_lengths => (a, b) ParetoGP works by selecting the first parent...
What might at least be considerable is the way the closure is performed. I personally don't care so much about how large the EPS is but how it is applied....