def __init__(self,
value=None,
shape=None,
dtype=None,
device=None,
name=''):
if not device:
device = use_default_device()
if (np.isscalar(value) or isinstance(value, np.ndarray)) and not shape:
shape = ()
if dtype is not None:
if isinstance(value, np.ndarray) and dtype != value.dtype:
value = np.array(value, dtype=dtype)
else:
if isinstance(value, np.ndarray):
dtype = value.dtype
else:
dtype = np.float32
if np.isscalar(value):
super(Constant, self).__init__(sanitize_shape(shape),
sanitize_dtype_cntk(dtype), value,
device, name)
else:
ndav = sanitize_value(shape, value, dtype, device)
super(Constant, self).__init__(ndav, name)
def __init__(self,
shape=None,
init=None,
dtype=None,
device=None,
name=''):
if not device:
device = use_default_device()
if dtype is not None:
if isinstance(init, np.ndarray) and dtype != init.dtype:
init = np.array(init, dtype=dtype)
else:
if np.isscalar(init) and not shape:
shape = ()
if isinstance(init, np.ndarray):
dtype = init.dtype
else:
dtype = np.float32
if init is None:
init = 0
if isinstance(init, (np.ndarray, list, float, int)):
ndav = sanitize_value(shape, init, dtype, device)
super(Parameter, self).__init__(ndav, name)
else:
shape = sanitize_shape(shape)
cntk_dtype = sanitize_dtype_cntk(dtype)
super(Parameter, self).__init__(shape, cntk_dtype, init, device,
name)
def __init__(self, shape, data_type, device=None):
from cntk.internal import sanitize_shape, sanitize_dtype_cntk
shape = sanitize_shape(shape)
data_type = sanitize_dtype_cntk(data_type)
if device is None:
device = use_default_device()
super(NDArrayView, self).__init__(data_type, cntk_py.StorageFormat_Dense, shape,
device)
def __init__(self, shape, data_type, device=None):
from cntk.internal import sanitize_shape, sanitize_dtype_cntk
shape = sanitize_shape(shape)
data_type = sanitize_dtype_cntk(data_type)
if device is None:
device = use_default_device()
super(NDArrayView, self).__init__(data_type, cntk_py.StorageFormat_Dense, shape,
device)
def __init__(self, shape=None, dtype=None, needs_gradient=False, is_sparse=False,
dynamic_axes=[cntk_py.Axis.default_batch_axis(), cntk_py.Axis.default_dynamic_axis()], name=''):
shape = sanitize_shape(shape)
if dtype is None:
dtype = np.float32
dtype = sanitize_dtype_cntk(dtype)
dynamic_axes = sanitize_dynamic_axes(dynamic_axes)
super(Variable, self).__init__(shape, is_sparse, dtype, needs_gradient, name, dynamic_axes)
def __init__(self, shape=None, dtype=None, needs_gradient=False, is_sparse=False,
dynamic_axes=[cntk_py.Axis.default_batch_axis(), cntk_py.Axis.default_dynamic_axis()], name=''):
shape = sanitize_shape(shape)
if dtype is None:
dtype = np.float32
dtype = sanitize_dtype_cntk(dtype)
dynamic_axes = sanitize_dynamic_axes(dynamic_axes)
super(Variable, self).__init__(shape, is_sparse, dtype, needs_gradient, name, dynamic_axes)
def __init__(self, value=None, shape=None, dtype=None, device=None, name=''):
if dtype is None:
if isinstance(value, np.ndarray):
dtype = value.dtype
else:
dtype = np.float32
if device is None:
device = DeviceDescriptor.use_default_device()
if np.isscalar(value):
super(Constant, self).__init__(sanitize_shape(shape),
sanitize_dtype_cntk(dtype), value, device, name)
else:
ndav = sanitize_value(shape, value, dtype, device)
super(Constant, self).__init__(ndav, name)
def __init__(self, shape=None, init=None, dtype=None,
device=None, name=''):
if dtype is None:
if isinstance(init, np.ndarray):
dtype = init.dtype
else:
dtype = np.float32
if init is None:
init = 0
if isinstance(init, (np.ndarray, list, float, int)):
ndav = sanitize_value(shape, init, dtype, device)
super(Parameter, self).__init__(ndav, name)
else:
shape = sanitize_shape(shape)
cntk_dtype = sanitize_dtype_cntk(dtype)
super(Parameter, self).__init__(shape, cntk_dtype, init,
