from builtins import range, super
import ctypes
import logging
import textwrap
from ctypes import (addressof, byref, cast, c_double, c_float, c_int8, c_int16,
c_int32, c_int64, c_void_p, POINTER)
from ..tecutil import _tecutil, _tecutil_connector
from ..constant import *
from ..exception import *
from .. import session, tecutil
from . import operate
log = logging.getLogger(__name__)
[docs]
class Array(c_void_p):
"""Low-level accessor for underlying data within a `Dataset <tecplot.data.Dataset>`.
This object exposes a list-like interface to the underlying data array.
Using it, values can be directly queried and modified. After any
modification to the data, the Tecplot Engine will have to be notified of
the change. This notification will happen automatically in most cases, but
can be turned off using the `suspend context <tecplot.session.suspend>` for
a significant performance increase on large data sets.
Accessing values within an `Array` is done through the standard ``[]``
syntax::
>>> print(array[3])
3.1415
The numbers passed are interpreted just like Python's built-in
:py:class:`slice` object::
>>> # print the values at indices: 5, 7, 9
>>> print(array[5:10:2])
[1.0, 1.0, 1.0]
Elements within an array can be manipulated in-place with the assignment
operator::
>>> array[3] = 5.0
>>> print(array[3])
5.0
Element-by-element access is *not* guaranteed to be performant and users
should avoid writing loops over indices in Python. Instead, whole arrays
should be used. This will effectively push the loop down to the underlying
native library and will be much faster in virtually all cases.
Consider this array of 10k elements::
>>> ds = frame.create_dataset('Dataset', ['x'])
>>> zn = ds.add_ordered_zone('Zone', 10000)
>>> array = zn.values('x')
The following loop, which takes the sine of all values in the array will
require several Python function calls per element which is a tremendous
overhead::
>>> import math
>>> for i in range(len(ar)):
... ar[i] = math.sin(ar[i])
An immediate improvement on this can be made by looping over the elements
in Python only when reading the values, but assigning them using the
whole array. This will be several times faster for even modest arrays::
>>> ar[:] = [math.sin(x) for x in ar]
But there is still a large performance penalty for looping over elements
directly in Python and PyTecplot supports two solutions for large arrays:
`tecplot.data.operate.execute_equation` and `Array.as_numpy_array()`.
Please refer to these for details. Continuing with the example above, we
could accomplish the same thing with either of the following using
`execute_equation()` (assuming the array is identified by the first zone,
first variable)::
>>> from tecplot.data.operate import execute_equation
>>> execute_equation('V1 = SIN(V1)', zones=[dataset.zone(0)])
or by using the `numpy` library::
>>> import numpy as np
>>> ar[:] = np.sin(ar[:])
In both of these cases, the calculation of the sine and loop over elements
is pushed to the low level library and is much faster. Note that only the
`execute_equation()` solution does the calculation within Tecplot and does
not require the data to copied out to Python so it will typically be the
fastest option.
.. note::
When modifying data using this class, it may be necessary to update the
range of any associated contouring with a call to
`ContourLevels.reset()` or similar. This will ensure that the total
range of the new values is presented in the plot.
"""
def __init__(self, zone, variable):
self.zone = zone
self.variable = variable
super().__init__(self._native_reference())
@property
def _cache(self):
if _tecutil_connector.suspended:
_tecutil_connector._delete_caches.append(self._delete_cache)
return True
else:
return False
def _delete_cache(self):
attrs = '''
_rnr _wnr _rrp _wrp
_location
_len
_data_type
'''.split()
for attr in attrs:
try:
delattr(self, attr)
except AttributeError:
pass
def _native_reference(self, writable=False):
args = (self.zone.dataset.uid, self.zone.index + 1, self.variable.index + 1)
if writable:
if self._cache:
if not hasattr(self, '_wnr'):
with tecutil.lock():
self._wnr = _tecutil.DataValueGetWritableNativeRefByUniqueID(*args)
return self._wnr
else:
with tecutil.lock():
return _tecutil.DataValueGetWritableNativeRefByUniqueID(*args)
else:
if self._cache:
if not hasattr(self, '_rnr'):
with tecutil.lock():
self._rnr = _tecutil.DataValueGetReadableNativeRefByUniqueID(*args)
return self._rnr
else:
with tecutil.lock():
return _tecutil.DataValueGetReadableNativeRefByUniqueID(*args)
@tecutil.lock()
def _raw_pointer(self, writable=False):
if _tecutil_connector.connected:
msg = 'raw pointer access only available in batch-mode'
raise TecplotLogicError(msg)
elif writable:
ref = self._native_reference(writable=True)
_tecutil.handle.tecUtilDataValueGetWritableRawPtrByRef.restype = \
POINTER(self.c_type)
wrp = _tecutil.DataValueGetWritableRawPtrByRef(ref)
if self._cache:
if not hasattr(self, '_wrp'):
self._wrp = wrp
return self._wrp
else:
return wrp
else:
_tecutil.handle.tecUtilDataValueGetReadableRawPtrByRef.restype = \
POINTER(self.c_type)
rrp = _tecutil.DataValueGetReadableRawPtrByRef(self)
if self._cache:
if not hasattr(self, '_rrp'):
self._rrp = rrp
return self._rrp
else:
return rrp
def __eq__(self, other):
self_addr = addressof(cast(self, POINTER(c_int64)).contents)
other_addr = addressof(cast(other, POINTER(c_int64)).contents)
return self_addr == other_addr
def __ne__(self, other):
return not (self == other)
def __len__(self):
"""The number of values in this array.
