/usr/lib/python2.7/dist-packages/joblib/memory.py is in python-joblib 0.11-1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 | """
A context object for caching a function's return value each time it
is called with the same input arguments.
"""
# Author: Gael Varoquaux <gael dot varoquaux at normalesup dot org>
# Copyright (c) 2009 Gael Varoquaux
# License: BSD Style, 3 clauses.
from __future__ import with_statement
import os
import shutil
import time
import pydoc
import re
import functools
import traceback
import warnings
import inspect
import json
import weakref
import io
import operator
import collections
import datetime
import threading
# Local imports
from . import hashing
from .func_inspect import get_func_code, get_func_name, filter_args
from .func_inspect import format_call
from .func_inspect import format_signature
from ._memory_helpers import open_py_source
from .logger import Logger, format_time, pformat
from . import numpy_pickle
from .disk import mkdirp, rm_subdirs, memstr_to_bytes
from ._compat import _basestring, PY3_OR_LATER
from .backports import concurrency_safe_rename
FIRST_LINE_TEXT = "# first line:"
CacheItemInfo = collections.namedtuple('CacheItemInfo',
'path size last_access')
# TODO: The following object should have a data store object as a sub
# object, and the interface to persist and query should be separated in
# the data store.
#
# This would enable creating 'Memory' objects with a different logic for
# pickling that would simply span a MemorizedFunc with the same
# store (or do we want to copy it to avoid cross-talks?), for instance to
# implement HDF5 pickling.
# TODO: Same remark for the logger, and probably use the Python logging
# mechanism.
def extract_first_line(func_code):
""" Extract the first line information from the function code
text if available.
"""
if func_code.startswith(FIRST_LINE_TEXT):
func_code = func_code.split('\n')
first_line = int(func_code[0][len(FIRST_LINE_TEXT):])
func_code = '\n'.join(func_code[1:])
else:
first_line = -1
return func_code, first_line
class JobLibCollisionWarning(UserWarning):
""" Warn that there might be a collision between names of functions.
"""
def _get_func_fullname(func):
"""Compute the part of part associated with a function.
See code of_cache_key_to_dir() for details
"""
modules, funcname = get_func_name(func)
modules.append(funcname)
return os.path.join(*modules)
def _cache_key_to_dir(cachedir, func, argument_hash):
"""Compute directory associated with a given cache key.
func can be a function or a string as returned by _get_func_fullname().
"""
parts = [cachedir]
if isinstance(func, _basestring):
parts.append(func)
else:
parts.append(_get_func_fullname(func))
if argument_hash is not None:
parts.append(argument_hash)
return os.path.join(*parts)
def _load_output(output_dir, func_name, timestamp=None, metadata=None,
mmap_mode=None, verbose=0):
"""Load output of a computation."""
if verbose > 1:
signature = ""
try:
if metadata is not None:
args = ", ".join(['%s=%s' % (name, value)
for name, value
in metadata['input_args'].items()])
signature = "%s(%s)" % (os.path.basename(func_name),
args)
else:
signature = os.path.basename(func_name)
except KeyError:
pass
if timestamp is not None:
t = "% 16s" % format_time(time.time() - timestamp)
else:
t = ""
if verbose < 10:
print('[Memory]%s: Loading %s...' % (t, str(signature)))
else:
print('[Memory]%s: Loading %s from %s' % (
t, str(signature), output_dir))
filename = os.path.join(output_dir, 'output.pkl')
if not os.path.isfile(filename):
raise KeyError(
"Non-existing cache value (may have been cleared).\n"
"File %s does not exist" % filename)
result = numpy_pickle.load(filename, mmap_mode=mmap_mode)
return result
def _get_cache_items(root_path):
"""Get cache information for reducing the size of the cache."""
cache_items = []
for dirpath, dirnames, filenames in os.walk(root_path):
is_cache_hash_dir = re.match('[a-f0-9]{32}', os.path.basename(dirpath))
if is_cache_hash_dir:
output_filename = os.path.join(dirpath, 'output.pkl')
try:
last_access = os.path.getatime(output_filename)
except OSError:
try:
last_access = os.path.getatime(dirpath)
except OSError:
# The directory has already been deleted
continue
last_access = datetime.datetime.fromtimestamp(last_access)
try:
full_filenames = [os.path.join(dirpath, fn)
for fn in filenames]
dirsize = sum(os.path.getsize(fn)
for fn in full_filenames)
except OSError:
# Either output_filename or one of the files in
# dirpath does not exist any more. We assume this
# directory is being cleaned by another process already
continue
cache_items.append(CacheItemInfo(dirpath, dirsize, last_access))
return cache_items
def _get_cache_items_to_delete(root_path, bytes_limit):
"""Get cache items to delete to keep the cache under a size limit."""
