def testOptim(self):
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap=bp_model)
calc = PyXtalFFCalculator(ff=ff)
si = bulk('Si', 'diamond', a=5.0, cubic=True)
si.set_calculator(calc)
si = optimize(si, box=True)
self.assertTrue(abs(si.get_cell()[0][0] - 5.469) < 1e-2)
def test_6_LR_calculator(self):
#calc = PyXtalFFCalculator(mliap='unittest/PolyReg-checkpoint.pth', logo=False)
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap='unittest/PolyReg-checkpoint.pth')
calc = PyXtalFFCalculator(ff=ff)
self.struc.set_calculator(calc)
self.struc.get_potential_energy()
self.struc.get_stress()
def test_elastic(self):
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap=bp_model)
calc = PyXtalFFCalculator(ff=ff)
si = bulk('Si', 'diamond', a=5.469, cubic=True)
si.set_calculator(calc)
C, C_err = fit_elastic_constants(si, symmetry='cubic', optimizer=BFGS)
C /= units.GPa
self.assertTrue(abs(C[0, 0] - 124.5) < 1.0)
class TestRegression(unittest.TestCase):
model = {
'system': system,
'hiddenlayers': [12, 12],
'epoch': 10,
'stress_coefficient': None,
'force_coefficient': 0.03,
'path': 'unittest/'
}
ff = PyXtal_FF(descriptors=descriptor, model=model)
struc = bulk('Si', 'diamond', a=5.0, cubic=True)
@classmethod
def tearDownClass(cls):
shutil.rmtree('unittest')
def test_1_NN_LBFGS(self):
self.ff.algorithm = 'NN'
self.ff._model['optimizer'] = {'method': 'lbfgs'}
(train_stat, _) = self.ff.run(mode='train', TrainData=TrainData)
def test_2_NN_ADAM(self):
self.ff.algorithm = 'NN'
self.ff._model['optimizer'] = {'method': 'ADAM'}
self.ff._MODEL(self.ff._model)
train_stat = self.ff.model.train('Train_db', self.ff.optimizer)
self.ff.model.save_checkpoint(des_info=self.ff._descriptors)
def test_3_lr(self):
self.ff.algorithm = 'PR'
self.ff._model['order'] = 1
self.ff._MODEL(self.ff._model)
train_stat = self.ff.model.train('Train_db', None)
self.ff.model.save_checkpoint(des_info=self.ff._descriptors)
def test_4_qr(self):
self.ff.algorithm = 'PR'
self.ff._model['order'] = 2
self.ff._MODEL(self.ff._model)
train_stat = self.ff.model.train('Train_db', None)
def test_5_NN_calculator(self):
#calc = PyXtalFFCalculator(mliap='unittest/12-12-checkpoint.pth', logo=False)
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap='unittest/12-12-checkpoint.pth')
calc = PyXtalFFCalculator(ff=ff)
self.struc.set_calculator(calc)
self.struc.get_potential_energy()
self.struc.get_stress()
def test_6_LR_calculator(self):
#calc = PyXtalFFCalculator(mliap='unittest/PolyReg-checkpoint.pth', logo=False)
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap='unittest/PolyReg-checkpoint.pth')
calc = PyXtalFFCalculator(ff=ff)
self.struc.set_calculator(calc)
self.struc.get_potential_energy()
self.struc.get_stress()
class TestRegressionComp(unittest.TestCase):
model = {'system' : system,
'stress_coefficient': None,
'force_coefficient': 0.03,
'path': 'unittest_comp/',
'algorithm': 'PR',
}
ff = PyXtal_FF(descriptors=descriptor_comp, model=model)
@classmethod
def tearDownClass(cls):
shutil.rmtree('unittest_comp')
def test_lr_comp(self):
self.ff._model['order'] = 1
(train_stat, _) = self.ff.run(mode='train', TrainData=TrainData)
},
'Rc': 5.0,
'parameters': {
'lmax': 4,
'nmax': 3
},
'base_potential': {
'inner': 1.5,
'outer': 2.0
}, #zbl potential
'ncpu': 1,
}
model = {
'system': ['Si'],
'hiddenlayers': [12, 12],
'path': folder,
#'restart': folder + '12-12-checkpoint.pth',
'optimizer': {
'method': 'lbfgs'
},
'force_coefficient': 2e-2,
'stress_coefficient': 2e-3,
"stress_group": ["Elastic"],
'alpha': 1e-6,
'epoch': 1000,
}
ff = PyXtal_FF(descriptors=descriptor, model=model)
ff.run(mode='train', TrainData=TrainData, TestData=TestData)
(options, args) = parser.parse_args()
def printenergy(a, it, t0):
"""Function to print the potential, kinetic and total energy"""
epot = a.get_potential_energy() / len(a)
ekin = a.get_kinetic_energy() / len(a)
t_now = time()
print('Step: %4d [%6.2f]: Epot = %.3feV Ekin = %.3feV (T=%3.0fK) '
'Etot = %.3feV ' % (\
it, t_now-t0, epot, ekin, ekin / (1.5 * units.kB), epot + ekin))
return t_now
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap=options.file)
calc = PyXtalFFCalculator(ff=ff)
si = bulk('Si', 'diamond', a=5.659, cubic=True)
si = si * 5
print("MD simulation for ", len(si), " atoms")
si.set_calculator(calc)
MaxwellBoltzmannDistribution(si, 1000 * units.kB)
dyn = Langevin(si, timestep=5 * units.fs, temperature_K=1000, friction=0.02)
#dyn = VelocityVerlet(si, 5*units.fs) # 2 fs time step.
