[FEATURE REQUEST] SkipAtom compositional featurizer

[sgbaird]

https://github.com/lantunes/skipatom

Intended usage:

from pymatgen.core.composition import Composition
compositions = [Composition("Al2O3"), Composition("CeCoGe3")]

from matminer.featurizers.composition.composite import ElementProperty
ep = ElementProperty.from_preset("skipatom")
comp_fingerprints = ep.featurize_many(compositions)

I think the following would be a good example to follow:

matminer/matminer/utils/data.py

Lines 386 to 411 in 886524a

	class MatscholarElementData(AbstractData):
	"""
	Class to get word embedding vectors of elements. These word embeddings were
	generated using NLP + Neural Network techniques on more than 3 million
	scientific abstracts.
	The data returned by this class are simply learned representations of the
	elements, taken from:
	Tshitoyan, V., Dagdelen, J., Weston, L. et al. Unsupervised word embeddings
	capture latent knowledge from materials science literature. Nature 571,
	95–98 (2019). https://doi.org/10.1038/s41586-019-1335-8
	"""
	def __init__(self):
	dfile = os.path.join(module_dir, "data_files/matscholar_els.json")
	with open(dfile) as fp:
	embeddings = json.load(fp)
	self.prop_names = [f"embedding {i}" for i in range(1, 201)]
	all_element_data = {}
	for el, embedding in embeddings.items():
	all_element_data[el] = dict(zip(self.prop_names, embedding))
	self.all_element_data = all_element_data
	def get_elemental_property(self, elem, property_name):
	return self.all_element_data[str(elem)][property_name]

Followed by the appropriate incorporation into:

