Aggregated Objective does not return expected outputs using `return_dict`

[arrjon]

Bug description The functions available in pypesto.visualize.model_fit cannot work with an AggregatedObjective and hence fail. There are two reasons:

• the amici model is not available by objective.amici_model (I already fixed that issue in the branch visualize_aggregated by adding a new property to the AggregatedObjective)
• calling pypesto_problem.objective(x, return_dict=True) does return empty rdatas. (this I could not fix right away)

Expected behavior The visuslization routines should still work properly. Also I would expect pypesto_problem.objective(x, return_dict=True) to return non empty rdatas. Calling pypesto_problem.objective._objectives[0](x, return_dict=True) instead works and does return non empty rdatas.

To reproduce I attached a minimal example using the Blasi model and PeTab. minimal example.zip

Environment

• Operating system: MacOs
• pypesto version: branch visualize_aggregated
• Python version: python 3.10

[arrjon]

maybe @dweindl can help here?

[dweindl]

maybe @dweindl can help here?

I am not too familiar with either AggregatedObjective or pypesto.visualize.model_fit, but I agree that it should either work, or raise an informative error.

the amici model is not available by objective.amici_model

That makes sense, because often AggregatedObjective won't have only a single model (https://github.com/ICB-DCM/pyPESTO/issues/439).

(I already fixed that issue in the branch visualize_aggregated by adding a new property to the AggregatedObjective)

I would rather handle AggregatedObjective separately in the visualization code instead of adding amici_model to AggregatedObjective, since there will probably be multiple models.

[arrjon]

I would rather handle AggregatedObjective separately in the visualization code instead of adding amici_model to AggregatedObjective, since there will probably be multiple models.

Okay, I thought that AggregatedObjective is usually an AmiciObjective and NegLogPriors and not multiple models. This makes it rather complicated to write a general visualization function.

[dweindl]

I don't know how it's currently used, but one of the goals would be using it with PEtab problems comprising multiple models (https://github.com/ICB-DCM/pyPESTO/issues/439) and I wouldn't add anything that will complicate this.

I think it's nevertheless possible to provide some meaningful visualizations. Just iterate over the objectives and visualize what's possible and give a warning if any of those cannot be handled.