def finite_difference_b_tensor_element(self, *cart_coords, **kwargs):
rigor = pop_multikwarg(kwargs, 'rigor', 'robustness') or self.b_tensor_finite_difference_rigor
forward = pop_multikwarg(kwargs, 'forward', 'use_forward') or False
key = (self, self.molecule.cartesian_units, rigor, forward)
manager = InternalCoordinate._b_finite_difference_managers.get(key, None)
if manager is None:
manager = _BTensorFiniteDifferenceManager(
robustness=rigor,
units=self.molecule.cartesian_units,
forward=forward
)
InternalCoordinate._b_finite_difference_managers[key] = manager
return manager.get_b_tensor_entry(self, cart_coords)
def __new__(cls, *args, **kwargs):
"""
TODO Move this to class level documentation, since it doesn't show up in Sphinx
Takes several different forms. Given a number of arguments (possibly nested lists), a Tensor is created as
expected. *This is the only form in which arguments may be specified without keywords.*
`Tensor` can also be initialized by giving a list of dimensions for the `shape` (aliased as `dimension`)
keyword argument, with a possible default value keyword argument `default_val`.
Unless otherwise specified via the `dtype` keyword argument, it is assumed that the input data should be cast
as a `numpy.float64`.
"""
# Pop off any special args or kwargs and store their values for later
has_indices = False
ret_val = None
indices = None
units = pop_kwarg(kwargs, 'units')
name = kwargs.pop('name', None)
#--------------------------------------------------------------------------------#
# pop off any kwargs that the subclass's __init__ takes...
subclass_kwargs = {}
for supercls in cls.__mro__:
if supercls is Tensor:
break
if hasattr(supercls, '__tensor_init__') and callable(supercls.__tensor_init__):
if hasattr(supercls.__tensor_init__, 'getargspec'):
argspec = supercls.__tensor_init__.getargspec()
else:
argspec = inspect.getargspec(supercls.__tensor_init__)
for arg in argspec.args[1:]:
subclass_kwargs[arg] = kwargs.pop(arg, None)
#--------------------------------------------------------------------------------#
# Check for indices...
indices_kwarg = kwargs.pop('indices', None)
if indices_kwarg is None and len(args) == 1 and isinstance(args[0], basestring):
args = list(args)
indices_kwarg = args.pop(0)
args = tuple(args)
index_range_set = pop_multikwarg(kwargs, 'index_range_set', 'in_set', 'set')
if indices_kwarg is not None:
has_indices = True
indices = EinsumTensor.split_indices(indices_kwarg)
if index_range_set is None:
index_range_set = IndexRange.global_index_range_set
shape = []
for idx in indices:
if idx not in index_range_set.known_ranges:
raise IndexError("unknown index '" + idx + "'")
shape.append(index_range_set.known_ranges[idx].size)
shape = tuple(shape)
if "shape" in kwargs:
kwshape = kwargs.pop('shape')
if shape != kwshape:
raise TypeError("inconsistent shape: indices '{}' indicate a shape of {}, but"
" the keyword 'shape' was given with the shape {}".format(
",".join(indices), shape, kwshape
))
kwargs['shape'] = shape
#--------------------------------------------------------------------------------#
# Now create a numpy.ndarray object...
def_val = pop_kwarg(kwargs, 'default_val', 'default_value', 'default') or 0.0
# Set the default data type to float, unless the user specifies
dtype = get_kwarg(kwargs, 'dtype') or float
if not callable(dtype) and grendel.show_warnings:
warn("dtype given to {0} constructor is not Callable and thus cannot be used for casting." \
" It is better to use callable types for dtype if possible; e.g. numpy.float64 instead" \
"of numpy.dtype('float64')".format(classname(cls)))
kwargs['dtype'] = dtype
# Typecast the default value
if callable(dtype):
def_val = dtype(def_val)
# See if we have a shape...
shape = pop_kwarg(kwargs, 'shape', 'dimension')
