Issues:
- IMPORTANT DECISIONS
    - own class/function for min-max scaler
    - own class/function for transform (actually integrates into min-max) -> could we even flag this in case the local function gradients are too large/too small to catch 10^x or log(x) behavior? - or if the data is not distributed well in the domain (e.g. for exp. y from linear x) !! (possibly even resample the data for a better training?)
- API:
    - the function call can only handle lists but not other shapes of inputs (e.g. meshgrids, or one scalar and the other a vector)
    - should it be able to handle meshgrids? it can handle dataframes now!
    - training data is also not perfect. Should we force it to be a dataframe? Or explicitly not?
- unittests
- test the transform on generation with simple domain-selected variables!
- jax implementation (simple way to select between jax and pytorch?)
- write own min-max-scaler to have all the parameters available for custom loss functions?
- documentation
- add examples!!! (Sin function)