def test_mini_lambda_table(self):
sp = SubstitutionProbability(lambda_table=get_table(), alpha= -5.)
o2 = Specie('O', -2)
s2 = Specie('S', -2)
li1 = Specie('Li', 1)
na1 = Specie('Na', 1)
self.assertAlmostEqual(sp.prob(s2, o2), 0.124342317272, 5
, "probability isn't correct")
self.assertAlmostEqual(sp.pair_corr(li1, na1), 1.65425296864, 5
, "correlation isn't correct")
prob = sp.cond_prob_list([o2, li1], [na1, li1])
self.assertAlmostEqual(prob, 0.00102673915742, 5
, "probability isn't correct")
def get_substitution_probability(self, el1, el2):
"""
Use pymatgen to calculate substitution probability for use in site mixing
:param el1: name of first element
:param el2: name of second element
:return: total probability of substitution between two elements
"""
sp = SubstitutionProbability()
total = 0
for o1 in el1.common_oxidation_states:
for o2 in el2.common_oxidation_states:
total += sp.prob(self.get_ionic_specie(el1, o1),
self.get_ionic_specie(el2, o2))
return total
def __init__(self, threshold=1e-3, symprec=0.1, **kwargs):
"""
This substitutor uses the substitution probability class to
find good substitutions for a given chemistry or structure.
Args:
threshold:
probability threshold for predictions
symprec:
symmetry precision to determine if two structures
are duplicates
kwargs:
kwargs for the SubstitutionProbability object
lambda_table, alpha
"""
self._kwargs = kwargs
self._sp = SubstitutionProbability(**kwargs)
self._threshold = threshold
self._symprec = symprec
def test_full_lambda_table(self):
"""
This test tests specific values in the data folder. If the
json is updated, these tests will have to be as well
"""
sp = SubstitutionProbability(alpha= -5.)
sp1 = Specie('Fe', 4)
sp3 = Specie('Mn', 3)
prob1 = sp.prob(sp1, sp3)
self.assertAlmostEqual(prob1, 1.69243954552e-05, 5
, "probability isn't correct")
sp2 = Specie('Pt', 4)
sp4 = Specie('Pd', 4)
prob2 = sp.prob(sp2, sp4)
self.assertAlmostEqual(prob2, 4.7174906021e-05, 5
, "probability isn't correct")
corr = sp.pair_corr(Specie("Cu", 2), Specie("Fe", 2))
self.assertAlmostEqual(corr, 6.82496631637, 5
, "probability isn't correct")
prob3 = sp.cond_prob_list([sp1, sp2], [sp3, sp4])
self.assertAlmostEqual(prob3, 0.000300298841302, 6
, "probability isn't correct")
class Substitutor(MSONable):
"""
This object uses a data mined ionic substitution approach to propose
compounds likely to be stable. It relies on an algorithm presented in
Hautier, G., Fischer, C., Ehrlacher, V., Jain, A., and Ceder, G. (2011).
Data Mined Ionic Substitutions for the Discovery of New Compounds.
Inorganic Chemistry, 50(2), 656-663. doi:10.1021/ic102031h
"""
def __init__(self, threshold=1e-3, symprec=0.1, **kwargs):
"""
This substitutor uses the substitution probability class to
find good substitutions for a given chemistry or structure.
Args:
threshold:
probability threshold for predictions
symprec:
symmetry precision to determine if two structures
are duplicates
kwargs:
kwargs for the SubstitutionProbability object
lambda_table, alpha
"""
self._kwargs = kwargs
self._sp = SubstitutionProbability(**kwargs)
self._threshold = threshold
self._symprec = symprec
def get_allowed_species(self):
"""
returns the species in the domain of the probability function
any other specie will not work
"""
return self._sp.species
def pred_from_structures(self, target_species, structures_list,
remove_duplicates=True):
"""
performs a structure prediction targeting compounds containing the
target_species and based on a list of structure (those structures
can for instance come from a database like the ICSD). It will return
all the structures formed by ionic substitutions with a probability
higher than the threshold
Args:
target_species:
a list of species with oxidation states
e.g., [Specie('Li',1),Specie('Ni',2), Specie('O',-2)]
structures_list:
a list of dictionnary of the form {'structure':Structure object
,'id':some id where it comes from}
the id can for instance refer to an ICSD id
Returns:
a list of TransformedStructure objects.
