Coverage for /home/martinb/.local/share/virtualenvs/camcops/lib/python3.6/site-packages/numpy/lib/format.py : 12%

Hot-keys on this page
r m x p toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1"""
2Binary serialization
4NPY format
5==========
7A simple format for saving numpy arrays to disk with the full
8information about them.
10The ``.npy`` format is the standard binary file format in NumPy for
11persisting a *single* arbitrary NumPy array on disk. The format stores all
12of the shape and dtype information necessary to reconstruct the array
13correctly even on another machine with a different architecture.
14The format is designed to be as simple as possible while achieving
15its limited goals.
17The ``.npz`` format is the standard format for persisting *multiple* NumPy
18arrays on disk. A ``.npz`` file is a zip file containing multiple ``.npy``
19files, one for each array.
21Capabilities
22------------
24- Can represent all NumPy arrays including nested record arrays and
25 object arrays.
27- Represents the data in its native binary form.
29- Supports Fortran-contiguous arrays directly.
31- Stores all of the necessary information to reconstruct the array
32 including shape and dtype on a machine of a different
33 architecture. Both little-endian and big-endian arrays are
34 supported, and a file with little-endian numbers will yield
35 a little-endian array on any machine reading the file. The
36 types are described in terms of their actual sizes. For example,
37 if a machine with a 64-bit C "long int" writes out an array with
38 "long ints", a reading machine with 32-bit C "long ints" will yield
39 an array with 64-bit integers.
41- Is straightforward to reverse engineer. Datasets often live longer than
42 the programs that created them. A competent developer should be
43 able to create a solution in their preferred programming language to
44 read most ``.npy`` files that he has been given without much
45 documentation.
47- Allows memory-mapping of the data. See `open_memmep`.
49- Can be read from a filelike stream object instead of an actual file.
51- Stores object arrays, i.e. arrays containing elements that are arbitrary
52 Python objects. Files with object arrays are not to be mmapable, but
53 can be read and written to disk.
55Limitations
56-----------
58- Arbitrary subclasses of numpy.ndarray are not completely preserved.
59 Subclasses will be accepted for writing, but only the array data will
60 be written out. A regular numpy.ndarray object will be created
61 upon reading the file.
63.. warning::
65 Due to limitations in the interpretation of structured dtypes, dtypes
66 with fields with empty names will have the names replaced by 'f0', 'f1',
67 etc. Such arrays will not round-trip through the format entirely
68 accurately. The data is intact; only the field names will differ. We are
69 working on a fix for this. This fix will not require a change in the
70 file format. The arrays with such structures can still be saved and
71 restored, and the correct dtype may be restored by using the
72 ``loadedarray.view(correct_dtype)`` method.
74File extensions
75---------------
77We recommend using the ``.npy`` and ``.npz`` extensions for files saved
78in this format. This is by no means a requirement; applications may wish
79to use these file formats but use an extension specific to the
80application. In the absence of an obvious alternative, however,
81we suggest using ``.npy`` and ``.npz``.
83Version numbering
84-----------------
86The version numbering of these formats is independent of NumPy version
87numbering. If the format is upgraded, the code in `numpy.io` will still
88be able to read and write Version 1.0 files.
90Format Version 1.0
91------------------
93The first 6 bytes are a magic string: exactly ``\\x93NUMPY``.
95The next 1 byte is an unsigned byte: the major version number of the file
96format, e.g. ``\\x01``.
98The next 1 byte is an unsigned byte: the minor version number of the file
99format, e.g. ``\\x00``. Note: the version of the file format is not tied
100to the version of the numpy package.
102The next 2 bytes form a little-endian unsigned short int: the length of
103the header data HEADER_LEN.
105The next HEADER_LEN bytes form the header data describing the array's
106format. It is an ASCII string which contains a Python literal expression
107of a dictionary. It is terminated by a newline (``\\n``) and padded with
108spaces (``\\x20``) to make the total of
109``len(magic string) + 2 + len(length) + HEADER_LEN`` be evenly divisible
110by 64 for alignment purposes.
112The dictionary contains three keys:
114 "descr" : dtype.descr
115 An object that can be passed as an argument to the `numpy.dtype`
116 constructor to create the array's dtype.