# See if we have data...
# This allows us to support the form Tensor(1, 2, 3, 4)
if len(args) == 1 and isinstance(args[0], np.ndarray):
data = args[0]
else:
data = listify_args(*args)
if 'data' in kwargs:
if data:
raise TypeError("`data` may be specified as a keyword argument or as " \
"the regular arguments to {0} constructor, but not both.".format(classname(cls)))
else:
data = pop_kwarg('data')
if shape and not isinstance(data, np.ndarray):
has_content = False
if len(args) != 0:
has_content = True
if not has_content:
if def_val == 0.0:
try:
ret_val = np.zeros(shape=shape, **kwargs)
except:
raise
else:
ret_val = (np.ones(shape=shape, **kwargs) * def_val)
else:
if grendel.sanity_checking_enabled:
# Check data length
tmp = np.array(data)
needed_data_size = 1
for dim in shape: needed_data_size *= dim
try:
tmp.reshape((needed_data_size,))
except ValueError:
raise ValueError("Data provided to {0} constructor is incompatible with the shape {1}".format(classname(cls), shape))
# Check data type
ret_val = np.array(data, **kwargs).reshape(shape)
else:
# Just pass on the data and any surviving kwargs
try:
if isinstance(data, np.ndarray) and (
len(kwargs) == 0
or (len(kwargs) == 1 and 'dtype' in kwargs and kwargs['dtype'] == data.dtype)
):
# Just do a view
ret_val = data.view(cls)
else:
# Otherwise, we need to call the numpy "constructor" (actually a factory function) of ndarray
ret_val = np.array(data, **kwargs)
except:
# debugging breakpoint hook
raise
if shape and ret_val.shape != shape:
raise ValueError("Shape mismatch: data shape {0} does not match specified shape {1}".format(
data.shape, shape
))
#--------------------------------------------------------------------------------#
# View-cast the ret_val to the class in question, but only if we haven't already
if not isinstance(ret_val, cls):
ret_val = ret_val.view(cls)
# Now assign stuff from any special args...
if has_indices:
ret_val.indices = indices
ret_val.index_range_set = index_range_set
else:
ret_val.indices = None
ret_val.units = units
ret_val.name = name
if name is None:
ret_val.name = "(unnamed tensor)"
# pass the remaining kwargs to the initializer...
ret_val.__tensor_init__(**subclass_kwargs)
return ret_val
def __new__(cls, *args, **kwargs):
"""
TODO Move this to class level documentation, since it doesn't show up in Sphinx
Takes several different forms. Given a number of arguments (possibly nested lists), a Tensor is created as
expected. *This is the only form in which arguments may be specified without keywords.*
`Tensor` can also be initialized by giving a list of dimensions for the `shape` (aliased as `dimension`)
keyword argument, with a possible default value keyword argument `default_val`.
Unless otherwise specified via the `dtype` keyword argument, it is assumed that the input data should be cast
as a `numpy.float64`.
"""
# Pop off any special args or kwargs and store their values for later
has_indices = False
ret_val = None
indices = None
units = pop_kwarg(kwargs, 'units')
name = kwargs.pop('name', None)
#--------------------------------------------------------------------------------#
# pop off any kwargs that the subclass's __init__ takes...
subclass_kwargs = {}
for supercls in cls.__mro__:
if supercls is Tensor:
break
if hasattr(supercls, '__tensor_init__') and callable(
supercls.__tensor_init__):
if hasattr(supercls.__tensor_init__, 'getargspec'):
argspec = supercls.__tensor_init__.getargspec()
else:
argspec = inspect.getargspec(supercls.__tensor_init__)
for arg in argspec.args[1:]:
subclass_kwargs[arg] = kwargs.pop(arg, None)
#--------------------------------------------------------------------------------#
