While neural networks and other machine learning algorithms are very good at finding accurate approximations of the underlying function relating a given input to an output, it is in general hard to interpret what that mapping is doing.

​

Genetic programming is technique to find analytic expressions that approximate the underlying function.

​

We have applied this technique to find analytic expressions that predict the star-formation rate density (SFRD) just a function of redshift and the value of the cosmological and astrophysical parameters. The below formulae can be used with the IllustrisTNG simulations.