117 "fortran_order" : bool
118 Whether the array data is Fortran-contiguous or not. Since
119 Fortran-contiguous arrays are a common form of non-C-contiguity,
120 we allow them to be written directly to disk for efficiency.
121 "shape" : tuple of int
122 The shape of the array.
124For repeatability and readability, the dictionary keys are sorted in
125alphabetic order. This is for convenience only. A writer SHOULD implement
126this if possible. A reader MUST NOT depend on this.
128Following the header comes the array data. If the dtype contains Python
129objects (i.e. ``dtype.hasobject is True``), then the data is a Python
130pickle of the array. Otherwise the data is the contiguous (either C-
131or Fortran-, depending on ``fortran_order``) bytes of the array.
132Consumers can figure out the number of bytes by multiplying the number
133of elements given by the shape (noting that ``shape=()`` means there is
1341 element) by ``dtype.itemsize``.
136Format Version 2.0
137------------------
139The version 1.0 format only allowed the array header to have a total size of
14065535 bytes. This can be exceeded by structured arrays with a large number of
141columns. The version 2.0 format extends the header size to 4 GiB.
142`numpy.save` will automatically save in 2.0 format if the data requires it,
143else it will always use the more compatible 1.0 format.
145The description of the fourth element of the header therefore has become:
146"The next 4 bytes form a little-endian unsigned int: the length of the header
147data HEADER_LEN."
149Format Version 3.0
150------------------
152This version replaces the ASCII string (which in practice was latin1) with
153a utf8-encoded string, so supports structured types with any unicode field
154names.
156Notes
157-----
158The ``.npy`` format, including motivation for creating it and a comparison of
159alternatives, is described in the `"npy-format" NEP
160<https://www.numpy.org/neps/nep-0001-npy-format.html>`_, however details have
161evolved with time and this document is more current.
163"""
164import numpy
165import io
166import warnings
167from numpy.lib.utils import safe_eval
168from numpy.compat import (
169 isfileobj, os_fspath, pickle
170 )
173__all__ = []
176MAGIC_PREFIX = b'\x93NUMPY'
177MAGIC_LEN = len(MAGIC_PREFIX) + 2
178ARRAY_ALIGN = 64 # plausible values are powers of 2 between 16 and 4096
179BUFFER_SIZE = 2**18 # size of buffer for reading npz files in bytes
181# difference between version 1.0 and 2.0 is a 4 byte (I) header length
182# instead of 2 bytes (H) allowing storage of large structured arrays
183_header_size_info = {
184 (1, 0): ('<H', 'latin1'),
185 (2, 0): ('<I', 'latin1'),
186 (3, 0): ('<I', 'utf8'),
187}
190def _check_version(version):
191 if version not in [(1, 0), (2, 0), (3, 0), None]:
192 msg = "we only support format version (1,0), (2,0), and (3,0), not %s"
193 raise ValueError(msg % (version,))
195def magic(major, minor):
196 """ Return the magic string for the given file format version.
198 Parameters
199 ----------
200 major : int in [0, 255]
201 minor : int in [0, 255]
203 Returns
204 -------
205 magic : str
207 Raises
208 ------
209 ValueError if the version cannot be formatted.
210 """
211 if major < 0 or major > 255:
212 raise ValueError("major version must be 0 <= major < 256")
213 if minor < 0 or minor > 255:
214 raise ValueError("minor version must be 0 <= minor < 256")
215 return MAGIC_PREFIX + bytes([major, minor])
217def read_magic(fp):
218 """ Read the magic string to get the version of the file format.
220 Parameters
221 ----------
222 fp : filelike object
224 Returns
225 -------
226 major : int
227 minor : int
228 """
229 magic_str = _read_bytes(fp, MAGIC_LEN, "magic string")
230 if magic_str[:-2] != MAGIC_PREFIX:
231 msg = "the magic string is not correct; expected %r, got %r"
232 raise ValueError(msg % (MAGIC_PREFIX, magic_str[:-2]))
233 major, minor = magic_str[-2:]
234 return major, minor
236def _has_metadata(dt):
237 if dt.metadata is not None:
238 return True
239 elif dt.names is not None:
240 return any(_has_metadata(dt[k]) for k in dt.names)
241 elif dt.subdtype is not None:
242 return _has_metadata(dt.base)
243 else:
244 return False
246def dtype_to_descr(dtype):
247 """
248 Get a serializable descriptor from the dtype.
250 The .descr attribute of a dtype object cannot be round-tripped through
251 the dtype() constructor. Simple types, like dtype('float32'), have
252 a descr which looks like a record array with one field with '' as
253 a name. The dtype() constructor interprets this as a request to give
254 a default name. Instead, we construct descriptor that can be passed to
255 dtype().
