What tool should I be using for mapping a function across the complex domain?

I have two datasets (A,B) that are the input and outputs of a complex set of functions. They are (or can be) contained within a poincaré disk. There are over two million samples, spread quite evenly over the problem domain. I want to use a fitting / prediction methodology to accurately (errors less than 1e-18) predict any number within the domain, both A->B and B->A. So far I have trained neural nets (via keras/tensorflow) and that works to a point. I am now wondering if I should be considering another approach altogether, like scikit-learn, or-tools, or something else. This might be a complex variant of kriging? I believe that this sort of problem is a regression rather than a classification, but maybe it's better modelled (and addressed) elsewhere? I guess that I'm asking what tools I should. really be considering for a complex black box function, given a large training set availability.

Apr 4, 2025 - 10:15

I have two datasets (A,B) that are the input and outputs of a complex set of functions. They are (or can be) contained within a poincaré disk. There are over two million samples, spread quite evenly over the problem domain. I want to use a fitting / prediction methodology to accurately (errors less than 1e-18) predict any number within the domain, both A->B and B->A.

So far I have trained neural nets (via keras/tensorflow) and that works to a point. I am now wondering if I should be considering another approach altogether, like scikit-learn, or-tools, or something else. This might be a complex variant of kriging?

I believe that this sort of problem is a regression rather than a classification, but maybe it's better modelled (and addressed) elsewhere?

I guess that I'm asking what tools I should. really be considering for a complex black box function, given a large training set availability.