Add support for serializing ase Atoms objects

[utf]

This PR adds support for serializing ase Atoms objects.

We've found this might be a useful feature to have in atomate2. But I think it could be useful more broadly. E.g., Atoms objects are used in the MD interface in matgl, in chgnet, and quacc (developed by @Andrew-S-Rosen). Although I understand if this sort of feature is outside the scope of monty.

[shyuep]

I agree this is useful. But shouldn't this be implemented in ASE itself? Eg pymatgen implements Structure serialization. Monty is not exactly a materials science code. It is strange to have ASE serialization here.

[codecov]

Codecov Report

Attention: 4 lines in your changes are missing coverage. Please review.

Comparison is base (3d6ba72) 75.56% compared to head (52a516f) 75.59%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #582      +/-   ##
==========================================
+ Coverage   75.56%   75.59%   +0.03%     
==========================================
  Files          27       27              
  Lines        1465     1475      +10     
  Branches      309      312       +3     
==========================================
+ Hits         1107     1115       +8     
- Misses        302      304       +2     
  Partials       56       56              

Files	Coverage Δ
monty/json.py	85.13% <85.71%> (-0.16%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[utf]

I agree that this feels quite specific for monty. I'll raise an issue on the ase repo and see if they would be happy with me adding monty support directly.

[Andrew-S-Rosen]

ASE does have its own .todict() and .fromdict() methods. That said, it doesn't currently work if there are constraints if I recall correctly. So, I usually have something like the following in my code that uses the ASE JSON encoder/decoder instead, although I believe the encoded object is a str type, which isn't necessarily ideal.

from ase.io.jsonio import decode, encode

def atoms_as_dict(s: Atoms) -> dict:
    return {"@module": "ase.atoms", "@class": "Atoms", "atoms_json": encode(s)}

def atoms_from_dict(d: dict) -> Atoms:
    return decode(d["atoms_json"])

Atoms.as_dict = atoms_as_dict
Atoms.from_dict = atoms_from_dict

Not sure of the pros/cons compared to your approach @utf. I'd need to investigate. But I just wanted to share what I already know about the topic.

[Andrew-S-Rosen]

I agree that this feels quite specific for monty. I'll raise an issue on the ase repo and see if they would be happy with me adding monty support directly.

I'm a maintainer of ASE now, so feel free to float something by me. I might be able to help get it through. My gitlab username is andrewrosen.

[shyuep]

I would be fine adding support for todict and fromdict as alternatives to as_dict and from_dict. Though it would be better if ASE simply adds these aliases.

[shyuep]

Closing this for now.

[Andrew-S-Rosen]

I know this is quite old, but I wanted to chime in to say that there is basically 0% chance such methods will be able to be added to ASE. It either has to happen in monty or monkeypatched in each code that wants to make Atoms an MSONable.

I would be fine adding support for todict and fromdict as alternatives to as_dict and from_dict.

Did we want to revisit this proposition? It is not necessarily ideal since monty is a general-purpose package, but it is indeed practical.

I will note that the method proposed by @utf is not ideal. The below will crash because FixAtoms is not JSON serializable.

from ase.build import bulk
from ase.constraints import FixAtoms
from monty.json import MontyEncoder

atoms = bulk("Cu")
c = FixAtoms([0])
atoms.set_constraint(c)
{"@module": "ase", "@class": "Atoms", "data": MontyEncoder().encode(atoms.todict())}

My approach using ASE's JSON encoder/decoder will always work, but it has the downside that the Atoms object data is stored as a JSON string, which is not readily parseable via a database query. You'd have to rehydrate it with an Atoms.from_dict() call before use.