Working with complex custom objectives

[DenisSvirin]

How can I add and evaluate function to the objective ?

Essentially, I want to also search for functions for polynomials in your example (https://astroautomata.com/PySR/examples/#9-custom-objectives). Something like this f(P(X)) / g(Q(X)), where f = P(X) ** y or something else.

I tried to reserve first l and r nodes for these functions and use the rest for polynomials:

    tree.degree != 2 && return L(Inf)
   
    attraction_func = tree.l
    repultion_func= tree.r
    
    attraction_func != 2 && return L(Inf)
    repulsion_func != 2 && return L(Inf)
   
    attraction = attraction_func.l
    repulsion = repultion_func.l

but it gives inf loss and only complexity of 1

[MilesCranmer]

Is this

attraction_func != 2

Supposed to be attraction_func.degree != 2?

[DenisSvirin]

yes, thanks.
But how i can extract the function?

[MilesCranmer]

What do you mean by extract?

[DenisSvirin]

As i said i want now to apply functions attraction_func and repulsion_func to the trees attraction and repulsion.

These trees are already just numbers. So i basically have only prediction of these trees and can't use Node method.

I suppose i can make new tree with a constant and these functions, but i wonder if there is any easier method

[MilesCranmer]

You mean like this?

    attraction_func = tree.l.r
    attraction = tree.l.l
    attraction_func_val, flag = eval_tree_array(attraction_func, dataset.X, options)
    !flag && return L(Inf)
    attraction_val, flag = eval_tree_array(attraction, dataset.X, options)
    !flag && return L(Inf)

    result = attraction_func_val ^ attraction_val