257 Parameters
258 ----------
259 dtype : dtype
260 The dtype of the array that will be written to disk.
262 Returns
263 -------
264 descr : object
265 An object that can be passed to `numpy.dtype()` in order to
266 replicate the input dtype.
268 """
269 if _has_metadata(dtype):
270 warnings.warn("metadata on a dtype may be saved or ignored, but will "
271 "raise if saved when read. Use another form of storage.",
272 UserWarning, stacklevel=2)
273 if dtype.names is not None:
274 # This is a record array. The .descr is fine. XXX: parts of the
275 # record array with an empty name, like padding bytes, still get
276 # fiddled with. This needs to be fixed in the C implementation of
277 # dtype().
278 return dtype.descr
279 else:
280 return dtype.str
282def descr_to_dtype(descr):
283 '''
284 descr may be stored as dtype.descr, which is a list of
285 (name, format, [shape]) tuples where format may be a str or a tuple.
286 Offsets are not explicitly saved, rather empty fields with
287 name, format == '', '|Vn' are added as padding.
289 This function reverses the process, eliminating the empty padding fields.
290 '''
291 if isinstance(descr, str):
292 # No padding removal needed
293 return numpy.dtype(descr)
294 elif isinstance(descr, tuple):
295 # subtype, will always have a shape descr[1]
296 dt = descr_to_dtype(descr[0])
297 return numpy.dtype((dt, descr[1]))
299 titles = []
300 names = []
301 formats = []
302 offsets = []
303 offset = 0
304 for field in descr:
305 if len(field) == 2:
306 name, descr_str = field
307 dt = descr_to_dtype(descr_str)
308 else:
309 name, descr_str, shape = field
310 dt = numpy.dtype((descr_to_dtype(descr_str), shape))
312 # Ignore padding bytes, which will be void bytes with '' as name
313 # Once support for blank names is removed, only "if name == ''" needed)
314 is_pad = (name == '' and dt.type is numpy.void and dt.names is None)
315 if not is_pad:
316 title, name = name if isinstance(name, tuple) else (None, name)
317 titles.append(title)
318 names.append(name)
319 formats.append(dt)
320 offsets.append(offset)
321 offset += dt.itemsize
323 return numpy.dtype({'names': names, 'formats': formats, 'titles': titles,
324 'offsets': offsets, 'itemsize': offset})
326def header_data_from_array_1_0(array):
327 """ Get the dictionary of header metadata from a numpy.ndarray.
329 Parameters
330 ----------
331 array : numpy.ndarray
333 Returns
334 -------
335 d : dict
336 This has the appropriate entries for writing its string representation
337 to the header of the file.
338 """
339 d = {'shape': array.shape}
340 if array.flags.c_contiguous:
341 d['fortran_order'] = False
342 elif array.flags.f_contiguous:
343 d['fortran_order'] = True
344 else:
345 # Totally non-contiguous data. We will have to make it C-contiguous
346 # before writing. Note that we need to test for C_CONTIGUOUS first
347 # because a 1-D array is both C_CONTIGUOUS and F_CONTIGUOUS.
348 d['fortran_order'] = False
350 d['descr'] = dtype_to_descr(array.dtype)
351 return d
354def _wrap_header(header, version):
355 """
356 Takes a stringified header, and attaches the prefix and padding to it
357 """
358 import struct
359 assert version is not None
360 fmt, encoding = _header_size_info[version]
361 if not isinstance(header, bytes): # always true on python 3
362 header = header.encode(encoding)
363 hlen = len(header) + 1
364 padlen = ARRAY_ALIGN - ((MAGIC_LEN + struct.calcsize(fmt) + hlen) % ARRAY_ALIGN)
365 try:
366 header_prefix = magic(*version) + struct.pack(fmt, hlen + padlen)
367 except struct.error:
368 msg = "Header length {} too big for version={}".format(hlen, version)
369 raise ValueError(msg)
371 # Pad the header with spaces and a final newline such that the magic
372 # string, the header-length short and the header are aligned on a
373 # ARRAY_ALIGN byte boundary. This supports memory mapping of dtypes
374 # aligned up to ARRAY_ALIGN on systems like Linux where mmap()
375 # offset must be page-aligned (i.e. the beginning of the file).