:rtype: `integer <int>`
Example showing size of ordered data::
>>> x = dataset.zone('Zone').values('X')
>>> print(x.shape)
(10, 10, 10)
>>> print(len(x))
1000
"""
if self._cache:
if not hasattr(self, '_len'):
self._len = _tecutil.DataValueGetCountByRef(self)
return self._len
else:
return _tecutil.DataValueGetCountByRef(self)
@property
def location(self):
"""`ValueLocation`: Data points location with respect to the elements.
Possible values are `ValueLocation.CellCentered` and
`ValueLocation.Nodal`. Example usage::
>>> print(dataset.zone(0).values('X').location)
ValueLocation.Nodal
"""
if self._cache:
if not hasattr(self, '_location'):
self._location = _tecutil.DataValueGetLocation(
self.zone.index + 1, self.variable.index + 1)
return self._location
else:
return _tecutil.DataValueGetLocation(self.zone.index + 1,
self.variable.index + 1)
@property
def shape(self):
"""`tuple` of `floats <float>`: ``(i, j, k)`` shape for this array.
This is defined by the parent zone and can be used to reshape arrays.
The following example assumes 32-bit floating point array and copies
the Tecplot-owned ``data`` into the `numpy`-owned ``array``::
>>> import numpy as np
>>> data = dataset.zone('Zone').values('X')
>>> array = np.empty(data.shape, dtype=np.float32)
>>> arr_ptr = array.ctypes.data_as(POINTER(data.c_type))
>>> memmove(arr_ptr, data.copy(), sizeof(data.c_type) * len(data))
The data array presented is normally one-dimensional. For ordered data,
you may wish to reshape the array indexing according to the
dimensionality given by the ``shape`` attribute:
.. code-block:: python
:emphasize-lines: 30
import numpy as np
import tecplot as tp
frame = tp.active_frame()
dataset = frame.create_dataset('Dataset', ['X'])
zone = dataset.add_ordered_zone('Zone', shape=(3,3,3))
'''
the following will print:
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
'''
x = np.array(zone.values('X')[:])
print(x)
'''
the following will print:
[[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]
[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]
[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]]
'''
x.shape = zone.values('X').shape
print(x)
"""
if self.zone.zone_type is ZoneType.Ordered:
array_shape = tuple(i for i in self.zone._shape if i > 1)
else:
array_shape = (self.zone.num_points,)
if self.location is ValueLocation.CellCentered:
array_shape = tuple(i - 1 for i in array_shape if i > 2)
if not array_shape:
array_shape = (1,)
return array_shape
@property
def c_type(self):
"""`ctypes` compatible data type of this array.
This is the `ctypes` equivalent of `Array.data_type` and will return
one of the following:
* `ctypes.c_float`
* `ctypes.c_double`
* `ctypes.c_int`
* `ctypes.c_int16`
* `ctypes.c_int8`
and can be used to create a `ctypes` array to store a copy of the
data:
.. code-block:: python
import tecplot as tp
frame = tp.active_frame()
dataset = frame.create_dataset('Dataset', ['x'])
dataset.add_ordered_zone('Zone', (3,3,3))
x = dataset.zone('Zone').values('x')
# allocate array using Python's ctypes
x_array = (x.c_type * len(x))()
# copy values from Dataset into ctypes array
x_array[:] = x[:]
"""
_ctypes = {
FieldDataType.Float: ctypes.c_float,
FieldDataType.Double: ctypes.c_double,
FieldDataType.Int32: ctypes.c_int32,
FieldDataType.Int16: ctypes.c_int16,
FieldDataType.Byte: ctypes.c_int8}
return _ctypes[self.data_type]
@property
def data_type(self):
"""`FieldDataType`: Indicating the underlying value type of this array.
