Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.10/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
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 io
165import os
166import pickle
167import warnings
169import numpy
170from numpy.lib._utils_impl import drop_metadata
173__all__ = []
175drop_metadata.__module__ = "numpy.lib.format"
177EXPECTED_KEYS = {'descr', 'fortran_order', 'shape'}
178MAGIC_PREFIX = b'\x93NUMPY'
179MAGIC_LEN = len(MAGIC_PREFIX) + 2
180ARRAY_ALIGN = 64 # plausible values are powers of 2 between 16 and 4096
181BUFFER_SIZE = 2**18 # size of buffer for reading npz files in bytes
182# allow growth within the address space of a 64 bit machine along one axis
183GROWTH_AXIS_MAX_DIGITS = 21 # = len(str(8*2**64-1)) hypothetical int1 dtype
185# difference between version 1.0 and 2.0 is a 4 byte (I) header length
186# instead of 2 bytes (H) allowing storage of large structured arrays
187_header_size_info = {
188 (1, 0): ('<H', 'latin1'),
189 (2, 0): ('<I', 'latin1'),
190 (3, 0): ('<I', 'utf8'),
191}
193# Python's literal_eval is not actually safe for large inputs, since parsing
194# may become slow or even cause interpreter crashes.
195# This is an arbitrary, low limit which should make it safe in practice.
196_MAX_HEADER_SIZE = 10000
198def _check_version(version):
199 if version not in [(1, 0), (2, 0), (3, 0), None]:
200 msg = "we only support format version (1,0), (2,0), and (3,0), not %s"
201 raise ValueError(msg % (version,))
203def magic(major, minor):
204 """ Return the magic string for the given file format version.
206 Parameters
207 ----------
208 major : int in [0, 255]
209 minor : int in [0, 255]
211 Returns
212 -------
213 magic : str
215 Raises
216 ------
217 ValueError if the version cannot be formatted.
218 """
219 if major < 0 or major > 255:
220 raise ValueError("major version must be 0 <= major < 256")
221 if minor < 0 or minor > 255:
222 raise ValueError("minor version must be 0 <= minor < 256")
223 return MAGIC_PREFIX + bytes([major, minor])
225def read_magic(fp):
226 """ Read the magic string to get the version of the file format.
228 Parameters
229 ----------
230 fp : filelike object
232 Returns
233 -------
234 major : int
235 minor : int
236 """
237 magic_str = _read_bytes(fp, MAGIC_LEN, "magic string")
238 if magic_str[:-2] != MAGIC_PREFIX:
239 msg = "the magic string is not correct; expected %r, got %r"
240 raise ValueError(msg % (MAGIC_PREFIX, magic_str[:-2]))
241 major, minor = magic_str[-2:]
242 return major, minor
245def dtype_to_descr(dtype):
246 """
247 Get a serializable descriptor from the dtype.
249 The .descr attribute of a dtype object cannot be round-tripped through
250 the dtype() constructor. Simple types, like dtype('float32'), have
251 a descr which looks like a record array with one field with '' as
252 a name. The dtype() constructor interprets this as a request to give
253 a default name. Instead, we construct descriptor that can be passed to
254 dtype().
256 Parameters
257 ----------
258 dtype : dtype
259 The dtype of the array that will be written to disk.
261 Returns
262 -------
263 descr : object
264 An object that can be passed to `numpy.dtype()` in order to
265 replicate the input dtype.
267 """
268 # NOTE: that drop_metadata may not return the right dtype e.g. for user
269 # dtypes. In that case our code below would fail the same, though.
270 new_dtype = drop_metadata(dtype)
271 if new_dtype is not dtype:
272 warnings.warn("metadata on a dtype is not saved to an npy/npz. "
273 "Use another format (such as pickle) to store it.",
274 UserWarning, stacklevel=2)
275 dtype = new_dtype
277 if dtype.names is not None:
278 # This is a record array. The .descr is fine. XXX: parts of the
279 # record array with an empty name, like padding bytes, still get
280 # fiddled with. This needs to be fixed in the C implementation of
281 # dtype().
