Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/numpy/lib/format.py: 12%
Shortcuts on this page
r m x toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
Shortcuts on this page
r m x 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 they have been given without much
45 documentation.
47- Allows memory-mapping of the data. See `open_memmap`.
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
160:doc:`"npy-format" NEP <neps:nep-0001-npy-format>`, however details have
161evolved with time and this document is more current.
163"""
164import numpy
165import warnings
166from numpy.lib.utils import safe_eval
167from numpy.compat import (
168 os_fspath, pickle
169 )
170from numpy.compat.py3k import _isfileobj
173__all__ = []
176EXPECTED_KEYS = {'descr', 'fortran_order', 'shape'}
177MAGIC_PREFIX = b'\x93NUMPY'
178MAGIC_LEN = len(MAGIC_PREFIX) + 2
179ARRAY_ALIGN = 64 # plausible values are powers of 2 between 16 and 4096
180BUFFER_SIZE = 2**18 # size of buffer for reading npz files in bytes
181# allow growth within the address space of a 64 bit machine along one axis
182GROWTH_AXIS_MAX_DIGITS = 21 # = len(str(8*2**64-1)) hypothetical int1 dtype
184# difference between version 1.0 and 2.0 is a 4 byte (I) header length
185# instead of 2 bytes (H) allowing storage of large structured arrays
186_header_size_info = {
187 (1, 0): ('<H', 'latin1'),
188 (2, 0): ('<I', 'latin1'),
189 (3, 0): ('<I', 'utf8'),
190}
192# Python's literal_eval is not actually safe for large inputs, since parsing
193# may become slow or even cause interpreter crashes.
194# This is an arbitrary, low limit which should make it safe in practice.
195_MAX_HEADER_SIZE = 10000
197def _check_version(version):
198 if version not in [(1, 0), (2, 0), (3, 0), None]:
199 msg = "we only support format version (1,0), (2,0), and (3,0), not %s"
200 raise ValueError(msg % (version,))
202def magic(major, minor):
203 """ Return the magic string for the given file format version.
205 Parameters
206 ----------
207 major : int in [0, 255]
208 minor : int in [0, 255]
210 Returns
211 -------
212 magic : str
214 Raises
215 ------
216 ValueError if the version cannot be formatted.
217 """
218 if major < 0 or major > 255:
219 raise ValueError("major version must be 0 <= major < 256")
220 if minor < 0 or minor > 255:
221 raise ValueError("minor version must be 0 <= minor < 256")
222 return MAGIC_PREFIX + bytes([major, minor])
224def read_magic(fp):
225 """ Read the magic string to get the version of the file format.
227 Parameters
228 ----------
229 fp : filelike object
231 Returns
232 -------
233 major : int
234 minor : int
235 """
236 magic_str = _read_bytes(fp, MAGIC_LEN, "magic string")
237 if magic_str[:-2] != MAGIC_PREFIX:
238 msg = "the magic string is not correct; expected %r, got %r"
239 raise ValueError(msg % (MAGIC_PREFIX, magic_str[:-2]))
240 major, minor = magic_str[-2:]
241 return major, minor
243def _has_metadata(dt):
244 if dt.metadata is not None:
245 return True
246 elif dt.names is not None:
247 return any(_has_metadata(dt[k]) for k in dt.names)
248 elif dt.subdtype is not None:
249 return _has_metadata(dt.base)
250 else:
251 return False
253def dtype_to_descr(dtype):
254 """
255 Get a serializable descriptor from the dtype.
257 The .descr attribute of a dtype object cannot be round-tripped through
258 the dtype() constructor. Simple types, like dtype('float32'), have
259 a descr which looks like a record array with one field with '' as
260 a name. The dtype() constructor interprets this as a request to give
261 a default name. Instead, we construct descriptor that can be passed to
262 dtype().
