def evaluate(self,
test_structures,
test_energies,
test_forces,
test_stresses=None):
"""
Evaluate energies, forces and stresses of structures with trained
machinea learning potentials.
Args:
test_structures ([Structure]): List of Pymatgen Structure Objects.
test_energies ([float]): List of DFT-calculated total energies of
each structure in structures list.
test_forces ([np.array]): List of DFT-calculated (m, 3) forces of
each structure with m atoms in structures list. m can be varied
with each single structure case.
test_stresses (list): List of DFT-calculated (6, ) viriral stresses
of each structure in structures list.
"""
predict_pool = pool_from(test_structures, test_energies, test_forces,
test_stresses)
_, df_orig = convert_docs(predict_pool)
_, df_predict = convert_docs(pool_from(test_structures))
outputs = self.model.predict_objs(objs=test_structures)
df_predict['y_orig'] = df_predict['n'] * outputs
return df_orig, df_predict
def evaluate(self,
test_structures,
test_energies,
test_forces,
test_stresses=None,
include_stress=False):
"""
Evaluate energies, forces and stresses of structures with trained
machinea learning potentials.
Args:
test_structures ([Structure]): List of Pymatgen Structure Objects.
test_energies ([float]): List of DFT-calculated total energies of
each structure in structures list.
test_forces ([np.array]): List of DFT-calculated (m, 3) forces of
each structure with m atoms in structures list. m can be varied
with each single structure case.
test_stresses (list): List of DFT-calculated (6, ) viriral stresses
of each structure in structures list.
include_stress (bool): Whether to include stress components.
"""
test_structures, test_forces, test_stresses = check_structures_forces_stresses(
test_structures, test_forces, test_stresses)
predict_pool = pool_from(test_structures, test_energies, test_forces,
test_stresses)
_, df_orig = convert_docs(predict_pool, include_stress=include_stress)
_, df_predict = convert_docs(pool_from(test_structures),
include_stress=include_stress)
outputs = self.model.predict_objs(objs=test_structures)
df_predict["y_orig"] = df_predict["n"] * outputs
return df_orig, df_predict
def train(self,
train_structures,
train_energies,
train_forces,
train_stresses=None,
include_stress=False,
**kwargs):
"""
Training data with models.
Args:
train_structures ([Structure]): The list of Pymatgen Structure object.
energies ([float]): The list of total energies of each structure
in structures list.
train_energies ([float]): List of total energies of each structure in
structures list.
train_forces ([np.array]): List of (m, 3) forces array of each
structure with m atoms in structures list. m can be varied with
each single structure case.
train_stresses (list): List of (6, ) virial stresses of each
structure in structures list.
include_stress (bool): Whether to include stress components.
"""
train_structures, train_forces, train_stresses = check_structures_forces_stresses(
train_structures, train_forces, train_stresses)
train_pool = pool_from(train_structures, train_energies, train_forces,
train_stresses)
_, df = convert_docs(train_pool, include_stress=include_stress)
ytrain = df["y_orig"] / df["n"]
xtrain = self.model.describer.transform(train_structures)
self.model.fit(features=xtrain, targets=ytrain, **kwargs)
def train(self,
train_structures,
train_energies,
train_forces,
train_stresses=None,
**kwargs):
"""
Training data with model.
Args:
train_structures ([Structure]): The list of Pymatgen Structure object.
energies ([float]): The list of total energies of each structure
in structures list.
train_energies ([float]): List of total energies of each structure in
structures list.
train_forces ([np.array]): List of (m, 3) forces array of each
structure with m atoms in structures list. m can be varied with
each single structure case.
train_stresses (list): List of (6, ) virial stresses of each
structure in structures list.
"""
train_pool = pool_from(train_structures, train_energies, train_forces,
train_stresses)
_, df = convert_docs(train_pool)
ytrain = df['y_orig'] / df['n']
xtrain = self.model.describer.transform(train_structures)
self.model.fit(features=xtrain, targets=ytrain, **kwargs)
def read_cfgs(self, filename="output.data"):
"""
Read the configuration file.
Args:
filename (str): The configuration file to be read.