282 return dtype.descr
283 elif not type(dtype)._legacy:
284 # this must be a user-defined dtype since numpy does not yet expose any
285 # non-legacy dtypes in the public API
286 #
287 # non-legacy dtypes don't yet have __array_interface__
288 # support. Instead, as a hack, we use pickle to save the array, and lie
289 # that the dtype is object. When the array is loaded, the descriptor is
290 # unpickled with the array and the object dtype in the header is
291 # discarded.
292 #
293 # a future NEP should define a way to serialize user-defined
294 # descriptors and ideally work out the possible security implications
295 warnings.warn("Custom dtypes are saved as python objects using the "
296 "pickle protocol. Loading this file requires "
297 "allow_pickle=True to be set.",
298 UserWarning, stacklevel=2)
299 return "|O"
300 else:
301 return dtype.str
303def descr_to_dtype(descr):
304 """
305 Returns a dtype based off the given description.
307 This is essentially the reverse of `~lib.format.dtype_to_descr`. It will
308 remove the valueless padding fields created by, i.e. simple fields like
309 dtype('float32'), and then convert the description to its corresponding
310 dtype.
312 Parameters
313 ----------
314 descr : object
315 The object retrieved by dtype.descr. Can be passed to
316 `numpy.dtype` in order to replicate the input dtype.
318 Returns
319 -------
320 dtype : dtype
321 The dtype constructed by the description.
323 """
324 if isinstance(descr, str):
325 # No padding removal needed
326 return numpy.dtype(descr)
327 elif isinstance(descr, tuple):
328 # subtype, will always have a shape descr[1]
329 dt = descr_to_dtype(descr[0])
330 return numpy.dtype((dt, descr[1]))
332 titles = []
333 names = []
334 formats = []
335 offsets = []
336 offset = 0
337 for field in descr:
338 if len(field) == 2:
339 name, descr_str = field
340 dt = descr_to_dtype(descr_str)
341 else:
342 name, descr_str, shape = field
343 dt = numpy.dtype((descr_to_dtype(descr_str), shape))
345 # Ignore padding bytes, which will be void bytes with '' as name
346 # Once support for blank names is removed, only "if name == ''" needed)
347 is_pad = (name == '' and dt.type is numpy.void and dt.names is None)
348 if not is_pad:
349 title, name = name if isinstance(name, tuple) else (None, name)
350 titles.append(title)
351 names.append(name)
352 formats.append(dt)
353 offsets.append(offset)
354 offset += dt.itemsize
356 return numpy.dtype({'names': names, 'formats': formats, 'titles': titles,
357 'offsets': offsets, 'itemsize': offset})
359def header_data_from_array_1_0(array):
360 """ Get the dictionary of header metadata from a numpy.ndarray.
362 Parameters
363 ----------
364 array : numpy.ndarray
366 Returns
367 -------
368 d : dict
369 This has the appropriate entries for writing its string representation
370 to the header of the file.
371 """
372 d = {'shape': array.shape}
373 if array.flags.c_contiguous:
374 d['fortran_order'] = False
375 elif array.flags.f_contiguous:
376 d['fortran_order'] = True
377 else:
378 # Totally non-contiguous data. We will have to make it C-contiguous
379 # before writing. Note that we need to test for C_CONTIGUOUS first
380 # because a 1-D array is both C_CONTIGUOUS and F_CONTIGUOUS.
381 d['fortran_order'] = False
383 d['descr'] = dtype_to_descr(array.dtype)
384 return d
387def _wrap_header(header, version):
388 """
389 Takes a stringified header, and attaches the prefix and padding to it
390 """
391 import struct
392 assert version is not None
393 fmt, encoding = _header_size_info[version]
394 header = header.encode(encoding)
395 hlen = len(header) + 1
396 padlen = ARRAY_ALIGN - ((MAGIC_LEN + struct.calcsize(fmt) + hlen) % ARRAY_ALIGN)
397 try:
398 header_prefix = magic(*version) + struct.pack(fmt, hlen + padlen)
399 except struct.error:
400 msg = "Header length {} too big for version={}".format(hlen, version)
401 raise ValueError(msg) from None
403 # Pad the header with spaces and a final newline such that the magic
404 # string, the header-length short and the header are aligned on a
405 # ARRAY_ALIGN byte boundary. This supports memory mapping of dtypes
406 # aligned up to ARRAY_ALIGN on systems like Linux where mmap()
407 # offset must be page-aligned (i.e. the beginning of the file).