376 return header_prefix + header + b' '*padlen + b'\n'
379def _wrap_header_guess_version(header):
380 """
381 Like `_wrap_header`, but chooses an appropriate version given the contents
382 """
383 try:
384 return _wrap_header(header, (1, 0))
385 except ValueError:
386 pass
388 try:
389 ret = _wrap_header(header, (2, 0))
390 except UnicodeEncodeError:
391 pass
392 else:
393 warnings.warn("Stored array in format 2.0. It can only be"
394 "read by NumPy >= 1.9", UserWarning, stacklevel=2)
395 return ret
397 header = _wrap_header(header, (3, 0))
398 warnings.warn("Stored array in format 3.0. It can only be "
399 "read by NumPy >= 1.17", UserWarning, stacklevel=2)
400 return header
403def _write_array_header(fp, d, version=None):
404 """ Write the header for an array and returns the version used
406 Parameters
407 ----------
408 fp : filelike object
409 d : dict
410 This has the appropriate entries for writing its string representation
411 to the header of the file.
412 version: tuple or None
413 None means use oldest that works
414 explicit version will raise a ValueError if the format does not
415 allow saving this data. Default: None
416 """
417 header = ["{"]
418 for key, value in sorted(d.items()):
419 # Need to use repr here, since we eval these when reading
420 header.append("'%s': %s, " % (key, repr(value)))
421 header.append("}")
422 header = "".join(header)
423 header = _filter_header(header)
424 if version is None:
425 header = _wrap_header_guess_version(header)
426 else:
427 header = _wrap_header(header, version)
428 fp.write(header)
430def write_array_header_1_0(fp, d):
431 """ Write the header for an array using the 1.0 format.
433 Parameters
434 ----------
435 fp : filelike object
436 d : dict
437 This has the appropriate entries for writing its string
438 representation to the header of the file.
439 """
440 _write_array_header(fp, d, (1, 0))
443def write_array_header_2_0(fp, d):
444 """ Write the header for an array using the 2.0 format.
445 The 2.0 format allows storing very large structured arrays.
447 .. versionadded:: 1.9.0
449 Parameters
450 ----------
451 fp : filelike object
452 d : dict
453 This has the appropriate entries for writing its string
454 representation to the header of the file.
455 """
456 _write_array_header(fp, d, (2, 0))
458def read_array_header_1_0(fp):
459 """
460 Read an array header from a filelike object using the 1.0 file format
461 version.
463 This will leave the file object located just after the header.
465 Parameters
466 ----------
467 fp : filelike object
468 A file object or something with a `.read()` method like a file.
470 Returns
471 -------
472 shape : tuple of int
473 The shape of the array.
474 fortran_order : bool
475 The array data will be written out directly if it is either
476 C-contiguous or Fortran-contiguous. Otherwise, it will be made
477 contiguous before writing it out.
478 dtype : dtype
479 The dtype of the file's data.
481 Raises
482 ------
483 ValueError
484 If the data is invalid.
486 """
487 return _read_array_header(fp, version=(1, 0))
489def read_array_header_2_0(fp):
490 """
491 Read an array header from a filelike object using the 2.0 file format
492 version.
494 This will leave the file object located just after the header.
496 .. versionadded:: 1.9.0
498 Parameters
499 ----------
500 fp : filelike object
501 A file object or something with a `.read()` method like a file.
503 Returns
504 -------
505 shape : tuple of int
506 The shape of the array.
507 fortran_order : bool
508 The array data will be written out directly if it is either
509 C-contiguous or Fortran-contiguous. Otherwise, it will be made
510 contiguous before writing it out.
511 dtype : dtype
512 The dtype of the file's data.
514 Raises
515 ------
516 ValueError
517 If the data is invalid.
519 """
520 return _read_array_header(fp, version=(2, 0))
523def _filter_header(s):
524 """Clean up 'L' in npz header ints.
