def check_delta(self, x=None, eps=1e-4, zero=None):
"""Test the analytical derivatives"""
if x is None:
x = self.x0
dxs = np.random.uniform(-eps, eps, (100, len(x)))
if zero is not None:
dxs[:,zero] = 0.0
check_delta(self.fun, x, dxs)
def check_delta(self, x=None, eps=1e-4, zero=None):
"""Test the analytical derivatives"""
if x is None:
x = self.x0
dxs = np.random.uniform(-eps, eps, (100, len(x)))
if zero is not None:
dxs[:, zero] = 0.0
check_delta(self.fun, x, dxs)
def test_switch3_deriv():
rcut = 10.0
for width in 1.0, 2.0, 4.0:
tr = Switch3(width)
def fn(x, do_gradient):
value, gradient = tr.trunc_fn(x[0], rcut)
if do_gradient:
return value, np.array([gradient])
else:
return value
for d in 5.0, 9.5:
x0 = np.array([d])
dxs = np.random.normal(0, 1e-4, 100)
check_delta(fn, x0, dxs)
def test_hammer_deriv():
rcut = 10.0
for tau in 0.5, 1.0, 2.0:
tr = Hammer(tau)
def fn(x, do_gradient):
value, gradient = tr.trunc_fn(x[0], rcut)
if do_gradient:
return value, np.array([gradient])
else:
return value
for d in 5.0, 9.0, 9.9:
x0 = np.array([d])
dxs = np.random.normal(0, 1e-4, 100)
check_delta(fn, x0, dxs)
def check_cell_jacobian(ff, DOFClass, kwargs):
kwargs['do_frozen'] = True
dof = DOFClass(ff, **kwargs)
def fun(celldofs, do_gradient=False):
rvecs = dof._cellvars_to_rvecs(dof._expand_celldofs(celldofs))
if do_gradient:
dof._update(celldofs)
return rvecs.ravel(), dof._get_celldofs_jacobian(celldofs)
else:
return rvecs.ravel()
x = dof._reduce_cellvars(dof.cellvars0)
eps = 1e-5
dxs = np.random.uniform(-eps, eps, (100, len(x)))
check_delta(fun, x, dxs)
def check_cell_jacobian(ff, DOFClass, kwargs):
kwargs['do_frozen'] = True
dof = DOFClass(ff, **kwargs)
def fun(celldofs, do_gradient=False):
rvecs = dof._cellvars_to_rvecs(dof._expand_celldofs(celldofs))
if do_gradient:
dof._update(celldofs)
return rvecs.ravel(), dof._get_celldofs_jacobian(celldofs)
else:
return rvecs.ravel()
x = dof._reduce_cellvars(dof.cellvars0)
eps = 1e-5
dxs = np.random.uniform(-eps, eps, (100, len(x)))
check_delta(fun, x, dxs)
def test_switch3_deriv():
rcut = 10.0
for width in 1.0, 2.0, 4.0:
tr = Switch3(width)
def fn(x, do_gradient):
value, gradient = tr.trunc_fn(x[0], rcut)
if do_gradient:
return value, np.array([gradient])
else:
return value
for d in 5.0, 9.5:
x0 = np.array([d])
dxs = np.random.normal(0, 1e-4, 100)
check_delta(fn, x0, dxs)
def test_hammer_deriv():
rcut = 10.0
for tau in 0.5, 1.0, 2.0:
tr = Hammer(tau)
def fn(x, do_gradient):
value, gradient = tr.trunc_fn(x[0], rcut)
if do_gradient:
return value, np.array([gradient])
else:
return value
for d in 5.0, 9.0, 9.9:
x0 = np.array([d])
dxs = np.random.normal(0, 1e-4, 100)
check_delta(fn, x0, dxs)