264 Parameters
265 ----------
266 dtype : dtype
267 The dtype of the array that will be written to disk.
269 Returns
270 -------
271 descr : object
272 An object that can be passed to `numpy.dtype()` in order to
273 replicate the input dtype.
275 """
276 if _has_metadata(dtype):
277 warnings.warn("metadata on a dtype may be saved or ignored, but will "
278 "raise if saved when read. Use another form of storage.",
279 UserWarning, stacklevel=2)
280 if dtype.names is not None:
281 # This is a record array. The .descr is fine. XXX: parts of the
282 # record array with an empty name, like padding bytes, still get
283 # fiddled with. This needs to be fixed in the C implementation of
284 # dtype().
285 return dtype.descr
286 else:
287 return dtype.str
289def descr_to_dtype(descr):
290 """
291 Returns a dtype based off the given description.
293 This is essentially the reverse of `dtype_to_descr()`. It will remove
294 the valueless padding fields created by, i.e. simple fields like
295 dtype('float32'), and then convert the description to its corresponding
296 dtype.
298 Parameters
299 ----------
300 descr : object
301 The object retrieved by dtype.descr. Can be passed to
302 `numpy.dtype()` in order to replicate the input dtype.
304 Returns
305 -------
306 dtype : dtype
307 The dtype constructed by the description.
309 """
310 if isinstance(descr, str):
311 # No padding removal needed
312 return numpy.dtype(descr)
313 elif isinstance(descr, tuple):
314 # subtype, will always have a shape descr[1]
315 dt = descr_to_dtype(descr[0])
316 return numpy.dtype((dt, descr[1]))
318 titles = []
319 names = []
320 formats = []
321 offsets = []
322 offset = 0
323 for field in descr:
324 if len(field) == 2:
325 name, descr_str = field
326 dt = descr_to_dtype(descr_str)
327 else:
328 name, descr_str, shape = field
329 dt = numpy.dtype((descr_to_dtype(descr_str), shape))
331 # Ignore padding bytes, which will be void bytes with '' as name
332 # Once support for blank names is removed, only "if name == ''" needed)
333 is_pad = (name == '' and dt.type is numpy.void and dt.names is None)
334 if not is_pad:
335 title, name = name if isinstance(name, tuple) else (None, name)
336 titles.append(title)
337 names.append(name)
338 formats.append(dt)
339 offsets.append(offset)
340 offset += dt.itemsize
342 return numpy.dtype({'names': names, 'formats': formats, 'titles': titles,
343 'offsets': offsets, 'itemsize': offset})
345def header_data_from_array_1_0(array):
346 """ Get the dictionary of header metadata from a numpy.ndarray.
348 Parameters
349 ----------
350 array : numpy.ndarray
352 Returns
353 -------
354 d : dict
355 This has the appropriate entries for writing its string representation
356 to the header of the file.
357 """
358 d = {'shape': array.shape}
359 if array.flags.c_contiguous:
360 d['fortran_order'] = False
361 elif array.flags.f_contiguous:
362 d['fortran_order'] = True
363 else:
364 # Totally non-contiguous data. We will have to make it C-contiguous
365 # before writing. Note that we need to test for C_CONTIGUOUS first
366 # because a 1-D array is both C_CONTIGUOUS and F_CONTIGUOUS.
367 d['fortran_order'] = False
369 d['descr'] = dtype_to_descr(array.dtype)
370 return d
373def _wrap_header(header, version):
374 """
375 Takes a stringified header, and attaches the prefix and padding to it
376 """
377 import struct
378 assert version is not None
379 fmt, encoding = _header_size_info[version]
380 header = header.encode(encoding)
381 hlen = len(header) + 1
382 padlen = ARRAY_ALIGN - ((MAGIC_LEN + struct.calcsize(fmt) + hlen) % ARRAY_ALIGN)
383 try:
384 header_prefix = magic(*version) + struct.pack(fmt, hlen + padlen)
385 except struct.error:
386 msg = "Header length {} too big for version={}".format(hlen, version)
387 raise ValueError(msg) from None
389 # Pad the header with spaces and a final newline such that the magic
390 # string, the header-length short and the header are aligned on a
391 # ARRAY_ALIGN byte boundary. This supports memory mapping of dtypes
392 # aligned up to ARRAY_ALIGN on systems like Linux where mmap()
393 # offset must be page-aligned (i.e. the beginning of the file).
