def __init__(self, exclude=None):
super(StructureFeaturizers, self).__init__(exclude=exclude)
self._fast_featurizers = [
sf.DensityFeatures(),
sf.GlobalSymmetryFeatures(),
sf.EwaldEnergy(),
sf.SineCoulombMatrix(flatten=True)
]
ssf = sf.SiteStatsFingerprint
self._slow_featurizers = [
ssf.from_preset('CrystalNNFingerprint_ops'),
ssf.from_preset("BondLength-dejong2016"),
ssf.from_preset("BondAngle-dejong2016"),
ssf.from_preset("Composition-dejong2016_SD"),
ssf.from_preset("Composition-dejong2016_AD"),
ssf.from_preset("CoordinationNumber_ward-prb-2017"),
ssf.from_preset("LocalPropertyDifference_ward-prb-2017"),
sf.ChemicalOrdering(),
sf.StructuralHeterogeneity(),
sf.MaximumPackingEfficiency(),
sf.XRDPowderPattern(),
sf.Dimensionality(),
sf.OrbitalFieldMatrix(flatten=True),
sf.JarvisCFID(),
]
# Prevent import errors
self._require_external = []
if torch and cgcnn:
self._require_external.append(sf.CGCNNFeaturizer())
if dscribe:
self._require_external.append(sf.SOAP())
self._need_fitting_featurizers = [
sf.PartialRadialDistributionFunction(),
sf.BondFractions(),
sf.BagofBonds(coulomb_matrix=sf.CoulombMatrix()),
sf.BagofBonds(coulomb_matrix=sf.SineCoulombMatrix()),
]
self._matrix_featurizers = [
sf.CoulombMatrix(flatten=False),
sf.RadialDistributionFunction(), # returns dict
sf.MinimumRelativeDistances(), # returns a list
sf.ElectronicRadialDistributionFunction()
]
# these are the same as _need_fitting_featurizers
self._many_features_featurizers = [
sf.PartialRadialDistributionFunction(),
sf.BondFractions(approx_bonds=False),
sf.BagofBonds(coulomb_matrix=sf.CoulombMatrix()),
sf.BagofBonds(coulomb_matrix=sf.SineCoulombMatrix()),
sf.OrbitalFieldMatrix(flatten=True),
sf.JarvisCFID()
]
def __init__(self, exclude=None):
super(StructureFeaturizers, self).__init__(exclude=exclude)
self._fast_featurizers = [
sf.DensityFeatures(),
sf.GlobalSymmetryFeatures(),
sf.EwaldEnergy()
]
self._slow_featurizers = [
sf.SiteStatsFingerprint.from_preset('CrystalNNFingerprint_ops'),
sf.ChemicalOrdering(),
sf.StructuralHeterogeneity(),
sf.MaximumPackingEfficiency(),
sf.XRDPowderPattern(),
sf.Dimensionality(),
sf.OrbitalFieldMatrix(flatten=True)
]
self._need_fitting_featurizers = [
sf.PartialRadialDistributionFunction(),
sf.BondFractions(),
sf.BagofBonds()
]
self._matrix_featurizers = [
sf.RadialDistributionFunction(), # returns dict
sf.CoulombMatrix(), # returns a matrix
sf.SineCoulombMatrix(), # returns a matrix
sf.OrbitalFieldMatrix(flatten=False), # returns a matrix
sf.MinimumRelativeDistances(), # returns a list
sf.ElectronicRadialDistributionFunction()
]
# these are the same as _need_fitting_featurizers
self._many_features_featurizers = [
sf.PartialRadialDistributionFunction(),
sf.BondFractions(),
sf.BagofBonds(),
sf.OrbitalFieldMatrix(flatten=True)
]
def heavy(self):
fs = [
self.ssf.from_preset('CrystalNNFingerprint_ops'),
sf.ChemicalOrdering(),
sf.StructuralHeterogeneity(),
sf.MaximumPackingEfficiency(),
sf.XRDPowderPattern(),
sf.Dimensionality(),
sf.OrbitalFieldMatrix(flatten=True),
sf.JarvisCFID(),
]
fs += self.express
fs = self._add_external(fs)
return self._get_featurizers(fs)
mae += np.mean(np.abs(Y_pred - Y_test) / N_test) / NUM_SPLITS
rmse += np.mean(((Y_pred - Y_test) / N_test)**2)**0.5 / NUM_SPLITS
r2 += sklearn.metrics.r2_score(Y_test / N_test,
Y_pred / N_test) / NUM_SPLITS
print("SCM RESULTS MAE = %f, RMSE = %f, R-SQUARED = %f" % (mae, rmse, r2))
finish = time.monotonic()
print("TIME TO TEST SCM %f SECONDS" % (finish - start))
print()
# OFM evaluation
for ROW in [False, True]:
print("ROW ELEMS", ROW)
# Featurize dataframe with OFM and time it
start = time.monotonic()
ofm = MM.OrbitalFieldMatrix(ROW)
ofm.set_n_jobs(NJOBS)
df = ofm.featurize_dataframe(df, 'structure')
df['orbital field matrix'] = pd.Series([s.flatten() \
for s in df['orbital field matrix']], df.index)
finish = time.monotonic()
print("TIME TO FEATURIZE OFM %f SECONDS" % (finish - start))
print()
# Get OFM descriptor and set up KRR model
krr = KernelRidge()
hpsel = GridSearchCV(krr,
params['orbital field matrix'],
cv=inner_cv,
refit=True)
X = df['orbital field matrix'].as_matrix()