Can I use the optimade JSON serialization to store ASE atoms?

[jan-janssen]

For my structures to be stored in an interoperable format - in the context of FAIR - I would like to use the JSON representation of the optimade structure.

I naively tired the following:

import json
from ase.build import bulk
from optimade.adapters.structures.ase import from_ase_atoms, get_ase_atoms
from optimade.models.structures import StructureResourceAttributes

al = bulk("Al", cubic=True)
struct_opt = from_ase_atoms(atoms=al)
json_dict = struct_opt.model_dump(mode="json")
structure_restore = StructureResourceAttributes.model_validate_json(json.dumps(json_dict))
get_ase_atoms(optimade_structure=structure_restore)

Unfortunately, this fails with the following error message:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[8], line 1
----> 1 get_ase_atoms(optimade_structure=structure_restore)

File ~/mambaforge/lib/python3.12/site-packages/optimade/adapters/structures/ase.py:55, in get_ase_atoms(optimade_structure)
     52     warn(ASE_NOT_FOUND, AdapterPackageNotFound)
     53     return None
---> 55 attributes = optimade_structure.attributes
     57 # Cannot handle partial occupancies
     58 if StructureFeatures.DISORDER in attributes.structure_features:

File ~/mambaforge/lib/python3.12/site-packages/pydantic/main.py:891, in BaseModel.__getattr__(self, item)
    888     return super().__getattribute__(item)  # Raises AttributeError if appropriate
    889 else:
    890     # this is the current error
--> 891     raise AttributeError(f'{type(self).__name__!r} object has no attribute {item!r}')

AttributeError: 'StructureResourceAttributes' object has no attribute 'attributes'

[edan-bainglass]

@jan-janssen try using the StructureResource class instead of StructureResourceAttributes

[jan-janssen]

That fails even one step before:

---------------------------------------------------------------------------
ValidationError                           Traceback (most recent call last)
Cell In[7], line 1
----> 1 structure_restore = StructureResource.model_validate_json(json.dumps(json_dict))

File ~/mambaforge/lib/python3.12/site-packages/pydantic/main.py:656, in BaseModel.model_validate_json(cls, json_data, strict, context)
    654 # `__tracebackhide__` tells pytest and some other tools to omit this function from tracebacks
    655 __tracebackhide__ = True
--> 656 return cls.__pydantic_validator__.validate_json(json_data, strict=strict, context=context)

ValidationError: 2 validation errors for StructureResource

[ml-evs]

It's only the restoration step that is failing in your first example, perhaps a bit confusing but from_ase_atoms gives you StructureResourceAttributes but get_ase_atoms expects a full StructureResource (i.e., with an ID, as you would get directly from OPTIMADE). Happy to make this detection automatic if that would be helpful. Something like:

import json
from ase.build import bulk
from optimade.adapters.structures.ase import from_ase_atoms, get_ase_atoms
from optimade.models.structures import StructureResourceAttributes, StructureResource

al = bulk("Al", cubic=True)
struct_opt = from_ase_atoms(atoms=al)
json_dict = struct_opt.model_dump(mode="json")
structure_restore = StructureResourceAttributes.model_validate_json(json.dumps(json_dict))
get_ase_atoms(optimade_structure=StructureResource(id="dummy", attributes=structure_restore))

will work even if its a bit cumbersome.

[jan-janssen]

@ml-evs Thanks a lot, now it works.

[blokhin]

@ml-evs however it won’t work for the disordered structures or for any other tags attached to the particular atoms, correct?

[ml-evs]

@ml-evs however it won’t work for the disordered structures or for any other tags attached to the particular atoms, correct?

AFAIK ASE itself doesn't support partial occupancies (it can read them from a CIF but does not handle them well when converted to Atoms). The adapter code here should not be used for anything load-bearing, ideally ASE etc. would instead support OPTIMADE so that it can be removed from here (I should remove the pymatgen support at some point, since we already added it in materialsproject/pymatgen#3876, likewise AiiDA which exists here: https://github.com/aiidateam/aiida-optimade). As ASE rarely updates on PyPI, its a low priority for me to remove it but any simplification of our deps improves the sustainability of this package given limited (and probably soon vanishing) time for maintenance.