def test_small_moddata_featurization():
""" This test creates a new MODData from the MP 2018.6 structures. """
data_file = Path(__file__).parent.joinpath("data/MP_2018.6_small.zip")
# Loading pickles can be dangerous, so lets at least check that the MD5 matches
# what it was when created
assert (get_sha512_of_file(data_file) ==
"37bd4f8ce6f29c904a13e5670dd53af9a8779094727052ec85ccd6362b1b3765"
"ac613426331811b3f626242896d87c3f6bc1884cc5545875b5ae66a712f9e218")
old = MODData.load(data_file)
structures = old.structures
targets = old.targets
names = old.names
new = MODData(structures, targets, target_names=names)
new.featurize(fast=False)
new_cols = sorted(new.df_featurized.columns.tolist())
old_cols = sorted(old.df_featurized.columns.tolist())
for i in range(len(old_cols)):
print(new_cols[i], old_cols[i])
assert new_cols[i] == old_cols[i]
np.testing.assert_array_equal(old_cols, new_cols)
for col in new.df_featurized.columns:
np.testing.assert_almost_equal(
new.df_featurized[col].to_numpy(),
old.df_featurized[col].to_numpy(),
)
def test_small_moddata_featurization(small_moddata):
""" This test creates a new MODData from the MP 2018.6 structures. """
old = small_moddata
structures = old.structures
targets = old.targets
names = old.names
new = MODData(structures, targets, target_names=names)
new.featurize(fast=False, n_jobs=1)
new_cols = sorted(new.df_featurized.columns.tolist())
old_cols = sorted(old.df_featurized.columns.tolist())
for i in range(len(old_cols)):
print(new_cols[i], old_cols[i])
assert new_cols[i] == old_cols[i]
np.testing.assert_array_equal(old_cols, new_cols)
for col in new.df_featurized.columns:
np.testing.assert_almost_equal(
new.df_featurized[col].to_numpy(),
old.df_featurized[col].to_numpy(),
)
def featurize(task, n_jobs=1):
import warnings
warnings.filterwarnings("ignore", category=RuntimeWarning)
from modnet.preprocessing import MODData
from modnet.featurizers.presets import DeBreuck2020Featurizer
from matminer.datasets import load_dataset
if task == "matbench_elastic":
df_g = load_dataset("matbench_log_gvrh")
df_k = load_dataset("matbench_log_kvrh")
df = df_g.join(df_k.drop("structure", axis=1))
else:
df = load_dataset(task)
mapping = {
col: col.replace(" ", "_").replace("(", "").replace(")", "")
for ind, col in enumerate(df.columns)
}
df.rename(columns=mapping, inplace=True)
targets = [
col for col in df.columns
if col not in ("id", "structure", "composition")
]
if "structure" not in df.columns:
featurizer = CompositionOnlyFeaturizer()
else:
featurizer = DeBreuck2020Featurizer(fast_oxid=True)
try:
materials = df["structure"] if "structure" in df.columns else df[
"composition"].map(Composition)
except KeyError:
raise RuntimeError(
f"Could not find any materials data dataset for task {task!r}!")
data = MODData(
materials=materials.tolist(),
targets=df[targets].values,
target_names=targets,
featurizer=featurizer,
)
data.featurize(n_jobs=n_jobs)
os.makedirs("./precomputed", exist_ok=True)
data.save(f"./precomputed/{task}_moddata.pkl.gz")
return data
def test_small_moddata_composition_featurization(small_moddata_composition):
""" This test creates a new MODData from the MP 2018.6 structures. """
reference = small_moddata_composition
compositions = reference.compositions
new = MODData(materials=compositions)
new.featurize(fast=False, n_jobs=1)
new_cols = sorted(new.df_featurized.columns.tolist())
ref_cols = sorted(reference.df_featurized.columns.tolist())
for i in range(len(ref_cols)):
# print(new_cols[i], ref_cols[i])
assert new_cols[i] == ref_cols[i]
for col in new.df_featurized.columns:
np.testing.assert_almost_equal(
new.df_featurized[col].to_numpy(),
reference.df_featurized[col].to_numpy(),
)
materials = train_df[
"structure"] if "structure" in train_df.columns else train_df[
"composition"].map(Composition)
except KeyError:
raise RuntimeError(
f"Could not find any materials data dataset for task {task!r}!"
)
fast_oxid_featurizer = DeBreuck2020Featurizer(fast_oxid=True)
train_data = MODData(
materials=materials.tolist(),
targets=train_df[targets].values,
target_names=targets,
featurizer=fast_oxid_featurizer,
)
train_data.featurize(n_jobs=32)
train_data.feature_selection(n=-1, use_precomputed_cross_nmi=True)
# create model
targets_hierarchy = [[[field for field in targets]]]
weights = {field: 1 for field in targets}
model = EnsembleMODNetModel(targets_hierarchy, weights)
# fit model
if USE_GA:
# you can either use a GA for hyper-parameter optimization or...
from modnet.hyper_opt import FitGenetic
ga = FitGenetic(train_data)
model = ga.run(
size_pop=20,
