Should unspecified optional strings be the empty string or NaN?

At the moment, there can be NaNs (after pd.read_csv) in optional PEtab string columns, such as observableNames, that, if interpreted as a string, are converted to the string literal 'nan'.

>>> import numpy as np
>>> str(np.nan)
'nan'

An issue can occur in the AMICI plotting functions. This issue can be fixed by replacing

elif model.getObservableNames()[iy] != '':

with

elif model.getObservableNames()[iy] in ['', 'nan']:

to correctly identify unspecified observable names. However, testing for the string 'nan' seems unintuitive, and this fix might cause another issue if an observable is named 'nan'.

Here's a solution, which could be implemented in PEtab, and might resolve the issue in AMICI.

$ cat test_str.csv
observableId    observableName
a_id    a_name
b_id    

>>> import pandas as pd
>>> df1 = pd.read_csv('test_str.csv', sep='\t')
>>> df2 = pd.read_csv('test_str.csv', sep='\t')
>>> df2['observableName'] = df2['observableName'].fillna('')
>>> df1
  observableId observableName
0         a_id         a_name
1         b_id            NaN
>>> df2
  observableId observableName
0         a_id         a_name
1         b_id               

Start by reviewing the PEtab optional string-column handling described in the issue and the affected entry point in AMICI's python/amici/plotting.py at line 81. Decide how unspecified values should be represented without confusing a literal 'nan' name, then verify the behavior across parsing and plotting integration.