1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
# -*- coding: utf-8 -*-
# Copyright 2007-2023 The HyperSpy developers
#
# This file is part of RosettaSciIO.
#
# RosettaSciIO is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# RosettaSciIO is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with RosettaSciIO. If not, see <https://www.gnu.org/licenses/#GPL>.
import os
import dask.array as da
import numpy as np
def get_chunk_slice(
shape,
chunks="auto",
block_size_limit=None,
dtype=None,
):
"""
Get chunk slices for the :func:`rsciio.utils.distributed.slice_memmap` function.
Takes a shape and chunks and returns a dask array of the slices to be used with the
:func:`rsciio.utils.distributed.slice_memmap` function. This is useful for loading data from a memmaped file in a
distributed manner.
Parameters
----------
shape : tuple
Shape of the data.
chunks : tuple, optional
Chunk shape. The default is "auto".
block_size_limit : int, optional
Maximum size of a block in bytes. The default is None. This is passed
to the :py:func:`dask.array.core.normalize_chunks` function when chunks == "auto".
dtype : numpy.dtype, optional
Data type. The default is None. This is passed to the
:py:func:`dask.array.core.normalize_chunks` function when chunks == "auto".
Returns
-------
dask.array.Array
Dask array of the slices.
tuple
Tuple of the chunks.
"""
chunks = da.core.normalize_chunks(
chunks=chunks, shape=shape, limit=block_size_limit, dtype=dtype
)
chunks_shape = tuple([len(c) for c in chunks])
slices = np.empty(
shape=chunks_shape + (len(chunks_shape), 2),
dtype=int,
)
for ind in np.ndindex(chunks_shape):
current_chunk = [chunk[i] for i, chunk in zip(ind, chunks)]
starts = [int(np.sum(chunk[:i])) for i, chunk in zip(ind, chunks)]
stops = [s + c for s, c in zip(starts, current_chunk)]
slices[ind] = [[start, stop] for start, stop in zip(starts, stops)]
return da.from_array(slices, chunks=(1,) * len(shape) + slices.shape[-2:]), chunks
def slice_memmap(slices, file, dtypes, shape, key=None, **kwargs):
"""
Slice a memory mapped file using a tuple of slices.
This is useful for loading data from a memory mapped file in a distributed manner. The function
first creates a memory mapped array of the entire dataset and then uses the ``slices`` to slice the
memory mapped array. The slices can be used to build a ``dask`` array as each slice translates to one
chunk for the ``dask`` array.
Parameters
----------
slices : array-like of int
An array of the slices to use. The dimensions of the array should be
(n,2) where n is the number of dimensions of the data. The first column
is the start of the slice and the second column is the stop of the slice.
file : str
Path to the file.
dtypes : numpy.dtype
Data type of the data for :class:`numpy.memmap` function.
shape : tuple
Shape of the entire dataset. Passed to the :class:`numpy.memmap` function.
key : None, str
For structured dtype only. Specify the key of the structured dtype to use.
**kwargs : dict
Additional keyword arguments to pass to the :class:`numpy.memmap` function.
Returns
-------
numpy.ndarray
Array of the data from the memory mapped file sliced using the provided slice.
"""
slices_ = np.squeeze(slices)[()]
data = np.memmap(file, dtypes, shape=shape, **kwargs)
if key is not None:
data = data[key]
slices_ = tuple([slice(s[0], s[1]) for s in slices_])
return data[slices_]
def memmap_distributed(
filename,
dtype,
offset=0,
shape=None,
order="C",
chunks="auto",
block_size_limit=None,
key=None,
):
"""
Drop in replacement for py:func:`numpy.memmap` allowing for distributed
loading of data.
This always loads the data using dask which can be beneficial in many
cases, but may not be ideal in others. The ``chunks`` and ``block_size_limit``
are for describing an ideal chunk shape and size as defined using the
:func:`dask.array.core.normalize_chunks` function.
Parameters
----------
filename : str
Path to the file.
dtype : numpy.dtype
Data type of the data for memmap function.
offset : int, optional
Offset in bytes. The default is 0.
shape : tuple, optional
Shape of the data to be read. The default is None.
order : str, optional
Order of the data. The default is "C" see :class:`numpy.memmap` for more details.
chunks : tuple, optional
Chunk shape. The default is "auto".
block_size_limit : int, optional
Maximum size of a block in bytes. The default is None.
key : None, str
For structured dtype only. Specify the key of the structured dtype to use.
Returns
-------
dask.array.Array
Dask array of the data from the memmaped file and with the specified chunks.
Notes
-----
Currently :func:`dask.array.map_blocks` does not allow for multiple outputs.
As a result, in case of structured dtype, the key of the structured dtype need
to be specified.
For example: with dtype = (("data", int, (128, 128)), ("sec", "<u4", 512)),
"data" or "sec" will need to be specified.
"""
if dtype.names is not None:
# Structured dtype
array_dtype = dtype[key].base
sub_array_shape = dtype[key].shape
else:
array_dtype = dtype.base
sub_array_shape = dtype.shape
if shape is None:
unit_size = np.dtype(dtype).itemsize
shape = int(os.path.getsize(filename) / unit_size)
if not isinstance(shape, tuple):
shape = (shape,)
# Separates slices into appropriately sized chunks.
chunked_slices, data_chunks = get_chunk_slice(
shape=shape + sub_array_shape,
chunks=chunks,
block_size_limit=block_size_limit,
dtype=array_dtype,
)
num_dim = len(shape)
data = da.map_blocks(
slice_memmap,
chunked_slices,
file=filename,
dtype=array_dtype,
shape=shape,
order=order,
mode="r",
dtypes=dtype,
offset=offset,
chunks=data_chunks,
drop_axis=(
num_dim,
num_dim + 1,
), # Dask 2021.10.0 minimum to use negative indexing
key=key,
)
return data
|