394 return header_prefix + header + b' '*padlen + b'\n'
397def _wrap_header_guess_version(header):
398 """
399 Like `_wrap_header`, but chooses an appropriate version given the contents
400 """
401 try:
402 return _wrap_header(header, (1, 0))
403 except ValueError:
404 pass
406 try:
407 ret = _wrap_header(header, (2, 0))
408 except UnicodeEncodeError:
409 pass
410 else:
411 warnings.warn("Stored array in format 2.0. It can only be"
412 "read by NumPy >= 1.9", UserWarning, stacklevel=2)
413 return ret
415 header = _wrap_header(header, (3, 0))
416 warnings.warn("Stored array in format 3.0. It can only be "
417 "read by NumPy >= 1.17", UserWarning, stacklevel=2)
418 return header
421def _write_array_header(fp, d, version=None):
422 """ Write the header for an array and returns the version used
424 Parameters
425 ----------
426 fp : filelike object
427 d : dict
428 This has the appropriate entries for writing its string representation
429 to the header of the file.
430 version : tuple or None
431 None means use oldest that works. Providing an explicit version will
432 raise a ValueError if the format does not allow saving this data.
433 Default: None
434 """
435 header = ["{"]
436 for key, value in sorted(d.items()):
437 # Need to use repr here, since we eval these when reading
438 header.append("'%s': %s, " % (key, repr(value)))
439 header.append("}")
440 header = "".join(header)
442 # Add some spare space so that the array header can be modified in-place
443 # when changing the array size, e.g. when growing it by appending data at
444 # the end.
445 shape = d['shape']
446 header += " " * ((GROWTH_AXIS_MAX_DIGITS - len(repr(
447 shape[-1 if d['fortran_order'] else 0]
448 ))) if len(shape) > 0 else 0)
450 if version is None:
451 header = _wrap_header_guess_version(header)
452 else:
453 header = _wrap_header(header, version)
454 fp.write(header)
456def write_array_header_1_0(fp, d):
457 """ Write the header for an array using the 1.0 format.
459 Parameters
460 ----------
461 fp : filelike object
462 d : dict
463 This has the appropriate entries for writing its string
464 representation to the header of the file.
465 """
466 _write_array_header(fp, d, (1, 0))
469def write_array_header_2_0(fp, d):
470 """ Write the header for an array using the 2.0 format.
471 The 2.0 format allows storing very large structured arrays.
473 .. versionadded:: 1.9.0
475 Parameters
476 ----------
477 fp : filelike object
478 d : dict
479 This has the appropriate entries for writing its string
480 representation to the header of the file.
481 """
482 _write_array_header(fp, d, (2, 0))
484def read_array_header_1_0(fp, max_header_size=_MAX_HEADER_SIZE):
485 """
486 Read an array header from a filelike object using the 1.0 file format
487 version.
489 This will leave the file object located just after the header.
491 Parameters
492 ----------
493 fp : filelike object
494 A file object or something with a `.read()` method like a file.
496 Returns
497 -------
498 shape : tuple of int
499 The shape of the array.
500 fortran_order : bool
501 The array data will be written out directly if it is either
502 C-contiguous or Fortran-contiguous. Otherwise, it will be made
503 contiguous before writing it out.
504 dtype : dtype
505 The dtype of the file's data.
506 max_header_size : int, optional
507 Maximum allowed size of the header. Large headers may not be safe
508 to load securely and thus require explicitly passing a larger value.
