Results from *.get() method differ from simulation result object

[LukasKoenemann]

Hi,
I'm using pyfmi to simulate my modelica models in python. First I'm calling the simulation in my declared class (result = selfmodel.simulate()). Later on I fetch my variables via the .get() method (value = model.get(variable)). Now I'm wondering, that the values for the get() sometimes differ from the values of the result- object:
code:

            result = self.model.simulate(
                input=input_object,
                start_time=start_time,
                final_time=stop_time,
                options=opts,
            )
            for variable in ['TAC', 'electrical_storage[1].F']:
                print(f'\t - result[{variable}]:', result[variable][-1])
                print(f'\t - model.final({variable})', result.final(variable))
                print(f'\t - get_variable({variable}):', self.get_variable(variable))

output:

         - result[TAC]: 123360.32496377944
         - model.final(TAC) 123360.32496377944
         - get_variable(TAC): 123857.68462234597
         - result[electrical_storage[1].F]: 0.9997065955338098
         - model.final(electrical_storage[1].F) 0.9997065955338098
         - get_variable(electrical_storage[1].F): 0.9997065955338098

Do you know where the difference come from? And which value is the "true" value?

• this is how my get_variable() method looks like:

    def get_variable(self, name) -> float:
        return self.model.get(name).item()

[PeterMeisrimelModelon]

I fail to reproduce the issue thus far. Can you provide us with a minimal reproducer?

[LukasKoenemann]

here is a jupyter notebook script, the fmu und the necessary resource data to reproduce the error.
minimal_example.zip

[PeterMeisrimelModelon]

here is a jupyter notebook script, the fmu und the necessary resource data to reproduce the error. minimal_example.zip

Thanks, can reproduce the issue, looking into it.

[PeterMeisrimelModelon]

Hi,

we suspect there is some issue in the result handling, possibly related to enumerations, we continue to investigate.

For the time being, we recommend to use result_handling = 'binary' & put a filter on the specific variables you want to look at (note: These are using glob patterns; and variables such as 'x[1]' will not be filtered correctly, use 'x?1?' instead). This should address the differences in the results with different result handlers (minus potential machine precision sized differences).

For the difference in res = fmu.simulate(...) vs. fmu.get(...) (var 'TAC') we suspect this may actually be due to the FMU. Is the evaluation of 'TAC' possibly linked to impure (external) functions or may be affected by specific fmu.get(...) calling sequences?