Fixed parameters

[yannikschaelte]

Implement fixed parameters, i.e. only a subset of the parameters is to be optimized. This can be handled in amici, but we should also enable this in pesto, probably in the problem, or the objective. One could have fields fixed_par_indices, fixed_par_values. The optimizer should not be aware of this.

[yannikschaelte]

question to all: say we have dim free parameters and dim_full >= dim parameters in total including fixed parameters.

in the return object, we would want to have all in size dim_full, right? then in grad and hess we would insert nans and in x insert the fixed values?

and we would also like lb and ub to be given in dim_full? this would make sense for the visualizations, but the bounds are not strictly required really. also, for user-defined parameter guesses, ones for the fixed parameters would not be required.

[yannikschaelte]

so what I would do would be: in problem fixed parameters and values are specified, and the objective is made capable of handling them. then internally the reduced vector is used, but in the result object everything is in dimension dim_full.

also, in problem.lb and .ub dim_full would be used, possibly extended with nan values.

[yannikschaelte]

is being worked on now in branch feature_fixedpars, see also here: https://github.com/ICB-DCM/pyPESTO/compare/feature_fixedpars?expand=1

[yannikschaelte]

done in pull request #37.