el_refs serialized as strings

[jmmshn]

emmet/emmet-core/emmet/core/thermo.py

Lines 251 to 253 in 31a2262

	phase_diagram: PhaseDiagram = Field(
	..., description="Phase diagram for the chemical system.",
	)

When these PhaseDiagramDocuments are pulled from the DB and deserialized the keys for el_refs are string but the pymatgen code uses Element the entries time. I can't find if a conversion occurred in emmet.

pd = loadfn("/Users/shen9/Desktop/pd_doc.json") # taken from DB
pd["phase_diagram"].el_refs.keys()
# gives dict_keys(['O', 'Ge', 'Co', 'Li'])

@kim-jiyoon reported this. @munrojm and @mkhorton can you please have a look? This is to get the conversion documents on the new site.

[munrojm]

Yup! Will take a look at this.

[munrojm]

@janosh, this seems to stem from the fact that computed_data can be passed when constructing a PhaseDiagram object, which contains the el_refs dictionary. If its keys are strings instead of Element objects it appears to make a fuss when de-serializing with monty, or when trying to plot with PDPlotter. It might be best to have it check and ensure Element object keys on construction in pymatgen?

[munrojm]

Also, qhull_data is assumed to be a numpy array, and causes similar issues for ternary+ plots. For now, I am manually patching in the API client. https://github.com/materialsproject/api/blob/ece1b3effeebae208a16d98776a3bb555ec8e459/mp_api/client/routes/thermo.py#L181

[janosh]

Was I tagged by accident?

[munrojm]

@janosh No, this just relates to a pymatgen issue I was about to open. We can discuss outside of this thread.