Add plot_structure_2d() in new module ml_matrics/struct_vis.py

[janosh]

Adds ml_matrics/struct_vis.py with ase-inspired plot_structure_2d() but works natively with pymatgen Structures, matplotlib being the only other dependency.

Example

from pymatgen.ext.matproj import MPRester
from ml_matrics.struct_vis import plot_structure_2d

struct = MPRester().get_structure_by_material_id('mp-12712')
plot_structure_2d(struct)

Aims to resemble output of ase.visualize.plot.plot_atoms with minor improvements, less conversion hassle and handling of disordered structures.

pymatgen is unable to convert disordered structures via AseAtomsAdaptor().get_atoms(struct):

ValueError: ASE Atoms only supports ordered structures

which appears to be untrue. ASE does allow for disorder.

ASE Example

from pymatgen.ext.matproj import MPRester
from pymatgen.io.ase import AseAtomsAdaptor
from ase.visualize.plot import plot_atoms

struct = MPRester().get_structure_by_material_id('mp-12712')
atoms = AseAtomsAdaptor().get_atoms(struct)

plot_atoms(atoms)

Disordered Example

import matplotlib.pyplot as plt
from pymatgen.transformations.standard_transformations import SubstitutionTransformation


struct = MPRester().get_structure_by_material_id("mp-19017")

disord_struct: Structure = SubstitutionTransformation(
    {"Fe": {"Fe": 0.75, "C": 0.25}}
).apply_transformation(struct)

plot_structure_2d(disord_struct, rotation="10x,10y,0z", atomic_radii=0.5)

[janosh]

@sgbaird @CompRhys Would be interested to hear your feedback if you encounter the need to get a quick 2d plot of a pymatgen structure and feel like giving this a try.

[sgbaird]

Great! Will give it a try

[janosh]

Structure with new annotate_sites=True option:

Room for improvement: element labels aren't yet occluded by sites with higher z-index.

[janosh]

Output seems suboptimal but maybe good enough for now across most of 20 random structures. I think the code can still be simplified quite a bit though so leaving WIP status.

structs = [
    (fname, Structure.from_file(fname))
    for fname in glob(f"{ROOT}/data/structures/*.yml")
]

fig, axs = plt.subplots(4, 5, figsize=(20, 20))

for (fname, struct), ax in zip(structs, axs.flat):
    ax = plot_structure_2d(struct, ax=ax)
    mp_id = os.path.basename(fname).split(".")[0]
    ax.set_title(f"{struct.formula}\n{mp_id}", fontsize=14)


fig.savefig(f"{ROOT}/assets/mp-structures-2d.svg", bbox_inches="tight")

[CompRhys]

Maybe a key off to the side? So then it doesn't overlayer the labels?

Does it give different colours to charge assigned elements?

[janosh]

Maybe a key off to the side? So then it doesn't overlayer the labels?

Could you be more specific?

Does it give different colours to charge assigned elements?

It doesn't. Do you think that's an important feature? I feel like there already aren't enough colors to differentiate elements clearly. Maybe better to include oxidation state in site labels?

[sgbaird]

@janosh if you decide to include it, I think as a site label might work. As a legend (e.g. blue: Ni, red: O) incorporating charge states is tough because of the color issue.