509 See :py:meth:`ast.literal_eval()` for details.
511 Raises
512 ------
513 ValueError
514 If the data is invalid.
516 """
517 return _read_array_header(
518 fp, version=(1, 0), max_header_size=max_header_size)
520def read_array_header_2_0(fp, max_header_size=_MAX_HEADER_SIZE):
521 """
522 Read an array header from a filelike object using the 2.0 file format
523 version.
525 This will leave the file object located just after the header.
527 .. versionadded:: 1.9.0
529 Parameters
530 ----------
531 fp : filelike object
532 A file object or something with a `.read()` method like a file.
533 max_header_size : int, optional
534 Maximum allowed size of the header. Large headers may not be safe
535 to load securely and thus require explicitly passing a larger value.
536 See :py:meth:`ast.literal_eval()` for details.
538 Returns
539 -------
540 shape : tuple of int
541 The shape of the array.
542 fortran_order : bool
543 The array data will be written out directly if it is either
544 C-contiguous or Fortran-contiguous. Otherwise, it will be made
545 contiguous before writing it out.
546 dtype : dtype
547 The dtype of the file's data.
549 Raises
550 ------
551 ValueError
552 If the data is invalid.
554 """
555 return _read_array_header(
556 fp, version=(2, 0), max_header_size=max_header_size)
559def _filter_header(s):
560 """Clean up 'L' in npz header ints.
562 Cleans up the 'L' in strings representing integers. Needed to allow npz
563 headers produced in Python2 to be read in Python3.
565 Parameters
566 ----------
567 s : string
568 Npy file header.
570 Returns
571 -------
572 header : str
573 Cleaned up header.
575 """
576 import tokenize
577 from io import StringIO
579 tokens = []
580 last_token_was_number = False
581 for token in tokenize.generate_tokens(StringIO(s).readline):
582 token_type = token[0]
583 token_string = token[1]
584 if (last_token_was_number and
585 token_type == tokenize.NAME and
586 token_string == "L"):
587 continue
588 else:
589 tokens.append(token)
590 last_token_was_number = (token_type == tokenize.NUMBER)
591 return tokenize.untokenize(tokens)
594def _read_array_header(fp, version, max_header_size=_MAX_HEADER_SIZE):
595 """
596 see read_array_header_1_0
597 """
598 # Read an unsigned, little-endian short int which has the length of the
599 # header.
600 import struct
601 hinfo = _header_size_info.get(version)
602 if hinfo is None:
603 raise ValueError("Invalid version {!r}".format(version))
604 hlength_type, encoding = hinfo
606 hlength_str = _read_bytes(fp, struct.calcsize(hlength_type), "array header length")
607 header_length = struct.unpack(hlength_type, hlength_str)[0]
608 header = _read_bytes(fp, header_length, "array header")
609 header = header.decode(encoding)
610 if len(header) > max_header_size:
611 raise ValueError(
612 f"Header info length ({len(header)}) is large and may not be safe "
613 "to load securely.\n"
614 "To allow loading, adjust `max_header_size` or fully trust "
615 "the `.npy` file using `allow_pickle=True`.\n"
616 "For safety against large resource use or crashes, sandboxing "
617 "may be necessary.")
619 # The header is a pretty-printed string representation of a literal
620 # Python dictionary with trailing newlines padded to a ARRAY_ALIGN byte
621 # boundary. The keys are strings.
622 # "shape" : tuple of int
623 # "fortran_order" : bool
624 # "descr" : dtype.descr
625 # Versions (2, 0) and (1, 0) could have been created by a Python 2
626 # implementation before header filtering was implemented.
627 if version <= (2, 0):
628 header = _filter_header(header)
629 try:
630 d = safe_eval(header)
631 except SyntaxError as e:
632 msg = "Cannot parse header: {!r}"
633 raise ValueError(msg.format(header)) from e
634 if not isinstance(d, dict):
635 msg = "Header is not a dictionary: {!r}"
636 raise ValueError(msg.format(d))
638 if EXPECTED_KEYS != d.keys():
639 keys = sorted(d.keys())
640 msg = "Header does not contain the correct keys: {!r}"
641 raise ValueError(msg.format(keys))
643 # Sanity-check the values.
