def getplan(self, size, dtype):
"""
Identify the plan for chunking along each dimension
"""
from numpy import dtype as gettype
plan = ones(len(self.value.shape), dtype=int)
if isinstance(size, tuple):
plan[self.value.axes] = size
else:
# convert from megabytes
size *= 1000.0
# calculate from dtype
element_size = gettype(dtype).itemsize
nelements = prod(self.value.shape)
total_size = nelements * element_size
moving_value_shapes = self.value.shape[self.value.mask]
if size <= element_size:
return moving_value_shapes
remaining_size = 1.0 * total_size
nchunks = ones(len(moving_value_shapes))
for (i, s) in enumerate(moving_value_shapes):
min_chunk_size = remaining_size / s
if min_chunk_size >= size:
nchunks[i] = s
remaining_size = min_chunk_size
continue
else:
nchunks[i] = ceil(remaining_size / size)
break
plan[self.value.axes] = nchunks
return plan
def getplan(self, size, dtype):
"""
Identify the plan for chunking along each dimension
"""
from numpy import dtype as gettype
plan = ones(len(self.value.shape), dtype=int)
if isinstance(size, tuple):
plan[self.value.axes] = size
else:
# convert from megabytes
size *= 1000.0
# calculate from dtype
element_size = gettype(dtype).itemsize
nelements = prod(self.value.shape)
total_size = nelements * element_size
moving_value_shapes = self.value.shape[self.value.mask]
if size <= element_size:
return moving_value_shapes
remaining_size = 1.0*total_size
nchunks = ones(len(moving_value_shapes))
for (i, s) in enumerate(moving_value_shapes):
min_chunk_size = remaining_size/s
if min_chunk_size >= size:
nchunks[i] = s
remaining_size = min_chunk_size
continue
else:
nchunks[i] = ceil(remaining_size/size)
break
plan[self.value.axes] = nchunks
return plan
def getplan(self, size="150", axes=None, padding=None):
"""
Identify a plan for chunking values along each dimension.
Generates an ndarray with the size (in number of elements) of chunks
in each dimension. If provided, will estimate chunks for only a
subset of axes, leaving all others to the full size of the axis.
Parameters
----------
size : string or tuple
If str, the average size (in KB) of the chunks in all value dimensions.
If int/tuple, an explicit specification of the number chunks in
each moving value dimension.
axes : tuple, optional, default=None
One or more axes to estimate chunks for, if provided any
other axes will use one chunk.
padding : tuple or int, option, default=None
Size over overlapping padding between chunks in each dimension.
If tuple, specifies padding along each chunked dimension; if int,
all dimensions use same padding; if None, no padding
"""
from numpy import dtype as gettype
# initialize with all elements in one chunk
plan = self.vshape
# check for subset of axes
if axes is None:
if isinstance(size, str):
axes = arange(len(self.vshape))
else:
axes = arange(len(size))
else:
axes = asarray(axes, 'int')
# set padding
pad = array(len(self.vshape)*[0, ])
if padding is not None:
pad[axes] = padding
# set the plan
if isinstance(size, tuple):
plan[axes] = size
elif isinstance(size, str):
# convert from kilobytes
size = 1000.0 * float(size)
# calculate from dtype
elsize = gettype(self.dtype).itemsize
nelements = prod(self.vshape)
dims = self.vshape[self.vmask(axes)]
if size <= elsize:
s = ones(len(axes))
else:
remsize = 1.0 * nelements * elsize
s = []
for (i, d) in enumerate(dims):
minsize = remsize/d
if minsize >= size:
s.append(1)
remsize = minsize
continue
else:
s.append(min(d, floor(size/minsize)))
s[i+1:] = plan[i+1:]
break
plan[axes] = s
else:
raise ValueError("Chunk size not understood, must be tuple or int")
return plan, pad
def getplan(self, size="150", axes=None, padding=None):
"""
Identify a plan for chunking values along each dimension.
Generates an ndarray with the size (in number of elements) of chunks
in each dimension. If provided, will estimate chunks for only a
subset of axes, leaving all others to the full size of the axis.
Parameters
----------
size : string or tuple
If str, the average size (in MB) of the chunks in all value dimensions.
If int/tuple, an explicit specification of the number chunks in
each moving value dimension.
axes : tuple, optional, default=None
One or more axes to estimate chunks for, if provided any
other axes will use one chunk.
padding : tuple or int, option, default=None
Size over overlapping padding between chunks in each dimension.
If tuple, specifies padding along each chunked dimension; if int,
all dimensions use same padding; if None, no padding
"""
from numpy import dtype as gettype
# initialize with all elements in one chunk
plan = self.vshape
# check for subset of axes
if axes is None:
if isinstance(size, str):
axes = arange(len(self.vshape))
else:
axes = arange(len(size))
else:
axes = asarray(axes, 'int')
# set padding
pad = array(len(self.vshape) * [
0,
])
if padding is not None:
pad[axes] = padding
# set the plan
if isinstance(size, tuple):
plan[axes] = size
elif isinstance(size, str):
# convert from megabytes
size = 1000.0 * float(size)
# calculate from dtype
elsize = gettype(self.dtype).itemsize
nelements = prod(self.vshape)
dims = self.vshape[self.vmask(axes)]
if size <= elsize:
s = ones(len(axes))
else:
remsize = 1.0 * nelements * elsize
s = []
for (i, d) in enumerate(dims):
minsize = remsize / d
if minsize >= size:
s.append(1)
remsize = minsize
continue
else:
s.append(min(d, floor(size / minsize)))
s[i + 1:] = plan[i + 1:]
break
plan[axes] = s
else:
raise ValueError("Chunk size not understood, must be tuple or int")
return plan, pad