Example usage::
>>> print(dataset.zone('Zone').values('X').data_type)
FieldDataType.Float
"""
return _tecutil.DataValueGetRefType(self)
@tecutil.lock()
def as_ctypes_array(self, offset=0, size=None, copy=True):
"""Present the underlying data array as a `ctypes.Array`.
If the **copy** parameter is `False`, this method will attempt to return
an array pointing to the actual data stored in the Tecplot Engine. This
will fail in connected mode or if the loader does not support
immediate loading of the entire array into memory. Care should be taken
to ensure the validity of the pointers to the data.
Parameters:
offset (`int`, optional): Zero-based offset into the array. This
will be the starting point of the resulting data. (default: 0)
size (`int`, optional): Number of elements in the resulting array.
The default (a value of `None`) is to go to the end of the
data.
copy (`bool`, optional): Copy the data out from the Tecplot Engine.
If `False`, an attempt is made to point to the underlying raw
data and an exception is thrown on error. (default: `True`)
Returns:
`ctypes.Array`
Example usage::
>>> x = dataset.zone('Zone').values('X').as_ctypes_array()
"""
size = (len(self) - offset) if size is None else size
CArray = (self.c_type * size)
if copy:
arr = CArray()
_tecutil.DataValueArrayGetByRef(self, offset + 1, size, arr)
return arr
else:
ptr = self._raw_pointer(True)
if offset:
vptr = ctypes.cast(ctypes.pointer(ptr),
ctypes.POINTER(ctypes.c_void_p))
vptr.contents.value += ctypes.sizeof(ptr._type_) * offset
ptr_addr = ctypes.addressof(ptr.contents)
arr = CArray.from_address(ptr_addr)
return arr
[docs]
def as_numpy_array(self, offset=0, size=None, copy=True):
"""Present the underlying data array as a `numpy.ndarray`.
If the **copy** parameter is `False`, this method will attempt to return
an array pointing to the actual data stored in the Tecplot Engine. This
will fail in connected mode or if the loader does not support
immediate loading of the entire array into memory. Care should be taken
to ensure the validity of the pointers to the data.
Parameters:
offset (`int`, optional): Zero-based offset into the array. This
will be the starting point of the resulting data. (default: 0)
size (`int`, optional): Number of elements in the resulting array.
The default (a value of `None`) is to go to the end of the
data.
copy (`bool`, optional): Copy the data out from the Tecplot Engine.
If `False`, an attempt is made to point to the underlying raw
data and an exception is thrown on error. (default: `True`)
Returns:
`numpy.ndarray`
Example usage::
>>> x = dataset.zone('Zone').values('X').as_numpy_array()
"""
import numpy as np
carr = self.as_ctypes_array(offset, size, copy)
return np.ctypeslib.as_array(carr)
[docs]
@tecutil.lock()
def copy(self, offset=0, size=None):
"""Copy the whole or part of the array into a ctypes array.
Parameters:
offset (`int`, optional): Zero-based offset for starting
index to copy. (default: 0)
size (`int`, optional): Number of values to copy into the
resulting array. A value of `None` will copy to the end of the
array. (default: `None`)
Here we will copy out chunks of the data, do some operation and set the
values back into the dataset:
.. code-block:: python
import tecplot as tp
tp.new_layout()
frame = tp.active_frame()
dataset = frame.create_dataset('Dataset', ['x'])
dataset.add_ordered_zone('Zone', (2, 2, 2))
x = dataset.zone('Zone').values('x')
# loop over array copying out 4 values at a time
for i, offset in enumerate(range(0, len(x), 4)):
x_array = x.copy(offset, 4)
x_array[:] = [i] * 4
x[offset:offset + 4] = x_array
# will print: [0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0]
print(x[:])
"""
try:
return self.as_numpy_array(offset, size, copy=True)
except ImportError:
msg = textwrap.dedent('''\
Falling back to using basic Python for data operations.
If installed, PyTecplot will make use of Numpy where
appropriate for significant performance gains.
''')
log.warning(msg)
return self.as_ctypes_array(offset, size, copy=True)
def _slice_range(self, s):
start = s.start or 0
if start < 0:
start += len(self)
stop = s.stop or len(self)
if stop < 0:
stop += len(self)
step = s.step or 1
return range(start, stop, step)
def __getitem__(self, i):
if not isinstance(i, slice):
return _tecutil.DataValueGetByRef(self, i + 1)
else:
s = self._slice_range(i)
if s.step > 1:
# i is a non-contiguous slice
return [_tecutil.DataValueGetByRef(self, ii + 1) for ii in s]
else:
# i is a contiguous slice
return self.copy(s.start, s.stop - s.start)
@tecutil.lock()
def __setitem__(self, i, val):
"""
Developers note:
This method avoids using raw pointers which may not be available
for different reasons -- backing data does not support it, limited
memory resources or we are in connected mode for example.