"""
data_pool = []
with zopen(filename, "rt") as f:
lines = f.read()
block_pattern = re.compile("begin\n(.*?)end", re.S)
lattice_pattern = re.compile("lattice(.*?)\n")
position_pattern = re.compile("atom(.*?)\n")
energy_pattern = re.compile("energy(.*?)\n")
for block in block_pattern.findall(lines):
d = {"outputs": {}}
lattice_str = lattice_pattern.findall(block)
lattice = Lattice(
np.array([latt.split() for latt in lattice_str],
dtype=np.float64) * self.bohr_to_angstrom)
position_str = position_pattern.findall(block)
positions = pd.DataFrame([pos.split() for pos in position_str])
positions.columns = [
"x", "y", "z", "specie", "charge", "atomic_energy", "fx", "fy",
"fz"
]
coords = np.array(positions.loc[:, ["x", "y", "z"]],
dtype=np.float64)
coords = coords * self.bohr_to_angstrom
species = np.array(positions["specie"])
forces = np.array(positions.loc[:, ["fx", "fy", "fz"]],
dtype=np.float64)
forces = forces / self.eV_to_Ha / self.bohr_to_angstrom
energy_str = energy_pattern.findall(block)[0]
energy = float(energy_str.lstrip()) / self.eV_to_Ha
struct = Structure(lattice=lattice,
species=species,
coords=coords,
coords_are_cartesian=True)
d["structure"] = struct.as_dict()
d["outputs"]["energy"] = energy
d["outputs"]["forces"] = forces
d["num_atoms"] = len(struct)
data_pool.append(d)
_, df = convert_docs(docs=data_pool)
return data_pool, df
def read_cfgs(self, filename='output.data'):
"""
Read the configuration file.
Args:
filename (str): The configuration file to be read.
"""
data_pool = []
with zopen(filename, 'rt') as f:
lines = f.read()
block_pattern = re.compile('begin\n(.*?)end', re.S)
lattice_pattern = re.compile('lattice(.*?)\n')
position_pattern = re.compile('atom(.*?)\n')
energy_pattern = re.compile('energy(.*?)\n')
for block in block_pattern.findall(lines):
d = {'outputs': {}}
lattice_str = lattice_pattern.findall(block)
lattice = Lattice(
np.array([latt.split() for latt in lattice_str],
dtype=np.float) * self.bohr_to_angstrom)
position_str = position_pattern.findall(block)
positions = pd.DataFrame([pos.split() for pos in position_str])
positions.columns = \
['x', 'y', 'z', 'specie', 'charge', 'atomic_energy', 'fx', 'fy', 'fz']
coords = np.array(positions.loc[:, ['x', 'y', 'z']],
dtype=np.float)
coords = coords * self.bohr_to_angstrom
species = np.array(positions['specie'])
forces = np.array(positions.loc[:, ['fx', 'fy', 'fz']],
dtype=np.float)
forces = forces / self.eV_to_Ha / self.bohr_to_angstrom
energy_str = energy_pattern.findall(block)[0]
energy = float(energy_str.lstrip()) / self.eV_to_Ha
struct = Structure(lattice=lattice,
species=species,
coords=coords,
coords_are_cartesian=True)
d['structure'] = struct.as_dict()
d['outputs']['energy'] = energy
d['outputs']['forces'] = forces
d['num_atoms'] = len(struct)
data_pool.append(d)
_, df = convert_docs(docs=data_pool)
return data_pool, df
def read_cfgs(filename, predict=False):
"""
Read the configuration file.
Args:
filename (str): The configuration file to be read.