device, name)
def __init__(self,
value=None,
shape=None,
dtype=None,
device=None,
name=''):
if dtype is None:
if isinstance(value, np.ndarray):
dtype = value.dtype
else:
dtype = np.float32
if device is None:
device = DeviceDescriptor.use_default_device()
if np.isscalar(value):
super(Constant, self).__init__(sanitize_shape(shape),
sanitize_dtype_cntk(dtype), value,
device, name)
else:
ndav = sanitize_value(shape, value, dtype, device)
super(Constant, self).__init__(ndav, name)
def __init__(self,
shape=None,
init=None,
dtype=None,
device=None,
name=''):
if dtype is None:
if isinstance(init, np.ndarray):
dtype = init.dtype
else:
dtype = np.float32
if init is None:
init = 0
if isinstance(init, (np.ndarray, list, float, int)):
ndav = sanitize_value(shape, init, dtype, device)
super(Parameter, self).__init__(ndav, name)
else:
shape = sanitize_shape(shape)
cntk_dtype = sanitize_dtype_cntk(dtype)
super(Parameter, self).__init__(shape, cntk_dtype, init, device,
name)
def __getitem__(self, shape):
shape = sanitize_shape(shape)
return Variable._Type(shape, **kwargs) # inject it for @Function
def create(var, data, seq_starts=None, device=None, read_only=False):
'''
Creates a :class:`~cntk.core.Value` object.
Args:
var (:class:`~cntk.variables.Variable`): variable into which
``data`` is passed
data: data for `var`.
It can be:
* a single NumPy array denoting the full minibatch
* a list of NumPy arrays or SciPy sparse CSR matrices
* a single NumPy array denoting one parameter or constant
seq_starts (list of `bool`\ s or None): if None, every sequence is
treated as a new sequence. Otherwise, it is interpreted as a list of
Booleans that tell whether a sequence is a new sequence (`True`) or a
continuation of the sequence in the same slot of the previous
minibatch (`False`)
device (:class:`~cntk.device.DeviceDescriptor`, default None): device
this value should be put on
read_only (bool, default False): whether the data is read only
Returns:
:class:`~cntk.core.Value` object.
'''
if not isinstance(var, cntk_py.Variable):
raise TypeError('Variable expected, but got "%s"' % type(var))
if not var.dynamic_axes:
# No dynamic axes -> we can pass everything in one go
data = Value._as_best_data_type(var, data)
# Since the core API's Value does not copy single NDArrayViews,
# we cannot borrow the memory here.
ndav = NDArrayView.from_data(data, device=device, borrow=False)
return cntk_py.Value(ndav)
elif len(var.dynamic_axes) <= 1 and isinstance(data,
list) and len(data) > 1:
warnings.warn('you provided the minibatch data as a list, but '
'your corresponding input variable (uid "%s") has '
'only one dynamic axis (batch axis). To speed up '
'graph execution, please convert the data '
'beforehand into one NumPy array to speed up '
'training.' % var.uid)
if isinstance(data, cntk_py.NDArrayView):
return cntk_py.Value(data)
if isinstance(data, np.ndarray):
# The outermost axis has to be Python list. If the user passes a
# full minibatch as one NumPy array, we have to convert it.
if data.dtype == object:
raise ValueError(
'dtype object is not supported. If this is a '
'batch of sequences, you need to pass them as a '
'pure-Python list of NumPy arrays')
if seq_starts:
data = list(np.atleast_1d(data))
else:
data = Value._as_best_data_type(var, data)
ndav = NDArrayView.from_data(data, device)
return cntk_py.Value(ndav)
if isinstance(data, sparse.csr_matrix):
if seq_starts:
raise ValueError(
'seq_starts are not supported with data specified as a single '
'SciPy sparse CSR matrix. Please provide a list instead')
else:
data = Value._as_best_data_type(var, data)
ndav = NDArrayView.from_data(data, device)
return cntk_py.Value(ndav)
if not isinstance(data, list):
raise ValueError('batch must be a list of NumPy arrays or '
'SciPy CSR matrices')
