Providing each governing equation its own Candidate Function Library

[mszymanski-tuhh]

Hello,
I am trying to figure out the underlying equations of a synthetic model, where I already know some parts of some governing equations (if not the entire algebraic representation of some equations), or even whether some inputs will appear within an equation or not. While I tried to customize and tensor multiple libraries, where I would include or exclude certain inputs, it didn't really work out. Basically, I want to steer the algorithm in such a way where I can provide each governing equation its own candidate functions library, example:

I have a system of differential equations consisting of three states x0, x1, x2 that I want to represent given multiple timeseries of those states. I already know that the diff. eq. for x2 is dot_x2 = x1. For the diff. eq. dot_x2, I develop a library which only accepts an IdentityLibrary with inputs x1

While this may appear redundant to define, it may be helpful in case parts of some governing equations are already known while, for example, trying to model and isolate some underlying "noise" or un-modelled dynamics. Has anyone ever done something like this before?

Edit: I can see that one can linearly constraint certain features to select targets using the ConstrainedSR3 method, but is there a way to include/exclude features for a target using the index logic?

Edit 2: I actually found a way to do it: one simply has to know how the indexing works for cvxpy!

Thanks!
Michael