644 if (not isinstance(d['shape'], tuple) or
645 not all(isinstance(x, int) for x in d['shape'])):
646 msg = "shape is not valid: {!r}"
647 raise ValueError(msg.format(d['shape']))
648 if not isinstance(d['fortran_order'], bool):
649 msg = "fortran_order is not a valid bool: {!r}"
650 raise ValueError(msg.format(d['fortran_order']))
651 try:
652 dtype = descr_to_dtype(d['descr'])
653 except TypeError as e:
654 msg = "descr is not a valid dtype descriptor: {!r}"
655 raise ValueError(msg.format(d['descr'])) from e
657 return d['shape'], d['fortran_order'], dtype
659def write_array(fp, array, version=None, allow_pickle=True, pickle_kwargs=None):
660 """
661 Write an array to an NPY file, including a header.
663 If the array is neither C-contiguous nor Fortran-contiguous AND the
664 file_like object is not a real file object, this function will have to
665 copy data in memory.
667 Parameters
668 ----------
669 fp : file_like object
670 An open, writable file object, or similar object with a
671 ``.write()`` method.
672 array : ndarray
673 The array to write to disk.
674 version : (int, int) or None, optional
675 The version number of the format. None means use the oldest
676 supported version that is able to store the data. Default: None
677 allow_pickle : bool, optional
678 Whether to allow writing pickled data. Default: True
679 pickle_kwargs : dict, optional
680 Additional keyword arguments to pass to pickle.dump, excluding
681 'protocol'. These are only useful when pickling objects in object
682 arrays on Python 3 to Python 2 compatible format.
684 Raises
685 ------
686 ValueError
687 If the array cannot be persisted. This includes the case of
688 allow_pickle=False and array being an object array.
689 Various other errors
690 If the array contains Python objects as part of its dtype, the
691 process of pickling them may raise various errors if the objects
692 are not picklable.
694 """
695 _check_version(version)
696 _write_array_header(fp, header_data_from_array_1_0(array), version)
698 if array.itemsize == 0:
699 buffersize = 0
700 else:
701 # Set buffer size to 16 MiB to hide the Python loop overhead.
702 buffersize = max(16 * 1024 ** 2 // array.itemsize, 1)
704 if array.dtype.hasobject:
705 # We contain Python objects so we cannot write out the data
706 # directly. Instead, we will pickle it out
707 if not allow_pickle:
708 raise ValueError("Object arrays cannot be saved when "
709 "allow_pickle=False")
710 if pickle_kwargs is None:
711 pickle_kwargs = {}
712 pickle.dump(array, fp, protocol=3, **pickle_kwargs)
713 elif array.flags.f_contiguous and not array.flags.c_contiguous:
714 if _isfileobj(fp):
715 array.T.tofile(fp)
716 else:
717 for chunk in numpy.nditer(
718 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
719 buffersize=buffersize, order='F'):
720 fp.write(chunk.tobytes('C'))
721 else:
722 if _isfileobj(fp):
723 array.tofile(fp)
724 else:
725 for chunk in numpy.nditer(
726 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
727 buffersize=buffersize, order='C'):
728 fp.write(chunk.tobytes('C'))
731def read_array(fp, allow_pickle=False, pickle_kwargs=None, *,
732 max_header_size=_MAX_HEADER_SIZE):
733 """
734 Read an array from an NPY file.
736 Parameters
737 ----------
738 fp : file_like object
739 If this is not a real file object, then this may take extra memory
740 and time.
