observables but not species supported in `noiseFormula`

[paulflang]

The documentations states that the noiseFormula in the observable table can be specified like so:
noiseParameter1_observable_pErk + noiseParameter2_observable_pErk*pErk

However, when I use a species in the observable formula, I get the following error:

TypeError                                 Traceback (most recent call last)
/media/sf_DPhil_Project/Project07_Parameter Fitting/PEtab/petab/calculate.py in evaluate_noise_formula(measurement, noise_formulas, parameter_df, simulation)
    187     try:
--> 188         noise_value = float(noise_value)
    189     except TypeError:

~/venvs/std/lib/python3.8/site-packages/sympy/core/expr.py in __float__(self)
    324             raise TypeError("can't convert complex to float")
--> 325         raise TypeError("can't convert expression to float")
    326 

TypeError: can't convert expression to float

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-11-8dc41b43303c> in <module>
     34 problem = core.DisFitProblem(PETAB_YAML)
     35 problem.write_jl_file()
---> 36 problem.optimize()
     37 problem.plot_results('c0', path='plot.pdf')
     38 problem.write_results()

/media/sf_DPhil_Project/Project07_Parameter Fitting/df_software/DisFit/DisFit/core.py in optimize(self)
    386         print(self.petab_problem.parameter_df)
    387 
--> 388         self._results['fval'] = -petab.calculate_llh(self.petab_problem.measurement_df.loc[:, cols],
    389             pd.concat([self.petab_problem.simulation_df.rename(columns={'measurement': 'simulation'}), ndf], axis=1),
    390             self.petab_problem.observable_df,

/media/sf_DPhil_Project/Project07_Parameter Fitting/PEtab/petab/calculate.py in calculate_llh(measurement_dfs, simulation_dfs, observable_dfs, parameter_dfs)
    272     for (measurement_df, simulation_df, observable_df, parameter_df) in zip(
    273             measurement_dfs, simulation_dfs, observable_dfs, parameter_dfs):
--> 274         _llh = calculate_llh_for_table(
    275             measurement_df, simulation_df, observable_df, parameter_df)
    276         llhs.append(_llh)

/media/sf_DPhil_Project/Project07_Parameter Fitting/PEtab/petab/calculate.py in calculate_llh_for_table(measurement_df, simulation_df, observable_df, parameter_df)
    314 
    315         # get noise standard deviation
--> 316         noise_value = evaluate_noise_formula(
    317             row, noise_formulas, parameter_df, petab.scale(simulation, scale))
    318 

/media/sf_DPhil_Project/Project07_Parameter Fitting/PEtab/petab/calculate.py in evaluate_noise_formula(measurement, noise_formulas, parameter_df, simulation)
    188         noise_value = float(noise_value)
    189     except TypeError:
--> 190         raise TypeError(
    191             f"Cannot replace all parameters in noise formula {noise_value} "
    192             f"for observable {observable_id}.")

TypeError: Cannot replace all parameters in noise formula 0.1*A + 0.5 for observable obs_a.

When I replace the species with an observable everything works fine.

[yannikschaelte]

Thanks for pointing this out. To me it seems like there is still an error in the example formula in https://petab.readthedocs.io/en/stable/documentation_data_format.html#id2 (@dweindl could you shortly confirm?). The noise formula should only support observable, but not species dependence, as the latter can be realized via the former via additional observables, if needed. This seems not documented properly.

[dweindl]

The noise formula should only support observable, but not species dependence, as the latter can be realized via the former via additional observables, if needed. This seems not documented properly.

Not sure about that. I thought we agreed, that for PEtab this could be anything that would be allowed in other math places in the SBML model. (For AMICI we kind of agreed that we want to support only parameters and observables there.) Either that needs to be revised, or petab.calculate.evaluate_noise_formula should mention that it can't handle that.

[yannikschaelte]

Ah right, that may also be. For me, both is fine. As the documentation is not specific in this regard, it would be fine for me to keep allowing all types of expressions, but add a warning to calculate.py. At some point, we should extend the test suite then.

@paulflang For your info: calculate.py was designed primarily to automatically compute objective functions for the petab test suite, not targeted to provide the capability of computing function values in every day use. I am not sure whether other scenarios are also not covered, but I would not outrule that.

[paulflang]

@paulflang For your info: calculate.py was designed primarily to automatically compute objective functions for the petab test suite, not targeted to provide the capability of computing function values in every day use. I am not sure whether other scenarios are also not covered, but I would not outrule that.

Thanks. This is good to know.