# Check for indices...
indices_kwarg = kwargs.pop('indices', None)
if indices_kwarg is None and len(args) == 1 and isinstance(
args[0], basestring):
args = list(args)
indices_kwarg = args.pop(0)
args = tuple(args)
index_range_set = pop_multikwarg(kwargs, 'index_range_set', 'in_set',
'set')
if indices_kwarg is not None:
has_indices = True
indices = EinsumTensor.split_indices(indices_kwarg)
if index_range_set is None:
index_range_set = IndexRange.global_index_range_set
shape = []
for idx in indices:
if idx not in index_range_set.known_ranges:
raise IndexError("unknown index '" + idx + "'")
shape.append(index_range_set.known_ranges[idx].size)
shape = tuple(shape)
if "shape" in kwargs:
kwshape = kwargs.pop('shape')
if shape != kwshape:
raise TypeError(
"inconsistent shape: indices '{}' indicate a shape of {}, but"
" the keyword 'shape' was given with the shape {}".
format(",".join(indices), shape, kwshape))
kwargs['shape'] = shape
#--------------------------------------------------------------------------------#
# Now create a numpy.ndarray object...
def_val = pop_kwarg(kwargs, 'default_val', 'default_value',
'default') or 0.0
# Set the default data type to float, unless the user specifies
dtype = get_kwarg(kwargs, 'dtype') or float
if not callable(dtype) and grendel.show_warnings:
warn("dtype given to {0} constructor is not Callable and thus cannot be used for casting." \
" It is better to use callable types for dtype if possible; e.g. numpy.float64 instead" \
"of numpy.dtype('float64')".format(classname(cls)))
kwargs['dtype'] = dtype
# Typecast the default value
if callable(dtype):
def_val = dtype(def_val)
# See if we have a shape...
shape = pop_kwarg(kwargs, 'shape', 'dimension')
# See if we have data...
# This allows us to support the form Tensor(1, 2, 3, 4)
if len(args) == 1 and isinstance(args[0], np.ndarray):
data = args[0]
else:
data = listify_args(*args)
if 'data' in kwargs:
if data:
raise TypeError("`data` may be specified as a keyword argument or as " \
"the regular arguments to {0} constructor, but not both.".format(classname(cls)))
else:
data = pop_kwarg('data')
if shape and not isinstance(data, np.ndarray):
has_content = False
if len(args) != 0:
has_content = True
if not has_content:
if def_val == 0.0:
try:
ret_val = np.zeros(shape=shape, **kwargs)
except:
raise
else:
ret_val = (np.ones(shape=shape, **kwargs) * def_val)
else:
if grendel.sanity_checking_enabled:
# Check data length
tmp = np.array(data)
needed_data_size = 1
for dim in shape:
needed_data_size *= dim
try:
tmp.reshape((needed_data_size, ))
except ValueError:
raise ValueError(
"Data provided to {0} constructor is incompatible with the shape {1}"
.format(classname(cls), shape))
# Check data type
ret_val = np.array(data, **kwargs).reshape(shape)
else:
# Just pass on the data and any surviving kwargs
try:
if isinstance(data, np.ndarray) and (
len(kwargs) == 0 or
(len(kwargs) == 1 and 'dtype' in kwargs
and kwargs['dtype'] == data.dtype)):
# Just do a view
ret_val = data.view(cls)
else:
# Otherwise, we need to call the numpy "constructor" (actually a factory function) of ndarray
ret_val = np.array(data, **kwargs)
except:
# debugging breakpoint hook
raise
if shape and ret_val.shape != shape:
raise ValueError(
"Shape mismatch: data shape {0} does not match specified shape {1}"
.format(data.shape, shape))
#--------------------------------------------------------------------------------#
# View-cast the ret_val to the class in question, but only if we haven't already
if not isinstance(ret_val, cls):
ret_val = ret_val.view(cls)
# Now assign stuff from any special args...
if has_indices:
ret_val.indices = indices
ret_val.index_range_set = index_range_set
else:
ret_val.indices = None
ret_val.units = units
ret_val.name = name
if name is None:
ret_val.name = "(unnamed tensor)"
# pass the remaining kwargs to the initializer...
ret_val.__tensor_init__(**subclass_kwargs)
return ret_val