408 return header_prefix + header + b' '*padlen + b'\n'
411def _wrap_header_guess_version(header):
412 """
413 Like `_wrap_header`, but chooses an appropriate version given the contents
414 """
415 try:
416 return _wrap_header(header, (1, 0))
417 except ValueError:
418 pass
420 try:
421 ret = _wrap_header(header, (2, 0))
422 except UnicodeEncodeError:
423 pass
424 else:
425 warnings.warn("Stored array in format 2.0. It can only be"
426 "read by NumPy >= 1.9", UserWarning, stacklevel=2)
427 return ret
429 header = _wrap_header(header, (3, 0))
430 warnings.warn("Stored array in format 3.0. It can only be "
431 "read by NumPy >= 1.17", UserWarning, stacklevel=2)
432 return header
435def _write_array_header(fp, d, version=None):
436 """ Write the header for an array and returns the version used
438 Parameters
439 ----------
440 fp : filelike object
441 d : dict
442 This has the appropriate entries for writing its string representation
443 to the header of the file.
444 version : tuple or None
445 None means use oldest that works. Providing an explicit version will
446 raise a ValueError if the format does not allow saving this data.
447 Default: None
448 """
449 header = ["{"]
450 for key, value in sorted(d.items()):
451 # Need to use repr here, since we eval these when reading
452 header.append("'%s': %s, " % (key, repr(value)))
453 header.append("}")
454 header = "".join(header)
456 # Add some spare space so that the array header can be modified in-place
457 # when changing the array size, e.g. when growing it by appending data at
458 # the end.
459 shape = d['shape']
460 header += " " * ((GROWTH_AXIS_MAX_DIGITS - len(repr(
461 shape[-1 if d['fortran_order'] else 0]
462 ))) if len(shape) > 0 else 0)
464 if version is None:
465 header = _wrap_header_guess_version(header)
466 else:
467 header = _wrap_header(header, version)
468 fp.write(header)
470def write_array_header_1_0(fp, d):
471 """ Write the header for an array using the 1.0 format.
473 Parameters
474 ----------
475 fp : filelike object
476 d : dict
477 This has the appropriate entries for writing its string
478 representation to the header of the file.
479 """
480 _write_array_header(fp, d, (1, 0))
483def write_array_header_2_0(fp, d):
484 """ Write the header for an array using the 2.0 format.
485 The 2.0 format allows storing very large structured arrays.
487 Parameters
488 ----------
489 fp : filelike object
490 d : dict
491 This has the appropriate entries for writing its string
492 representation to the header of the file.
493 """
494 _write_array_header(fp, d, (2, 0))
496def read_array_header_1_0(fp, max_header_size=_MAX_HEADER_SIZE):
497 """
498 Read an array header from a filelike object using the 1.0 file format
499 version.
501 This will leave the file object located just after the header.
503 Parameters
504 ----------
505 fp : filelike object
506 A file object or something with a `.read()` method like a file.
508 Returns
509 -------
510 shape : tuple of int
511 The shape of the array.
512 fortran_order : bool
513 The array data will be written out directly if it is either
514 C-contiguous or Fortran-contiguous. Otherwise, it will be made
515 contiguous before writing it out.
516 dtype : dtype
517 The dtype of the file's data.
518 max_header_size : int, optional
519 Maximum allowed size of the header. Large headers may not be safe
520 to load securely and thus require explicitly passing a larger value.
521 See :py:func:`ast.literal_eval()` for details.
523 Raises
524 ------
525 ValueError
526 If the data is invalid.
528 """
529 return _read_array_header(
530 fp, version=(1, 0), max_header_size=max_header_size)
532def read_array_header_2_0(fp, max_header_size=_MAX_HEADER_SIZE):
533 """
534 Read an array header from a filelike object using the 2.0 file format
535 version.