# NDArrayViews are all created on CPU. The Value object later then will
# move it to the requested device.
# As Value will later create copies anyways, we do not create copies in
# NDArrayView itself. Because of that, we need to keep around the
# instances _as_best_data_type() until we have passed them to
# Value_create() where it will be copied further.
data = [Value._as_best_data_type(var, sample) for sample in data]
device = device or use_default_device()
borrow = device.type() == DeviceKind.CPU
list_of_ndavs = [
NDArrayView.from_data(sample, device=cpu(), borrow=borrow)
for sample in data
]
from cntk.internal import sanitize_shape
value = cntk_py.Value_create(sanitize_shape(var.shape), list_of_ndavs,
seq_starts or [], device, read_only,
True) # always create a copy in Value
return value
def create(var, data, seq_starts=None, device=None, read_only=False):
'''
Creates a :class:`~cntk.core.Value` object.
Args:
var (:class:`~cntk.variables.Variable`): variable into which
``data`` is passed
data: data for `var`.
It can be:
* a single NumPy array denoting the full minibatch
* a list of NumPy arrays or SciPy sparse CSR matrices
* a single NumPy array denoting one parameter or constant
seq_starts (list of `bool`\ s or None): if None, every sequence is
treated as a new sequence. Otherwise, it is interpreted as a list of
Booleans that tell whether a sequence is a new sequence (`True`) or a
continuation of the sequence in the same slot of the previous
minibatch (`False`)
device (:class:`~cntk.device.DeviceDescriptor`, default None): device
this value should be put on
read_only (bool, default False): whether the data is read only
Returns:
:class:`~cntk.core.Value` object.
'''
if not isinstance(var, cntk_py.Variable):
raise TypeError('Variable expected, but got "%s"' % type(var))
if not var.dynamic_axes:
# No dynamic axes -> we can pass everything in one go
data = Value._as_best_data_type(var, data)
# Since the core API's Value does not copy single NDArrayViews,
# we cannot borrow the memory here.
ndav = NDArrayView.from_data(data, device=device, borrow=False)
return cntk_py.Value(ndav)
elif len(var.dynamic_axes) <= 1 and isinstance(data, list) and len(data) > 1:
warnings.warn('you provided the minibatch data as a list, but '
'your corresponding input variable (uid "%s") has '
'only one dynamic axis (batch axis). To speed up '
'graph execution, please convert the data '
'beforehand into one NumPy array to speed up '
'training.' % var.uid)
if isinstance(data, cntk_py.NDArrayView):
return cntk_py.Value(data)
if isinstance(data, np.ndarray):
# The outermost axis has to be Python list. If the user passes a
# full minibatch as one NumPy array, we have to convert it.
if data.dtype == object:
raise ValueError('dtype object is not supported. If this is a '
'batch of sequences, you need to pass them as a '
'pure-Python list of NumPy arrays')
if seq_starts:
data = list(np.atleast_1d(data))
else:
data = Value._as_best_data_type(var, data)
ndav = NDArrayView.from_data(data, device)
return cntk_py.Value(ndav)
if isinstance(data, sparse.csr_matrix):
if seq_starts:
raise ValueError('seq_starts are not supported with data specified as a single '
'SciPy sparse CSR matrix. Please provide a list instead')
else:
data = Value._as_best_data_type(var, data)
ndav = NDArrayView.from_data(data, device)
return cntk_py.Value(ndav)
if not isinstance(data, list):
raise ValueError('batch must be a list of NumPy arrays or '
'SciPy CSR matrices')
# NDArrayViews are all created on CPU. The Value object later then will
# move it to the requested device.
# As Value will later create copies anyways, we do not create copies in
# NDArrayView itself. Because of that, we need to keep around the
# instances _as_best_data_type() until we have passed them to
# Value_create() where it will be copied further.
data = [Value._as_best_data_type(var, sample) for sample in data]
device = device or use_default_device()
borrow = device.type() == DeviceKind.CPU
list_of_ndavs = [NDArrayView.from_data(sample, device=cpu(),
borrow=borrow)
for sample in data]
from cntk.internal import sanitize_shape
value = cntk_py.Value_create(
sanitize_shape(var.shape),
list_of_ndavs,
seq_starts or [],
device,
read_only,
True) # always create a copy in Value
return value