741 allow_pickle : bool, optional
742 Whether to allow writing pickled data. Default: False
744 .. versionchanged:: 1.16.3
745 Made default False in response to CVE-2019-6446.
747 pickle_kwargs : dict
748 Additional keyword arguments to pass to pickle.load. These are only
749 useful when loading object arrays saved on Python 2 when using
750 Python 3.
751 max_header_size : int, optional
752 Maximum allowed size of the header. Large headers may not be safe
753 to load securely and thus require explicitly passing a larger value.
754 See :py:meth:`ast.literal_eval()` for details.
755 This option is ignored when `allow_pickle` is passed. In that case
756 the file is by definition trusted and the limit is unnecessary.
758 Returns
759 -------
760 array : ndarray
761 The array from the data on disk.
763 Raises
764 ------
765 ValueError
766 If the data is invalid, or allow_pickle=False and the file contains
767 an object array.
769 """
770 if allow_pickle:
771 # Effectively ignore max_header_size, since `allow_pickle` indicates
772 # that the input is fully trusted.
773 max_header_size = 2**64
775 version = read_magic(fp)
776 _check_version(version)
777 shape, fortran_order, dtype = _read_array_header(
778 fp, version, max_header_size=max_header_size)
779 if len(shape) == 0:
780 count = 1
781 else:
782 count = numpy.multiply.reduce(shape, dtype=numpy.int64)
784 # Now read the actual data.
785 if dtype.hasobject:
786 # The array contained Python objects. We need to unpickle the data.
787 if not allow_pickle:
788 raise ValueError("Object arrays cannot be loaded when "
789 "allow_pickle=False")
790 if pickle_kwargs is None:
791 pickle_kwargs = {}
792 try:
793 array = pickle.load(fp, **pickle_kwargs)
794 except UnicodeError as err:
795 # Friendlier error message
796 raise UnicodeError("Unpickling a python object failed: %r\n"
797 "You may need to pass the encoding= option "
798 "to numpy.load" % (err,)) from err
799 else:
800 if _isfileobj(fp):
801 # We can use the fast fromfile() function.
802 array = numpy.fromfile(fp, dtype=dtype, count=count)
803 else:
804 # This is not a real file. We have to read it the
805 # memory-intensive way.
806 # crc32 module fails on reads greater than 2 ** 32 bytes,
807 # breaking large reads from gzip streams. Chunk reads to
808 # BUFFER_SIZE bytes to avoid issue and reduce memory overhead
809 # of the read. In non-chunked case count < max_read_count, so
810 # only one read is performed.
812 # Use np.ndarray instead of np.empty since the latter does
813 # not correctly instantiate zero-width string dtypes; see
814 # https://github.com/numpy/numpy/pull/6430
815 array = numpy.ndarray(count, dtype=dtype)
817 if dtype.itemsize > 0:
818 # If dtype.itemsize == 0 then there's nothing more to read
819 max_read_count = BUFFER_SIZE // min(BUFFER_SIZE, dtype.itemsize)
821 for i in range(0, count, max_read_count):
822 read_count = min(max_read_count, count - i)
823 read_size = int(read_count * dtype.itemsize)
824 data = _read_bytes(fp, read_size, "array data")
825 array[i:i+read_count] = numpy.frombuffer(data, dtype=dtype,
826 count=read_count)
828 if fortran_order:
829 array.shape = shape[::-1]
830 array = array.transpose()
831 else:
832 array.shape = shape
834 return array
837def open_memmap(filename, mode='r+', dtype=None, shape=None,
838 fortran_order=False, version=None, *,
839 max_header_size=_MAX_HEADER_SIZE):
840 """
841 Open a .npy file as a memory-mapped array.
843 This may be used to read an existing file or create a new one.
845 Parameters
846 ----------
847 filename : str or path-like
848 The name of the file on disk. This may *not* be a file-like
849 object.
850 mode : str, optional
851 The mode in which to open the file; the default is 'r+'. In
852 addition to the standard file modes, 'c' is also accepted to mean
853 "copy on write." See `memmap` for the available mode strings.