537 This will leave the file object located just after the header.
539 Parameters
540 ----------
541 fp : filelike object
542 A file object or something with a `.read()` method like a file.
543 max_header_size : int, optional
544 Maximum allowed size of the header. Large headers may not be safe
545 to load securely and thus require explicitly passing a larger value.
546 See :py:func:`ast.literal_eval()` for details.
548 Returns
549 -------
550 shape : tuple of int
551 The shape of the array.
552 fortran_order : bool
553 The array data will be written out directly if it is either
554 C-contiguous or Fortran-contiguous. Otherwise, it will be made
555 contiguous before writing it out.
556 dtype : dtype
557 The dtype of the file's data.
559 Raises
560 ------
561 ValueError
562 If the data is invalid.
564 """
565 return _read_array_header(
566 fp, version=(2, 0), max_header_size=max_header_size)
569def _filter_header(s):
570 """Clean up 'L' in npz header ints.
572 Cleans up the 'L' in strings representing integers. Needed to allow npz
573 headers produced in Python2 to be read in Python3.
575 Parameters
576 ----------
577 s : string
578 Npy file header.
580 Returns
581 -------
582 header : str
583 Cleaned up header.
585 """
586 import tokenize
587 from io import StringIO
589 tokens = []
590 last_token_was_number = False
591 for token in tokenize.generate_tokens(StringIO(s).readline):
592 token_type = token[0]
593 token_string = token[1]
594 if (last_token_was_number and
595 token_type == tokenize.NAME and
596 token_string == "L"):
597 continue
598 else:
599 tokens.append(token)
600 last_token_was_number = (token_type == tokenize.NUMBER)
601 return tokenize.untokenize(tokens)
604def _read_array_header(fp, version, max_header_size=_MAX_HEADER_SIZE):
605 """
606 see read_array_header_1_0
607 """
608 # Read an unsigned, little-endian short int which has the length of the
609 # header.
610 import ast
611 import struct
612 hinfo = _header_size_info.get(version)
613 if hinfo is None:
614 raise ValueError("Invalid version {!r}".format(version))
615 hlength_type, encoding = hinfo
617 hlength_str = _read_bytes(fp, struct.calcsize(hlength_type), "array header length")
618 header_length = struct.unpack(hlength_type, hlength_str)[0]
619 header = _read_bytes(fp, header_length, "array header")
620 header = header.decode(encoding)
621 if len(header) > max_header_size:
622 raise ValueError(
623 f"Header info length ({len(header)}) is large and may not be safe "
624 "to load securely.\n"
625 "To allow loading, adjust `max_header_size` or fully trust "
626 "the `.npy` file using `allow_pickle=True`.\n"
627 "For safety against large resource use or crashes, sandboxing "
628 "may be necessary.")
630 # The header is a pretty-printed string representation of a literal
631 # Python dictionary with trailing newlines padded to a ARRAY_ALIGN byte
632 # boundary. The keys are strings.
633 # "shape" : tuple of int
634 # "fortran_order" : bool
635 # "descr" : dtype.descr
636 # Versions (2, 0) and (1, 0) could have been created by a Python 2
637 # implementation before header filtering was implemented.
638 #
639 # For performance reasons, we try without _filter_header first though
640 try:
641 d = ast.literal_eval(header)
642 except SyntaxError as e:
643 if version <= (2, 0):
644 header = _filter_header(header)
645 try:
646 d = ast.literal_eval(header)
647 except SyntaxError as e2:
648 msg = "Cannot parse header: {!r}"
649 raise ValueError(msg.format(header)) from e2
650 else:
651 warnings.warn(
652 "Reading `.npy` or `.npz` file required additional "
653 "header parsing as it was created on Python 2. Save the "
654 "file again to speed up loading and avoid this warning.",
655 UserWarning, stacklevel=4)
656 else:
657 msg = "Cannot parse header: {!r}"
658 raise ValueError(msg.format(header)) from e
659 if not isinstance(d, dict):
660 msg = "Header is not a dictionary: {!r}"
661 raise ValueError(msg.format(d))
663 if EXPECTED_KEYS != d.keys():
664 keys = sorted(d.keys())
665 msg = "Header does not contain the correct keys: {!r}"
666 raise ValueError(msg.format(keys))
668 # Sanity-check the values.
