EIM for part of a LincombOperator

[niklasreich]

I have a full order model fom with one operator that is a LincombOperator, which itself includes multiple operators in its attribute operators. I want to apply the EIM only to the one operator within my linear combination that is nonlinear.
To do this I am trying to use the function interpolate_operators(). This function already has a parameter operator_names but I am struggling to pass only part of my linear combination (so something like ["operator.operators[2]"] if fom.operator.operators[2] is the desired operator) to the function, because the getattr() function within interpolate_operators() says this is not an attribute of my fom anymore.
How can I solve this?

[sdrave]

Yes, interpolate_operators is not smart enough to only treat parts of operators. I'm not sure if it's worth the effort to add that feature. If you strip the features that you don't need, interpolate_operators reduces to like 10 lines of code. So I would recommend to copy-paste the relevant parts and adapt them to handle the desired operator.

Something like:

op = fom.operator.operators[2]
evaluations = op.range.empty()
for mu in parameter_sample:
    U = fom.solve(mu)
    evaluations.append(op.apply(U, mu=mu)
dofs, basis, data = ei_greedy(evaluations, copy=False, **further_options)
ei_op = EmpiricalInterpolatedOperator(op, dofs, basis, triangular=True)  #False for DEIM
new_ops = [ei_op if i == 2 else op for i, op in enumerate(fom.operator.operators)]
ei_fom = fom.with_(operator=fom.operator.with_(operators=new_ops))

[niklasreich]

Thanks for the insight, I will do that. I just wanted to make sure that this isn't a feature that is already included and I am just not calling the function in the correct way.