How to set fill_fixed_parameters=True for petab parameter mappings?

[FFroehlich]

So there is

pyPESTO/pypesto/petab/importer.py

Line 476 in ead29b3

fill_fixed_parameters=False,

, which I don't like as this means that edata.plist will always be all parameters instead of subsetting to the ones that actually require sensitivities.

Now the best way that I found to fix this is

pypesto_problem = importer.create_problem()
simulation_conditions = petab.get_simulation_conditions(
    importer.petab_problem.measurement_df
)
pypesto_problem.objective.parameter_mapping = amici.petab.parameter_mapping.create_parameter_mapping(
    petab_problem=importer.petab_problem,
    simulation_conditions=simulation_conditions,
    scaled_parameters=True,
    amici_model=pypesto_problem.objective.amici_model,
    fill_fixed_parameters=True,
)

which is not great, because it requires me to have pretty deep knowledge of the code and requires a couple more lines of code than what I had hoped for.

Is there any specific reason why we are setting fill_fixed_parameters=True? If there is could we either expose this option to the user or set smarter defaults when this is really necessary?

Looks like this was introduced in #912, but seems unrelated to the linked issue #882. Unclear what "fix warning from fixed overrides" refers to.

[dweindl]

Sorry for the late reply.

Looks like this was introduced in #912, but seems unrelated to the linked issue #882.

The linked issue is unrelated to fill_fixed_parameters, correct. It was linked because that other change in this PR (#882 (comment)).

Unclear what "fix warning from fixed overrides" refers to.

I don't recall that either. My guess would be: Before the change, pypesto was calling the amici/pypesto functions for filling parameter values according to the parameter mapping. The parameter mapping already set the values for the fixed parameters (fill_fixed_parameters=True). Thus, their symbols didn't show up in the parameter mapping anymore. Yet, pypesto still provided values for those parameters which triggered a warning that values were passed for unknown parameters. The PR was to prevent this warnings.

For anything based on the PEtab importer, changing fill_fixed_parameters should not be necessary.

this means that edata.plist will always be all parameters instead of subsetting to the ones that actually require sensitivities

This is very suprising. I am pretty sure this was working at some point and that there were tests in place. Needs further investigation.

[dweindl]

this means that edata.plist will always be all parameters instead of subsetting to the ones that actually require sensitivities

This is very suprising. I am pretty sure this was working at some point and that there were tests in place. Needs further investigation.

So my current understanding is:

• this was handled in pypesto until fb82392#diff-9d078ac1d9ce34d78fbddabf5615006e38f17881dd528debc4f61e118e0a9c17R477 , then Re-use plist in AmiciObjective #226 was opened
• Re-use plist in AmiciObjective #226 was supposed to be fixed by account for plist in edata when slicing amici results #621 and construct condition specific plist in parameter mapping AMICI-dev/AMICI#1487. However, the "fixed" create_plist_from_par_opt_to_par_sim wasn't used anywhere in pypesto, and the respective test didn't test what it was supposed to test. create_plist_from_par_opt_to_par_sim is still lying around here unused.

[dweindl]

After a closer look: 1) plist is set correctly/efficiently if there are no pypesto-fixed parameters. I.e. condition-specific requirements are properly handled. 2) However, if there are pypesto-fixed parameters, they are not excluded from plist. For the latter I can't tell if this ever worked correctly or not.

I will try to get (2) fixed and will submit an effective test case shortly.

[dweindl]

2. However, if there are pypesto-fixed parameters, they are not excluded from plist. For the latter I can't tell if this ever worked correctly or not.

I am pretty confident this never worked.