t0 = time()
for i in range(10):
dyn.run(steps=1)
t_now = printenergy(si, i, t0)
warnings.simplefilter("ignore")
des, folder = "sna", "SiO2-snap"
mliap = folder + "/30-30-checkpoint.pth"
lmpiap = folder + "/NN_weights.txt"
lmpdes = folder + "/DescriptorParam.txt"
# initial silicon crystal
si = read('lt_quartz.cif')
# set the ordering
si.set_tags([2]*3+[1]*6)
si.positions[0,0] += (random() - 0.5)
# ase pyxtal_ff calculator
ff = PyXtal_FF(model={'system': ["Si", "O"]}, logo=False)
ff.run(mode='predict', mliap=mliap)
calc_pff = PyXtalFFCalculator(ff=ff)
# ase lammps calculatoor
lammps_name=''
comm=None
log_file='lammps.log'
cmd_args = ['-echo', 'log', '-log', log_file,
'-screen', 'none', '-nocite']
lmp = lammps(lammps_name, cmd_args, comm)
parameters = ["mass * 1.0",
"pair_style mliap model nn " + lmpiap + " descriptor " + des + " " + lmpdes,
"pair_coeff * * O Si"
]
print('downloading the training data')
os.system(
'wget https://raw.githubusercontent.com/MDIL-SNU/SIMPLE-NN/master/examples/SiO2/ab_initio_output/OUTCAR_comp'
)
descriptor = {
'Rc': 4.9,
'parameters': {
'lmax': 3
},
'N_train': 250,
'ncpu': 4,
}
model = {
'system': ['Si', 'O'],
'hiddenlayers': [30, 30],
'activation': ['Tanh', 'Tanh', 'Linear'],
'force_coefficient': 0.1,
'epoch': 500,
'path': 'SiO2-quick/',
#'restart': 'SiO2-quick/30-30-checkpoint.pth',
'optimizer': {
'method': 'lbfgs'
},
}
#------------------------- Run NN calculation ------------------------------
ff = PyXtal_FF(descriptors=descriptor, model=model)
ff.run(mode='train', TrainData=train_data)
'ncpu': 1,
}
model = {
'system': ['Si'],
'hiddenlayers': [20, 20],
'random_seed': 12345,
'activation': ['Tanh', 'Tanh', 'Linear'],
'optimizer': {
'method': 'lbfgs'
},
'force_coefficient': 1.,
'stress_coefficient': 1.,
'alpha': 1e-6,
'epoch': 500,
}
ff = PyXtal_FF(descriptors=descriptor, model=model)
ff.run(mode='train', TrainData=TrainData, TestData=None)
shutil.copyfile('Si-SNAP/20-20-checkpoint.pth', "potentials/Si-20-20-0K.pth")
shutil.rmtree('Si-SNAP')
Ts = [500, 750, 1000, 1250, 1500]
for i, T in enumerate(Ts):
print("Temperature: ", T)
ase_db = f"Si{T}.db"
struc = bulk("Si", 'diamond', a=5.468728, cubic=True) * (2, 2, 2)
struc.info['md_step'] = 0
mliap = f'potentials/Si-20-20-0K.pth' if i == 0 else f'potentials/Si-20-20-{Ts[i-1]}K.pth'
ff = PyXtal_FF(model={'system': ["Si"]}, logo=False)
ff.run(mode='predict', mliap=mliap)
calc = PyXtalFFCalculator(ff=ff)
struc.set_calculator(calc)