"""
result = []
transmuter = StandardTransmuter([])
if len(list(set(target_species) & set(self.get_allowed_species()))) \
!= len(target_species):
raise ValueError("the species in target_species are not allowed"
+ "for the probability model you are using")
for permut in itertools.permutations(target_species):
for s in structures_list:
#check if: species are in the domain,
#and the probability of subst. is above the threshold
els = s['structure'].composition.elements
if len(els) == len(permut) and \
len(list(set(els) & set(self.get_allowed_species()))) == \
len(els) and self._sp.cond_prob_list(permut, els) > \
self._threshold:
clean_subst = {els[i]: permut[i]
for i in xrange(0, len(els))
if els[i] != permut[i]}
if len(clean_subst) == 0:
continue
transf = SubstitutionTransformation(clean_subst)
if Substitutor._is_charge_balanced(
transf.apply_transformation(s['structure'])):
ts = TransformedStructure(
s['structure'], [transf], history=[s['id']],
other_parameters={
'type': 'structure_prediction',
'proba': self._sp.cond_prob_list(permut, els)}
)
result.append(ts)
transmuter.append_transformed_structures([ts])
if remove_duplicates:
transmuter.apply_filter(RemoveDuplicatesFilter(
symprec=self._symprec))
return transmuter.transformed_structures
@staticmethod
def _is_charge_balanced(struct):
"""
checks if the structure object is charge balanced
"""
if sum([s.specie. oxi_state for s in struct.sites]) == 0.0:
return True
else:
return False
def pred_from_list(self, species_list):
"""
There are an exceptionally large number of substitutions to
look at (260^n), where n is the number of species in the
list. We need a more efficient than brute force way of going
through these possibilities. The brute force method would be::
output = []
for p in itertools.product(self._sp.species_list
, repeat = len(species_list)):
if self._sp.conditional_probability_list(p, species_list)
> self._threshold:
output.append(dict(zip(species_list,p)))
return output
Instead of that we do a branch and bound.
Args:
species_list:
list of species in the starting structure
Returns:
list of dictionaries, each including a substitutions
dictionary, and a probability value
"""
#calculate the highest probabilities to help us stop the recursion
max_probabilities = []
for s2 in species_list:
max_p = 0
for s1 in self._sp.species:
max_p = max([self._sp.cond_prob(s1, s2), max_p])
max_probabilities.append(max_p)
output = []
def _recurse(output_prob, output_species):
best_case_prob = list(max_probabilities)
best_case_prob[:len(output_prob)] = output_prob
if reduce(mul, best_case_prob) > self._threshold:
if len(output_species) == len(species_list):
odict = {
'substitutions':
dict(zip(species_list, output_species)),
'probability': reduce(mul, best_case_prob)}
output.append(odict)
return
for sp in self._sp.species:
i = len(output_prob)
prob = self._sp.cond_prob(sp, species_list[i])
_recurse(output_prob + [prob], output_species + [sp])
_recurse([], [])
logging.info('{} substitutions found'.format(len(output)))
return output
def pred_from_comp(self, composition):
"""
Similar to pred_from_list except this method returns a list after
checking that compositions are charge balanced.
"""
output = []
predictions = self.pred_from_list(composition.elements)
for p in predictions:
subs = p['substitutions']
charge = 0
for i_el in composition.elements:
f_el = subs[i_el]
charge += f_el.oxi_state * composition[i_el]
if charge == 0:
output.append(p)
logging.info('{} charge balanced '
'compositions found'.format(len(output)))
return output
@property
def to_dict(self):
return {"name": self.__class__.__name__, "version": __version__,
"kwargs": self._kwargs, "threshold": self._threshold,
"@module": self.__class__.__module__,
"@class": self.__class__.__name__}
@classmethod
def from_dict(cls, d):
t = d['threshold']
kwargs = d['kwargs']
return cls(threshold=t, **kwargs)