669 if (not isinstance(d['shape'], tuple) or
670 not all(isinstance(x, int) for x in d['shape'])):
671 msg = "shape is not valid: {!r}"
672 raise ValueError(msg.format(d['shape']))
673 if not isinstance(d['fortran_order'], bool):
674 msg = "fortran_order is not a valid bool: {!r}"
675 raise ValueError(msg.format(d['fortran_order']))
676 try:
677 dtype = descr_to_dtype(d['descr'])
678 except TypeError as e:
679 msg = "descr is not a valid dtype descriptor: {!r}"
680 raise ValueError(msg.format(d['descr'])) from e
682 return d['shape'], d['fortran_order'], dtype
684def write_array(fp, array, version=None, allow_pickle=True, pickle_kwargs=None):
685 """
686 Write an array to an NPY file, including a header.
688 If the array is neither C-contiguous nor Fortran-contiguous AND the
689 file_like object is not a real file object, this function will have to
690 copy data in memory.
692 Parameters
693 ----------
694 fp : file_like object
695 An open, writable file object, or similar object with a
696 ``.write()`` method.
697 array : ndarray
698 The array to write to disk.
699 version : (int, int) or None, optional
700 The version number of the format. None means use the oldest
701 supported version that is able to store the data. Default: None
702 allow_pickle : bool, optional
703 Whether to allow writing pickled data. Default: True
704 pickle_kwargs : dict, optional
705 Additional keyword arguments to pass to pickle.dump, excluding
706 'protocol'. These are only useful when pickling objects in object
707 arrays on Python 3 to Python 2 compatible format.
709 Raises
710 ------
711 ValueError
712 If the array cannot be persisted. This includes the case of
713 allow_pickle=False and array being an object array.
714 Various other errors
715 If the array contains Python objects as part of its dtype, the
716 process of pickling them may raise various errors if the objects
717 are not picklable.
719 """
720 _check_version(version)
721 _write_array_header(fp, header_data_from_array_1_0(array), version)
723 if array.itemsize == 0:
724 buffersize = 0
725 else:
726 # Set buffer size to 16 MiB to hide the Python loop overhead.
727 buffersize = max(16 * 1024 ** 2 // array.itemsize, 1)
729 dtype_class = type(array.dtype)
731 if array.dtype.hasobject or not dtype_class._legacy:
732 # We contain Python objects so we cannot write out the data
733 # directly. Instead, we will pickle it out
734 if not allow_pickle:
735 if array.dtype.hasobject:
736 raise ValueError("Object arrays cannot be saved when "
737 "allow_pickle=False")
738 if not dtype_class._legacy:
739 raise ValueError("User-defined dtypes cannot be saved "
740 "when allow_pickle=False")
741 if pickle_kwargs is None:
742 pickle_kwargs = {}
743 pickle.dump(array, fp, protocol=4, **pickle_kwargs)
744 elif array.flags.f_contiguous and not array.flags.c_contiguous:
745 if isfileobj(fp):
746 array.T.tofile(fp)
747 else:
748 for chunk in numpy.nditer(
749 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
750 buffersize=buffersize, order='F'):
751 fp.write(chunk.tobytes('C'))
752 else:
753 if isfileobj(fp):
754 array.tofile(fp)
755 else:
756 for chunk in numpy.nditer(
757 array, flags=['external_loop', 'buffered', 'zerosize_ok'],
758 buffersize=buffersize, order='C'):
759 fp.write(chunk.tobytes('C'))
762def read_array(fp, allow_pickle=False, pickle_kwargs=None, *,
763 max_header_size=_MAX_HEADER_SIZE):
764 """
765 Read an array from an NPY file.
767 Parameters
768 ----------
769 fp : file_like object
770 If this is not a real file object, then this may take extra memory
771 and time.
