3 ms·

You don’t even need a training set imo just enough permutations to get a distribution of putative functions and outputs.