526 Cleans up the 'L' in strings representing integers. Needed to allow npz
527 headers produced in Python2 to be read in Python3.
529 Parameters
530 ----------
531 s : string
532 Npy file header.
534 Returns
535 -------
536 header : str
537 Cleaned up header.
539 """
540 import tokenize
541 from io import StringIO
543 tokens = []
544 last_token_was_number = False
545 for token in tokenize.generate_tokens(StringIO(s).readline):
546 token_type = token[0]
547 token_string = token[1]
548 if (last_token_was_number and
549 token_type == tokenize.NAME and
550 token_string == "L"):
551 continue
552 else:
553 tokens.append(token)
554 last_token_was_number = (token_type == tokenize.NUMBER)
555 return tokenize.untokenize(tokens)
558def _read_array_header(fp, version):
559 """
560 see read_array_header_1_0
561 """
562 # Read an unsigned, little-endian short int which has the length of the
563 # header.
564 import struct
565 hinfo = _header_size_info.get(version)
566 if hinfo is None:
567 raise ValueError("Invalid version {!r}".format(version))
568 hlength_type, encoding = hinfo
570 hlength_str = _read_bytes(fp, struct.calcsize(hlength_type), "array header length")
571 header_length = struct.unpack(hlength_type, hlength_str)[0]
572 header = _read_bytes(fp, header_length, "array header")
573 header = header.decode(encoding)
575 # The header is a pretty-printed string representation of a literal
576 # Python dictionary with trailing newlines padded to a ARRAY_ALIGN byte
577 # boundary. The keys are strings.
578 # "shape" : tuple of int
579 # "fortran_order" : bool
580 # "descr" : dtype.descr
581 header = _filter_header(header)
582 try:
583 d = safe_eval(header)
584 except SyntaxError as e:
585 msg = "Cannot parse header: {!r}\nException: {!r}"
586 raise ValueError(msg.format(header, e))
587 if not isinstance(d, dict):
588 msg = "Header is not a dictionary: {!r}"
589 raise ValueError(msg.format(d))
590 keys = sorted(d.keys())
591 if keys != ['descr', 'fortran_order', 'shape']:
592 msg = "Header does not contain the correct keys: {!r}"
593 raise ValueError(msg.format(keys))
595 # Sanity-check the values.
596 if (not isinstance(d['shape'], tuple) or
597 not numpy.all([isinstance(x, int) for x in d['shape']])):
598 msg = "shape is not valid: {!r}"
599 raise ValueError(msg.format(d['shape']))
600 if not isinstance(d['fortran_order'], bool):
601 msg = "fortran_order is not a valid bool: {!r}"
602 raise ValueError(msg.format(d['fortran_order']))
603 try:
604 dtype = descr_to_dtype(d['descr'])
605 except TypeError:
606 msg = "descr is not a valid dtype descriptor: {!r}"
607 raise ValueError(msg.format(d['descr']))
609 return d['shape'], d['fortran_order'], dtype
611def write_array(fp, array, version=None, allow_pickle=True, pickle_kwargs=None):
612 """
613 Write an array to an NPY file, including a header.
615 If the array is neither C-contiguous nor Fortran-contiguous AND the
616 file_like object is not a real file object, this function will have to
617 copy data in memory.
619 Parameters
620 ----------
621 fp : file_like object
622 An open, writable file object, or similar object with a
623 ``.write()`` method.
624 array : ndarray
625 The array to write to disk.
626 version : (int, int) or None, optional
627 The version number of the format. None means use the oldest
628 supported version that is able to store the data. Default: None
629 allow_pickle : bool, optional
630 Whether to allow writing pickled data. Default: True
631 pickle_kwargs : dict, optional
632 Additional keyword arguments to pass to pickle.dump, excluding
633 'protocol'. These are only useful when pickling objects in object
634 arrays on Python 3 to Python 2 compatible format.
636 Raises
637 ------
638 ValueError
639 If the array cannot be persisted. This includes the case of
640 allow_pickle=False and array being an object array.
641 Various other errors
642 If the array contains Python objects as part of its dtype, the
643 process of pickling them may raise various errors if the objects
644 are not picklable.