"""
if not isinstance(i, slice):
# i is an index
ref = self._native_reference(True)
_tecutil.DataValueSetByRef(ref, i + 1, val)
session.data_altered(self.zone, self.variable, i)
elif isinstance(val, Array) and val == self:
# self assignment no-op
return
else:
if callable(getattr(val, 'ravel', None)):
val = val.ravel()
s = self._slice_range(i)
if len(s) != len(val):
msg = 'Array length mismatch {} != {}'
raise TecplotIndexError(msg.format(len(s), len(val)))
if s.step > 1:
# i is a non-contiguous slice
# the following works, but is slow!
for a, b in enumerate(s):
self[b] = val[a]
else:
# i is a contiguous slice
offset = s.start
size = s.stop - s.start
if isinstance(val, Array):
if (
offset == 0 and
size in (None, len(self)) and
val.zone.dataset == self.zone.dataset
):
# copy whole array
if val.zone == self.zone:
eqn = 'V{0} = V{1}'.format(self.variable.index + 1,
val.variable.index + 1)
operate.execute_equation(eqn, self.zone)
else:
src_zone_idx = self.zone.index
tgt_zone_idx = val.zone.index
var_idx = val.variable.index
if not _tecutil.DataValueCopy(src_zone_idx + 1,
tgt_zone_idx + 1,
var_idx + 1):
raise TecplotSystemError()
return
else:
# either sub array or different datasets
val = val.as_ctypes_array(copy=True)
ctype = self.c_type
if isinstance(val, ctypes.Array) and val._type_ == ctype:
arr = val
else:
# coerce val to a ctypes array, using numpy if available
try:
import numpy as np
nparr = np.asarray(val, dtype=ctype)
ptarr = nparr.ctypes.data_as(POINTER(ctype))
ptaddr = addressof(ptarr.contents)
arr = (ctype * size).from_address(ptaddr)
except ImportError:
msg = textwrap.dedent('''\
Falling back to using basic Python for data
operations. If installed, PyTecplot will make use
of Numpy where appropriate for significant
performance gains.
''')
log.warning(msg)
arr = (ctype * size)(*val)
ref = self._native_reference(True)
_tecutil.DataValueArraySetByRef(ref, offset + 1, size, arr)
session.data_altered(self.zone, self.variable)
def __iter__(self):
self.current_index = -1
self.current_length = len(self)
return self
def __next__(self):
self.current_index += 1
if self.current_index < self.current_length:
return self.__getitem__(self.current_index)
else:
del self.current_index
del self.current_length
raise StopIteration
[docs]
def minmax(self):
"""Limits of the values stored in this array.
:rtype: `tuple` of `floats <float>`
This always returns `floats <float>` regardless of the underlying data
type::
>>> print(dataset.zone('Zone').values('x').minmax())
(0, 10)
"""
return _tecutil.DataValueGetMinMaxByRef(self)
[docs]
def min(self):
"""Lower bound of the values stored in this array.
:rtype: `float`
This always returns a `float` regardless of the underlying data type::
>>> print(dataset.zone('Zone').values('x').min())
0
"""
return self.minmax()[0]
[docs]
def max(self):
"""Upper bound of the values stored in this array.
:rtype: `float`
This always returns a `float` regardless of the underlying data type::
>>> print(dataset.zone('Zone').values('x').max())
10
"""
return self.minmax()[1]
@property
def shared_zones(self):
"""`list` of `Zones <data_access>`: All `Zones <data_access>` sharing this array.
Example usage::
>>> dataset.zone('My Zone').copy(share_variables=True)
>>> for z in dataset.zone('My Zone').values(0).shared_zones:
... print(z.index)
0
1
"""
indices = _tecutil.DataValueGetShareZoneSet(self.zone.index + 1,
self.variable.index + 1)
ret = [self.zone.dataset.zone(i) for i in indices]
indices.dealloc()
return ret
@property
def passive(self):
"""`bool`: An unallocated zone-variable combination.
Passive variables are placeholders where no data is defined for a zone
variable combination. Passive variables will always return zero when
queried:
.. code-block:: python
import tecplot as tp
ds = tp.active_page().add_frame().create_dataset('D', ['x','y'])
z = ds.add_ordered_zone('Z1', (3,))
assert not z.values(0).passive
"""
return _tecutil.DataValueIsPassive(self.zone.index + 1,
self.variable.index + 1)