772 allow_pickle : bool, optional
773 Whether to allow writing pickled data. Default: False
774 pickle_kwargs : dict
775 Additional keyword arguments to pass to pickle.load. These are only
776 useful when loading object arrays saved on Python 2 when using
777 Python 3.
778 max_header_size : int, optional
779 Maximum allowed size of the header. Large headers may not be safe
780 to load securely and thus require explicitly passing a larger value.
781 See :py:func:`ast.literal_eval()` for details.
782 This option is ignored when `allow_pickle` is passed. In that case
783 the file is by definition trusted and the limit is unnecessary.
785 Returns
786 -------
787 array : ndarray
788 The array from the data on disk.
790 Raises
791 ------
792 ValueError
793 If the data is invalid, or allow_pickle=False and the file contains
794 an object array.
796 """
797 if allow_pickle:
798 # Effectively ignore max_header_size, since `allow_pickle` indicates
799 # that the input is fully trusted.
800 max_header_size = 2**64
802 version = read_magic(fp)
803 _check_version(version)
804 shape, fortran_order, dtype = _read_array_header(
805 fp, version, max_header_size=max_header_size)
806 if len(shape) == 0:
807 count = 1
808 else:
809 count = numpy.multiply.reduce(shape, dtype=numpy.int64)
811 # Now read the actual data.
812 if dtype.hasobject:
813 # The array contained Python objects. We need to unpickle the data.
814 if not allow_pickle:
815 raise ValueError("Object arrays cannot be loaded when "
816 "allow_pickle=False")
817 if pickle_kwargs is None:
818 pickle_kwargs = {}
819 try:
820 array = pickle.load(fp, **pickle_kwargs)
821 except UnicodeError as err:
822 # Friendlier error message
823 raise UnicodeError("Unpickling a python object failed: %r\n"
824 "You may need to pass the encoding= option "
825 "to numpy.load" % (err,)) from err
826 else:
827 if isfileobj(fp):
828 # We can use the fast fromfile() function.
829 array = numpy.fromfile(fp, dtype=dtype, count=count)
830 else:
831 # This is not a real file. We have to read it the
832 # memory-intensive way.
833 # crc32 module fails on reads greater than 2 ** 32 bytes,
834 # breaking large reads from gzip streams. Chunk reads to
835 # BUFFER_SIZE bytes to avoid issue and reduce memory overhead
836 # of the read. In non-chunked case count < max_read_count, so
837 # only one read is performed.
839 # Use np.ndarray instead of np.empty since the latter does
840 # not correctly instantiate zero-width string dtypes; see
841 # https://github.com/numpy/numpy/pull/6430
842 array = numpy.ndarray(count, dtype=dtype)
844 if dtype.itemsize > 0:
845 # If dtype.itemsize == 0 then there's nothing more to read
846 max_read_count = BUFFER_SIZE // min(BUFFER_SIZE, dtype.itemsize)
848 for i in range(0, count, max_read_count):
849 read_count = min(max_read_count, count - i)
850 read_size = int(read_count * dtype.itemsize)
851 data = _read_bytes(fp, read_size, "array data")
852 array[i:i+read_count] = numpy.frombuffer(data, dtype=dtype,
853 count=read_count)
855 if fortran_order:
856 array.shape = shape[::-1]
857 array = array.transpose()
858 else:
859 array.shape = shape
861 return array
864def open_memmap(filename, mode='r+', dtype=None, shape=None,
865 fortran_order=False, version=None, *,
866 max_header_size=_MAX_HEADER_SIZE):
867 """
868 Open a .npy file as a memory-mapped array.
870 This may be used to read an existing file or create a new one.
872 Parameters
873 ----------
874 filename : str or path-like
875 The name of the file on disk. This may *not* be a file-like
876 object.
877 mode : str, optional
878 The mode in which to open the file; the default is 'r+'. In
879 addition to the standard file modes, 'c' is also accepted to mean
880 "copy on write." See `memmap` for the available mode strings.
881 dtype : data-type, optional
882 The data type of the array if we are creating a new file in "write"
883 mode, if not, `dtype` is ignored. The default value is None, which
884 results in a data-type of `float64`.