matminer/matminer/featurizers/composition/composite.py

Lines 18 to 244 in 886524a

	class ElementProperty(BaseFeaturizer):
	"""
	Class to calculate elemental property attributes.
	To initialize quickly, use the from_preset() method.
	Features: Based on the statistics of the data_source chosen, computed
	by element stoichiometry. The format generally is:
	"{data source} {statistic} {property}"
	For example:
	"PymetgenData range X" # Range of electronegativity from Pymatgen data
	For a list of all statistics, see the PropertyStats documentation; for a
	list of all attributes available for a given data_source, see the
	documentation for the data sources (e.g., PymatgenData, MagpieData,
	MatscholarElementData, etc.).
	Args:
	data_source (AbstractData or str): source from which to retrieve
	element property data (or use str for preset: "pymatgen",
	"magpie", or "deml")
	features (list of strings): List of elemental properties to use
	(these must be supported by data_source)
	stats (list of strings): a list of weighted statistics to compute to for each
	property (see PropertyStats for available stats)
	"""
	def __init__(self, data_source, features, stats):
	if data_source == "pymatgen":
	self.data_source = PymatgenData()
	elif data_source == "magpie":
	self.data_source = MagpieData()
	elif data_source == "deml":
	self.data_source = DemlData()
	elif data_source == "matscholar_el":
	self.data_source = MatscholarElementData()
	elif data_source == "megnet_el":
	self.data_source = MEGNetElementData()
	else:
	self.data_source = data_source
	self.features = features
	self.stats = stats
	# Initialize stats computer
	self.pstats = PropertyStats()
	@classmethod
	def from_preset(cls, preset_name):
	"""
	Return ElementProperty from a preset string
	Args:
	preset_name: (str) can be one of "magpie", "deml", "matminer",
	"matscholar_el", or "megnet_el".
	Returns:
	ElementProperty based on the preset name.
	"""
	if preset_name == "magpie":
	data_source = "magpie"
	features = [
	"Number",
	"MendeleevNumber",
	"AtomicWeight",
	"MeltingT",
	"Column",
	"Row",
	"CovalentRadius",
	"Electronegativity",
	"NsValence",
	"NpValence",
	"NdValence",
	"NfValence",
	"NValence",
	"NsUnfilled",
	"NpUnfilled",
	"NdUnfilled",
	"NfUnfilled",
	"NUnfilled",
	"GSvolume_pa",
	"GSbandgap",
	"GSmagmom",
	"SpaceGroupNumber",
	]
	stats = ["minimum", "maximum", "range", "mean", "avg_dev", "mode"]
	elif preset_name == "deml":
	data_source = "deml"
	stats = ["minimum", "maximum", "range", "mean", "std_dev"]
	features = [
	"atom_num",
	"atom_mass",
	"row_num",
	"col_num",
	"atom_radius",
	"molar_vol",
	"heat_fusion",
	"melting_point",
	"boiling_point",
	"heat_cap",
	"first_ioniz",
	"electronegativity",
	"electric_pol",
	"GGAU_Etot",
	"mus_fere",
	"FERE correction",
	]
	elif preset_name == "matminer":
	data_source = "pymatgen"
	stats = ["minimum", "maximum", "range", "mean", "std_dev"]
	features = [
	"X",
	"row",
	"group",
	"block",
	"atomic_mass",
	"atomic_radius",
	"mendeleev_no",
	"electrical_resistivity",
	"velocity_of_sound",
	"thermal_conductivity",
	"melting_point",
	"bulk_modulus",
	"coefficient_of_linear_thermal_expansion",
	]
	elif preset_name == "matscholar_el":
	data_source = "matscholar_el"
	stats = ["minimum", "maximum", "range", "mean", "std_dev"]
	features = MatscholarElementData().prop_names
	elif preset_name == "megnet_el":
	data_source = "megnet_el"
	stats = ["minimum", "maximum", "range", "mean", "std_dev"]
	features = MEGNetElementData().prop_names
	else:
	raise ValueError("Invalid preset_name specified!")
	return cls(data_source, features, stats)
	def featurize(self, comp):
	"""
	Get elemental property attributes
	Args:
	comp: Pymatgen composition object
	Returns:
	all_attributes: Specified property statistics of features
	"""
	all_attributes = []
	# Get the element names and fractions
	elements, fractions = zip(*comp.element_composition.items())
	for attr in self.features:
	elem_data = [self.data_source.get_elemental_property(e, attr) for e in elements]
	for stat in self.stats:
	all_attributes.append(self.pstats.calc_stat(elem_data, stat, fractions))
	return all_attributes
	def feature_labels(self):
	labels = []
	for attr in self.features:
	src = self.data_source.__class__.__name__
	for stat in self.stats:
	labels.append(f"{src} {stat} {attr}")
	return labels
	def citations(self):
	if self.data_source.__class__.__name__ == "MagpieData":
	citation = [
	"@article{ward_agrawal_choudary_wolverton_2016, title={A general-purpose "
	"machine learning framework for predicting properties of inorganic materials}, "
	"volume={2}, DOI={10.1038/npjcompumats.2017.28}, number={1}, journal={npj "
	"Computational Materials}, author={Ward, Logan and Agrawal, Ankit and Choudhary, "
	"Alok and Wolverton, Christopher}, year={2016}}"
	]
	elif self.data_source.__class__.__name__ == "DemlData":
	citation = [
	"@article{deml_ohayre_wolverton_stevanovic_2016, title={Predicting density "
	"functional theory total energies and enthalpies of formation of metal-nonmetal "
	"compounds by linear regression}, volume={47}, DOI={10.1002/chin.201644254}, "
	"number={44}, journal={ChemInform}, author={Deml, Ann M. and Ohayre, Ryan and "
	"Wolverton, Chris and Stevanovic, Vladan}, year={2016}}"
	]
	elif self.data_source.__class__.__name__ == "PymatgenData":
	citation = [
	"@article{Ong2013, author = {Ong, Shyue Ping and Richards, William Davidson and Jain, Anubhav and Hautier, "
	"Geoffroy and Kocher, Michael and Cholia, Shreyas and Gunter, Dan and Chevrier, Vincent L. and Persson, "
	"Kristin A. and Ceder, Gerbrand}, doi = {10.1016/j.commatsci.2012.10.028}, issn = {09270256}, "
	"journal = {Computational Materials Science}, month = {feb}, pages = {314--319}, "
	"publisher = {Elsevier B.V.}, title = {{Python Materials Genomics (pymatgen): A robust, open-source python "
	"library for materials analysis}}, url = {http://linkinghub.elsevier.com/retrieve/pii/S0927025612006295}, "
	"volume = {68}, year = {2013} } "
	]
	elif self.data_source.__class__.__name__ == "MEGNetElementData":
	# TODO: Cite MEGNet publication (not preprint) once released!
	citation = [
	"@ARTICLE{2018arXiv181205055C,"
	"author = {{Chen}, Chi and {Ye}, Weike and {Zuo}, Yunxing and {Zheng}, Chen and {Ong}, Shyue Ping},"
	"title = '{Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals}',"
	"journal = {arXiv e-prints},"
	"keywords = {Condensed Matter - Materials Science, Physics - Computational Physics},"
	"year = '2018',"
	"month = 'Dec',"
	"eid = {arXiv:1812.05055},"
	"pages = {arXiv:1812.05055},"
	"archivePrefix = {arXiv},"
	"eprint = {1812.05055},"
	"primaryClass = {cond-mat.mtrl-sci},"
	r"adsurl = {https://ui.adsabs.harvard.edu/\#abs/2018arXiv181205055C},"
	"adsnote = {Provided by the SAO/NASA Astrophysics Data System}}"
	]
	else:
	citation = []
	return citation
	def implementors(self):
	return ["Jiming Chen", "Logan Ward", "Anubhav Jain", "Alex Dunn"]