AutodE uses subclasses of ndarray (such as autode.values.Coordinates), with extra attributes such as units, other than the array. However, the attributes are not passed through when pickling. This means that the units etc. are lost when any of those objects are returned from parallel processes (as multiprocessing Queues use pickle to serialize from workers).
Minimal example:
>>> import autode as ade
>>> mol = ade.Molecule(smiles='CCO')
>>> x = mol.coordinates
>>> x
Coordinates([[ 0.8737 0.1872 0.0478]
[-0.4674 -0.481 -0.0469]
[-1.4902 0.3552 -0.417 ]
[ 0.797 1.3 0.0437]
[ 1.3353 -0.1674 1.0122]
[ 1.5848 -0.1419 -0.747 ]
[-0.7607 -0.9924 0.9124]
[-0.3963 -1.3063 -0.793 ]
[-1.4762 1.2465 -0.012 ]] Å)
>>> type(x)
<class 'autode.values.Coordinates'>
>>> pickle.loads(pickle.dumps(x,pickle.HIGHEST_PROTOCOL))
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/cha17srm/miniconda3/envs/main/lib/python3.10/site-packages/autode/values.py", line 645, in __repr__
return f'Coordinates({np.ndarray.__str__(self)} {self.units.name})'
AttributeError: 'Coordinates' object has no attribute 'units'
OS: Ubuntu 22.0.4. Also shows same error on Windows 10.
shoubhikraj
commented
1 year ago
AutodE uses subclasses of ndarray (such as
autode.values.Coordinates), with extra attributes such as units, other than the array. However, the attributes are not passed through when pickling. This means that the units etc. are lost when any of those objects are returned from parallel processes (as multiprocessing Queues use pickle to serialize from workers).Minimal example:
OS: Ubuntu 22.0.4. Also shows same error on Windows 10.