885 shape : tuple of int
886 The shape of the array if we are creating a new file in "write"
887 mode, in which case this parameter is required. Otherwise, this
888 parameter is ignored and is thus optional.
889 fortran_order : bool, optional
890 Whether the array should be Fortran-contiguous (True) or
891 C-contiguous (False, the default) if we are creating a new file in
892 "write" mode.
893 version : tuple of int (major, minor) or None
894 If the mode is a "write" mode, then this is the version of the file
895 format used to create the file. None means use the oldest
896 supported version that is able to store the data. Default: None
897 max_header_size : int, optional
898 Maximum allowed size of the header. Large headers may not be safe
899 to load securely and thus require explicitly passing a larger value.
900 See :py:func:`ast.literal_eval()` for details.
902 Returns
903 -------
904 marray : memmap
905 The memory-mapped array.
907 Raises
908 ------
909 ValueError
910 If the data or the mode is invalid.
911 OSError
912 If the file is not found or cannot be opened correctly.
914 See Also
915 --------
916 numpy.memmap
918 """
919 if isfileobj(filename):
920 raise ValueError("Filename must be a string or a path-like object."
921 " Memmap cannot use existing file handles.")
923 if 'w' in mode:
924 # We are creating the file, not reading it.
925 # Check if we ought to create the file.
926 _check_version(version)
927 # Ensure that the given dtype is an authentic dtype object rather
928 # than just something that can be interpreted as a dtype object.
929 dtype = numpy.dtype(dtype)
930 if dtype.hasobject:
931 msg = "Array can't be memory-mapped: Python objects in dtype."
932 raise ValueError(msg)
933 d = dict(
934 descr=dtype_to_descr(dtype),
935 fortran_order=fortran_order,
936 shape=shape,
937 )
938 # If we got here, then it should be safe to create the file.
939 with open(os.fspath(filename), mode+'b') as fp:
940 _write_array_header(fp, d, version)
941 offset = fp.tell()
942 else:
943 # Read the header of the file first.
944 with open(os.fspath(filename), 'rb') as fp:
945 version = read_magic(fp)
946 _check_version(version)
948 shape, fortran_order, dtype = _read_array_header(
949 fp, version, max_header_size=max_header_size)
950 if dtype.hasobject:
951 msg = "Array can't be memory-mapped: Python objects in dtype."
952 raise ValueError(msg)
953 offset = fp.tell()
955 if fortran_order:
956 order = 'F'
957 else:
958 order = 'C'
960 # We need to change a write-only mode to a read-write mode since we've
961 # already written data to the file.
962 if mode == 'w+':
963 mode = 'r+'
965 marray = numpy.memmap(filename, dtype=dtype, shape=shape, order=order,
966 mode=mode, offset=offset)
968 return marray
971def _read_bytes(fp, size, error_template="ran out of data"):
972 """
973 Read from file-like object until size bytes are read.
974 Raises ValueError if not EOF is encountered before size bytes are read.
975 Non-blocking objects only supported if they derive from io objects.
977 Required as e.g. ZipExtFile in python 2.6 can return less data than
978 requested.
979 """
980 data = bytes()
981 while True:
982 # io files (default in python3) return None or raise on
983 # would-block, python2 file will truncate, probably nothing can be
984 # done about that. note that regular files can't be non-blocking
985 try:
986 r = fp.read(size - len(data))
987 data += r
988 if len(r) == 0 or len(data) == size:
989 break
990 except BlockingIOError:
991 pass
992 if len(data) != size:
993 msg = "EOF: reading %s, expected %d bytes got %d"
994 raise ValueError(msg % (error_template, size, len(data)))
995 else:
996 return data
999def isfileobj(f):
1000 if not isinstance(f, (io.FileIO, io.BufferedReader, io.BufferedWriter)):
1001 return False
1002 try:
1003 # BufferedReader/Writer may raise OSError when
1004 # fetching `fileno()` (e.g. when wrapping BytesIO).
1005 f.fileno()
1006 return True
1007 except OSError:
1008 return False