646 """
647 _check_version(version)
648 _write_array_header(fp, header_data_from_array_1_0(array), version)
650 if array.itemsize == 0:
651 buffersize = 0
652 else:
653 # Set buffer size to 16 MiB to hide the Python loop overhead.
654 buffersize = max(16 * 1024 ** 2 // array.itemsize, 1)
656 if array.dtype.hasobject:
657 # We contain Python objects so we cannot write out the data
658 # directly. Instead, we will pickle it out
659 if not allow_pickle:
660 raise ValueError("Object arrays cannot be saved when "
661 "allow_pickle=False")
662 if pickle_kwargs is None:
663 pickle_kwargs = {}
664 pickle.dump(array, fp, protocol=3, **pickle_kwargs)
665 elif array.flags.f_contiguous and not array.flags.c_contiguous:
666 if isfileobj(fp):
667 array.T.tofile(fp)
668 else:
669 for chunk in numpy.nditer(
670 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
671 buffersize=buffersize, order='F'):
672 fp.write(chunk.tobytes('C'))
673 else:
674 if isfileobj(fp):
675 array.tofile(fp)
676 else:
677 for chunk in numpy.nditer(
678 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
679 buffersize=buffersize, order='C'):
680 fp.write(chunk.tobytes('C'))
683def read_array(fp, allow_pickle=False, pickle_kwargs=None):
684 """
685 Read an array from an NPY file.
687 Parameters
688 ----------
689 fp : file_like object
690 If this is not a real file object, then this may take extra memory
691 and time.
692 allow_pickle : bool, optional
693 Whether to allow writing pickled data. Default: False
695 .. versionchanged:: 1.16.3
696 Made default False in response to CVE-2019-6446.
698 pickle_kwargs : dict
699 Additional keyword arguments to pass to pickle.load. These are only
700 useful when loading object arrays saved on Python 2 when using
701 Python 3.
703 Returns
704 -------
705 array : ndarray
706 The array from the data on disk.
708 Raises
709 ------
710 ValueError
711 If the data is invalid, or allow_pickle=False and the file contains
712 an object array.
714 """
715 version = read_magic(fp)
716 _check_version(version)
717 shape, fortran_order, dtype = _read_array_header(fp, version)
718 if len(shape) == 0:
719 count = 1
720 else:
721 count = numpy.multiply.reduce(shape, dtype=numpy.int64)
723 # Now read the actual data.
724 if dtype.hasobject:
725 # The array contained Python objects. We need to unpickle the data.
726 if not allow_pickle:
727 raise ValueError("Object arrays cannot be loaded when "
728 "allow_pickle=False")
729 if pickle_kwargs is None:
730 pickle_kwargs = {}
731 try:
732 array = pickle.load(fp, **pickle_kwargs)
733 except UnicodeError as err:
734 # Friendlier error message
735 raise UnicodeError("Unpickling a python object failed: %r\n"
736 "You may need to pass the encoding= option "
737 "to numpy.load" % (err,))
738 else:
739 if isfileobj(fp):
740 # We can use the fast fromfile() function.
741 array = numpy.fromfile(fp, dtype=dtype, count=count)
742 else:
743 # This is not a real file. We have to read it the
744 # memory-intensive way.
745 # crc32 module fails on reads greater than 2 ** 32 bytes,
746 # breaking large reads from gzip streams. Chunk reads to
747 # BUFFER_SIZE bytes to avoid issue and reduce memory overhead
748 # of the read. In non-chunked case count < max_read_count, so
749 # only one read is performed.
751 # Use np.ndarray instead of np.empty since the latter does
752 # not correctly instantiate zero-width string dtypes; see
753 # https://github.com/numpy/numpy/pull/6430
754 array = numpy.ndarray(count, dtype=dtype)
756 if dtype.itemsize > 0:
757 # If dtype.itemsize == 0 then there's nothing more to read
758 max_read_count = BUFFER_SIZE // min(BUFFER_SIZE, dtype.itemsize)
760 for i in range(0, count, max_read_count):
761 read_count = min(max_read_count, count - i)
762 read_size = int(read_count * dtype.itemsize)
763 data = _read_bytes(fp, read_size, "array data")
764 array[i:i+read_count] = numpy.frombuffer(data, dtype=dtype,
765 count=read_count)
767 if fortran_order:
768 array.shape = shape[::-1]
769 array = array.transpose()
770 else:
771 array.shape = shape
773 return array
776def open_memmap(filename, mode='r+', dtype=None, shape=None,
777 fortran_order=False, version=None):
778 """
779 Open a .npy file as a memory-mapped array.
781 This may be used to read an existing file or create a new one.
783 Parameters
784 ----------
785 filename : str or path-like
786 The name of the file on disk. This may *not* be a file-like
787 object.