854 dtype : data-type, optional
855 The data type of the array if we are creating a new file in "write"
856 mode, if not, `dtype` is ignored. The default value is None, which
857 results in a data-type of `float64`.
858 shape : tuple of int
859 The shape of the array if we are creating a new file in "write"
860 mode, in which case this parameter is required. Otherwise, this
861 parameter is ignored and is thus optional.
862 fortran_order : bool, optional
863 Whether the array should be Fortran-contiguous (True) or
864 C-contiguous (False, the default) if we are creating a new file in
865 "write" mode.
866 version : tuple of int (major, minor) or None
867 If the mode is a "write" mode, then this is the version of the file
868 format used to create the file. None means use the oldest
869 supported version that is able to store the data. Default: None
870 max_header_size : int, optional
871 Maximum allowed size of the header. Large headers may not be safe
872 to load securely and thus require explicitly passing a larger value.
873 See :py:meth:`ast.literal_eval()` for details.
875 Returns
876 -------
877 marray : memmap
878 The memory-mapped array.
880 Raises
881 ------
882 ValueError
883 If the data or the mode is invalid.
884 OSError
885 If the file is not found or cannot be opened correctly.
887 See Also
888 --------
889 numpy.memmap
891 """
892 if _isfileobj(filename):
893 raise ValueError("Filename must be a string or a path-like object."
894 " Memmap cannot use existing file handles.")
896 if 'w' in mode:
897 # We are creating the file, not reading it.
898 # Check if we ought to create the file.
899 _check_version(version)
900 # Ensure that the given dtype is an authentic dtype object rather
901 # than just something that can be interpreted as a dtype object.
902 dtype = numpy.dtype(dtype)
903 if dtype.hasobject:
904 msg = "Array can't be memory-mapped: Python objects in dtype."
905 raise ValueError(msg)
906 d = dict(
907 descr=dtype_to_descr(dtype),
908 fortran_order=fortran_order,
909 shape=shape,
910 )
911 # If we got here, then it should be safe to create the file.
912 with open(os_fspath(filename), mode+'b') as fp:
913 _write_array_header(fp, d, version)
914 offset = fp.tell()
915 else:
916 # Read the header of the file first.
917 with open(os_fspath(filename), 'rb') as fp:
918 version = read_magic(fp)
919 _check_version(version)
921 shape, fortran_order, dtype = _read_array_header(
922 fp, version, max_header_size=max_header_size)
923 if dtype.hasobject:
924 msg = "Array can't be memory-mapped: Python objects in dtype."
925 raise ValueError(msg)
926 offset = fp.tell()
928 if fortran_order:
929 order = 'F'
930 else:
931 order = 'C'
933 # We need to change a write-only mode to a read-write mode since we've
934 # already written data to the file.
935 if mode == 'w+':
936 mode = 'r+'
938 marray = numpy.memmap(filename, dtype=dtype, shape=shape, order=order,
939 mode=mode, offset=offset)
941 return marray
944def _read_bytes(fp, size, error_template="ran out of data"):
945 """
946 Read from file-like object until size bytes are read.
947 Raises ValueError if not EOF is encountered before size bytes are read.
948 Non-blocking objects only supported if they derive from io objects.
950 Required as e.g. ZipExtFile in python 2.6 can return less data than
951 requested.
952 """
953 data = bytes()
954 while True:
955 # io files (default in python3) return None or raise on
956 # would-block, python2 file will truncate, probably nothing can be
957 # done about that. note that regular files can't be non-blocking
958 try:
959 r = fp.read(size - len(data))
960 data += r
961 if len(r) == 0 or len(data) == size:
962 break
963 except BlockingIOError:
964 pass
965 if len(data) != size:
966 msg = "EOF: reading %s, expected %d bytes got %d"
967 raise ValueError(msg % (error_template, size, len(data)))
968 else:
969 return data