parameters in formula interface

[kkmann]

Hi,

I was just wondering whether someone considered adding support for parameterized functions to the formula interface. With the excellent automatic differentiation support in julia it should be possible to incorporate that in a genric fitting algorithm, right?

So, for instance, you could write something like

@formula(y ~ 1 + x + f(x, :alpha)

which would indicate that :alpha is a learnable parameter and needs to be fitted as well. Are there any technical/logical reasons why that shouldn't work?