Intelegentny_Pszczelarz/.venv/Lib/site-packages/h5py/_hl/dims.py

189 lines
5.3 KiB
Python
Raw Normal View History

2023-06-19 00:49:18 +02:00
# This file is part of h5py, a Python interface to the HDF5 library.
#
# http://www.h5py.org
#
# Copyright 2008-2013 Andrew Collette and contributors
#
# License: Standard 3-clause BSD; see "license.txt" for full license terms
# and contributor agreement.
"""
Implements support for HDF5 dimension scales.
"""
import warnings
from .. import h5ds
from ..h5py_warnings import H5pyDeprecationWarning
from . import base
from .base import phil, with_phil
from .dataset import Dataset
class DimensionProxy(base.CommonStateObject):
"""
Represents an HDF5 "dimension".
"""
@property
@with_phil
def label(self):
""" Get or set the dimension scale label """
#return h5ds.get_label(self._id, self._dimension)
# Produces a segfault for a non-existent label (Fixed in hdf5-1.8.8).
# Here is a workaround:
try:
dset = Dataset(self._id)
return dset.attrs['DIMENSION_LABELS'][self._dimension]
except (KeyError, IndexError):
return ''
@label.setter
@with_phil
def label(self, val):
# pylint: disable=missing-docstring
h5ds.set_label(self._id, self._dimension, self._e(val))
@with_phil
def __init__(self, id_, dimension):
self._id = id_
self._dimension = dimension
@with_phil
def __hash__(self):
return hash((type(self), self._id, self._dimension))
@with_phil
def __eq__(self, other):
return hash(self) == hash(other)
@with_phil
def __iter__(self):
for k in self.keys():
yield k
@with_phil
def __len__(self):
return h5ds.get_num_scales(self._id, self._dimension)
@with_phil
def __getitem__(self, item):
if isinstance(item, int):
scales = []
h5ds.iterate(self._id, self._dimension, scales.append, 0)
return Dataset(scales[item])
else:
def f(dsid):
""" Iterate over scales to find a matching name """
if h5ds.get_scale_name(dsid) == self._e(item):
return dsid
res = h5ds.iterate(self._id, self._dimension, f, 0)
if res is None:
raise KeyError(item)
return Dataset(res)
def attach_scale(self, dset):
""" Attach a scale to this dimension.
Provide the Dataset of the scale you would like to attach.
"""
with phil:
h5ds.attach_scale(self._id, dset.id, self._dimension)
def detach_scale(self, dset):
""" Remove a scale from this dimension.
Provide the Dataset of the scale you would like to remove.
"""
with phil:
h5ds.detach_scale(self._id, dset.id, self._dimension)
def items(self):
""" Get a list of (name, Dataset) pairs with all scales on this
dimension.
"""
with phil:
scales = []
# H5DSiterate raises an error if there are no dimension scales,
# rather than iterating 0 times. See #483.
if len(self) > 0:
h5ds.iterate(self._id, self._dimension, scales.append, 0)
return [
(self._d(h5ds.get_scale_name(x)), Dataset(x))
for x in scales
]
def keys(self):
""" Get a list of names for the scales on this dimension. """
with phil:
return [key for (key, _) in self.items()]
def values(self):
""" Get a list of Dataset for scales on this dimension. """
with phil:
return [val for (_, val) in self.items()]
@with_phil
def __repr__(self):
if not self._id:
return "<Dimension of closed HDF5 dataset>"
return ('<"%s" dimension %d of HDF5 dataset at %s>'
% (self.label, self._dimension, id(self._id)))
class DimensionManager(base.CommonStateObject):
"""
Represents a collection of dimension associated with a dataset.
Like AttributeManager, an instance of this class is returned when
accessing the ".dims" property on a Dataset.
"""
@with_phil
def __init__(self, parent):
""" Private constructor.
"""
self._id = parent.id
@with_phil
def __getitem__(self, index):
""" Return a Dimension object
"""
if index > len(self) - 1:
raise IndexError('Index out of range')
return DimensionProxy(self._id, index)
@with_phil
def __len__(self):
""" Number of dimensions associated with the dataset. """
return self._id.rank
@with_phil
def __iter__(self):
""" Iterate over the dimensions. """
for i in range(len(self)):
yield self[i]
@with_phil
def __repr__(self):
if not self._id:
return "<Dimensions of closed HDF5 dataset>"
return "<Dimensions of HDF5 object at %s>" % id(self._id)
def create_scale(self, dset, name=''):
""" Create a new dimension, from an initial scale.
Provide the dataset and a name for the scale.
"""
warnings.warn("other_ds.dims.create_scale(ds, name) is deprecated. "
"Use ds.make_scale(name) instead.",
H5pyDeprecationWarning, stacklevel=2,
)
dset.make_scale(name)