"""
type_convert = {'R': np.float32, 'I': np.int, 'S': np.str}
data_pool = []
with zopen(filename, 'rt') as f:
lines = f.read()
repl = re.compile('AT ')
lines = repl.sub('', string=lines)
block_pattern = re.compile(
r'(\n[0-9]+\n|^[0-9]+\n)(.+?)(?=\n[0-9]+\n|$)', re.S)
lattice_pattern = re.compile(r'Lattice="(.+)"')
# energy_pattern = re.compile('dft_energy=(-?[0-9]+.[0-9]+)', re.I)
energy_pattern = re.compile(
r'(?<=\S{3}\s|dft_)energy=(-?[0-9]+.[0-9]+)')
# stress_pattern = re.compile('dft_virial={(.+)}')
stress_pattern = re.compile(r'dft_virial=({|)(.+?)(}|) \S.*')
properties_pattern = re.compile(r'properties=(\S+)', re.I)
# position_pattern = re.compile('\n(.+)', re.S)
position_pattern = re.compile('\n(.+?)(?=\nE.*|\n\n.*|$)', re.S)
# formatify = lambda string: [float(s) for s in string.split()]
for (size, block) in block_pattern.findall(lines):
d = {'outputs': {}}
size = int(size)
lattice_str = lattice_pattern.findall(block)[0]
lattice = Lattice(
list(map(lambda s: float(s), lattice_str.split())))
energy_str = energy_pattern.findall(block)[-1]
energy = float(energy_str)
stress_str = stress_pattern.findall(block)[0][1]
virial_stress = np.array(
list(map(lambda s: float(s), stress_str.split())))
virial_stress = [virial_stress[i] for i in [0, 4, 8, 1, 5, 6]]
properties = properties_pattern.findall(block)[0].split(":")
labels_columns = OrderedDict()
labels = defaultdict()
for i in range(0, len(properties), 3):
labels_columns[properties[i]] = [
int(properties[i + 2]), properties[i + 1]
]
position_str = position_pattern.findall(block)[0].split('\n')
position = np.array([p.split() for p in position_str])
column_index = 0
for key in labels_columns:
num_columns, dtype = labels_columns[key]
labels[key] = position[:, column_index:column_index +
num_columns].astype(type_convert[dtype])
column_index += num_columns
struct = Structure(lattice=lattice,
species=labels['species'].ravel(),
coords=labels['pos'],
coords_are_cartesian=True)
if predict:
forces = labels['force']
else:
forces = labels['dft_force']
d['structure'] = struct.as_dict()
d['outputs']['energy'] = energy
assert size == struct.num_sites
d['num_atoms'] = size
d['outputs']['forces'] = forces
d['outputs']['virial_stress'] = virial_stress
data_pool.append(d)
_, df = convert_docs(docs=data_pool)
return data_pool, df
def read_cfgs(self, filename):
"""
Args:
filename (str): The configuration file to be read.
"""
def formatify(string):
return [float(s) for s in string.split()]
if not self.elements:
raise ValueError("No species given.")
data_pool = []
with zopen(filename, 'rt') as f:
lines = f.read()
block_pattern = re.compile('BEGIN_CFG\n(.*?)\nEND_CFG', re.S)
size_pattern = re.compile('Size\n(.*?)\n SuperCell', re.S | re.I)
lattice_pattern = re.compile('SuperCell\n(.*?)\n AtomData',
re.S | re.I)
position_pattern = re.compile('fz\n(.*?)\n Energy', re.S)
energy_pattern = re.compile('Energy\n(.*?)\n (?=PlusStress|Stress)',
re.S)
stress_pattern = re.compile('xy\n(.*?)(?=\n|$)', re.S)
for block in block_pattern.findall(lines):
d = {'outputs': {}}
size_str = size_pattern.findall(block)[0]
size = int(size_str.lstrip())
lattice_str = lattice_pattern.findall(block)[0]
lattice = Lattice(
np.array(list(map(formatify, lattice_str.split('\n')))))
position_str = position_pattern.findall(block)[0]
position = np.array(list(map(formatify, position_str.split('\n'))))
species = np.array(self.elements)[position[:, 1].astype(np.int)]
forces = position[:, 5:8].tolist()
position = position[:, 2:5]
energy_str = energy_pattern.findall(block)[0]
energy = float(energy_str.lstrip())
stress_str = stress_pattern.findall(block)[0]
virial_stress = np.array(list(map(formatify,
stress_str.split()))).reshape(
6, ).tolist()
virial_stress = [
virial_stress[self.mtp_stress_order.index(n)]
for n in self.vasp_stress_order
]
struct = Structure(lattice=lattice,
species=species,
coords=position,
coords_are_cartesian=True)
d['structure'] = struct.as_dict()
d['outputs']['energy'] = energy
assert size == struct.num_sites
d['num_atoms'] = size
d['outputs']['forces'] = forces
d['outputs']['virial_stress'] = virial_stress
data_pool.append(d)
_, df = convert_docs(docs=data_pool)
return data_pool, df