788 mode : str, optional
789 The mode in which to open the file; the default is 'r+'. In
790 addition to the standard file modes, 'c' is also accepted to mean
791 "copy on write." See `memmap` for the available mode strings.
792 dtype : data-type, optional
793 The data type of the array if we are creating a new file in "write"
794 mode, if not, `dtype` is ignored. The default value is None, which
795 results in a data-type of `float64`.
796 shape : tuple of int
797 The shape of the array if we are creating a new file in "write"
798 mode, in which case this parameter is required. Otherwise, this
799 parameter is ignored and is thus optional.
800 fortran_order : bool, optional
801 Whether the array should be Fortran-contiguous (True) or
802 C-contiguous (False, the default) if we are creating a new file in
803 "write" mode.
804 version : tuple of int (major, minor) or None
805 If the mode is a "write" mode, then this is the version of the file
806 format used to create the file. None means use the oldest
807 supported version that is able to store the data. Default: None
809 Returns
810 -------
811 marray : memmap
812 The memory-mapped array.
814 Raises
815 ------
816 ValueError
817 If the data or the mode is invalid.
818 IOError
819 If the file is not found or cannot be opened correctly.
821 See Also
822 --------
823 memmap
825 """
826 if isfileobj(filename):
827 raise ValueError("Filename must be a string or a path-like object."
828 " Memmap cannot use existing file handles.")
830 if 'w' in mode:
831 # We are creating the file, not reading it.
832 # Check if we ought to create the file.
833 _check_version(version)
834 # Ensure that the given dtype is an authentic dtype object rather
835 # than just something that can be interpreted as a dtype object.
836 dtype = numpy.dtype(dtype)
837 if dtype.hasobject:
838 msg = "Array can't be memory-mapped: Python objects in dtype."
839 raise ValueError(msg)
840 d = dict(
841 descr=dtype_to_descr(dtype),
842 fortran_order=fortran_order,
843 shape=shape,
844 )
845 # If we got here, then it should be safe to create the file.
846 with open(os_fspath(filename), mode+'b') as fp:
847 _write_array_header(fp, d, version)
848 offset = fp.tell()
849 else:
850 # Read the header of the file first.
851 with open(os_fspath(filename), 'rb') as fp:
852 version = read_magic(fp)
853 _check_version(version)
855 shape, fortran_order, dtype = _read_array_header(fp, version)
856 if dtype.hasobject:
857 msg = "Array can't be memory-mapped: Python objects in dtype."
858 raise ValueError(msg)
859 offset = fp.tell()
861 if fortran_order:
862 order = 'F'
863 else:
864 order = 'C'
866 # We need to change a write-only mode to a read-write mode since we've
867 # already written data to the file.
868 if mode == 'w+':
869 mode = 'r+'
871 marray = numpy.memmap(filename, dtype=dtype, shape=shape, order=order,
872 mode=mode, offset=offset)
874 return marray
877def _read_bytes(fp, size, error_template="ran out of data"):
878 """
879 Read from file-like object until size bytes are read.
880 Raises ValueError if not EOF is encountered before size bytes are read.
881 Non-blocking objects only supported if they derive from io objects.
883 Required as e.g. ZipExtFile in python 2.6 can return less data than
884 requested.
885 """
886 data = bytes()
887 while True:
888 # io files (default in python3) return None or raise on
889 # would-block, python2 file will truncate, probably nothing can be
890 # done about that. note that regular files can't be non-blocking
891 try:
892 r = fp.read(size - len(data))
893 data += r
894 if len(r) == 0 or len(data) == size:
895 break
896 except io.BlockingIOError:
897 pass
898 if len(data) != size:
899 msg = "EOF: reading %s, expected %d bytes got %d"
900 raise ValueError(msg % (error_template, size, len(data)))
901 else:
902 return data