if isinstance(bytes_limit, _basestring):
bytes_limit = memstr_to_bytes(bytes_limit)
cache_items = _get_cache_items(root_path)
cache_size = sum(item.size for item in cache_items)
to_delete_size = cache_size - bytes_limit
if to_delete_size < 0:
return []
# We want to delete first the cache items that were accessed a
# long time ago
cache_items.sort(key=operator.attrgetter('last_access'))
cache_items_to_delete = []
size_so_far = 0
for item in cache_items:
if size_so_far > to_delete_size:
break
cache_items_to_delete.append(item)
size_so_far += item.size
return cache_items_to_delete
def concurrency_safe_write(to_write, filename, write_func):
"""Writes an object into a file in a concurrency-safe way."""
thread_id = id(threading.current_thread())
temporary_filename = '{}.thread-{}-pid-{}'.format(
filename, thread_id, os.getpid())
write_func(to_write, temporary_filename)
concurrency_safe_rename(temporary_filename, filename)
# An in-memory store to avoid looking at the disk-based function
# source code to check if a function definition has changed
_FUNCTION_HASHES = weakref.WeakKeyDictionary()
###############################################################################
# class `MemorizedResult`
###############################################################################
class MemorizedResult(Logger):
"""Object representing a cached value.
Attributes
----------
cachedir: string
path to root of joblib cache
func: function or string
function whose output is cached. The string case is intended only for
instanciation based on the output of repr() on another instance.
(namely eval(repr(memorized_instance)) works).
argument_hash: string
hash of the function arguments
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}
The memmapping mode used when loading from cache numpy arrays. See
numpy.load for the meaning of the different values.
verbose: int
verbosity level (0 means no message)
timestamp, metadata: string
for internal use only
"""
def __init__(self, cachedir, func, argument_hash,
mmap_mode=None, verbose=0, timestamp=None, metadata=None):
Logger.__init__(self)
if isinstance(func, _basestring):
self.func = func
else:
self.func = _get_func_fullname(func)
self.argument_hash = argument_hash
self.cachedir = cachedir
self.mmap_mode = mmap_mode
self._output_dir = _cache_key_to_dir(cachedir, self.func,
argument_hash)
if metadata is not None:
self.metadata = metadata
else:
self.metadata = {}
# No error is relevant here.
try:
with open(os.path.join(self._output_dir, 'metadata.json'),
'rb') as f:
self.metadata = json.load(f)
except:
pass
self.duration = self.metadata.get('duration', None)
self.verbose = verbose
self.timestamp = timestamp
def get(self):
"""Read value from cache and return it."""
return _load_output(self._output_dir, _get_func_fullname(self.func),
timestamp=self.timestamp,
metadata=self.metadata, mmap_mode=self.mmap_mode,
verbose=self.verbose)
def clear(self):
"""Clear value from cache"""
shutil.rmtree(self._output_dir, ignore_errors=True)
def __repr__(self):
return ('{class_name}(cachedir="{cachedir}", func="{func}", '
'argument_hash="{argument_hash}")'.format(
class_name=self.__class__.__name__,
cachedir=self.cachedir,
func=self.func,
argument_hash=self.argument_hash
))
def __reduce__(self):
return (self.__class__, (self.cachedir, self.func, self.argument_hash),
{'mmap_mode': self.mmap_mode})
class NotMemorizedResult(object):
"""Class representing an arbitrary value.
This class is a replacement for MemorizedResult when there is no cache.
"""
__slots__ = ('value', 'valid')
def __init__(self, value):
self.value = value
self.valid = True
def get(self):
if self.valid:
return self.value
else:
raise KeyError("No value stored.")
def clear(self):
self.valid = False
self.value = None
def __repr__(self):
if self.valid:
return '{class_name}({value})'.format(
class_name=self.__class__.__name__,
value=pformat(self.value)
)
else:
return self.__class__.__name__ + ' with no value'
# __getstate__ and __setstate__ are required because of __slots__
def __getstate__(self):
return {"valid": self.valid, "value": self.value}
def __setstate__(self, state):
self.valid = state["valid"]
self.value = state["value"]
###############################################################################
# class `NotMemorizedFunc`
###############################################################################
class NotMemorizedFunc(object):
"""No-op object decorating a function.
This class replaces MemorizedFunc when there is no cache. It provides an
identical API but does not write anything on disk.
Attributes
----------
func: callable
Original undecorated function.
"""
# Should be a light as possible (for speed)
def __init__(self, func):
self.func = func
def __call__(self, *args, **kwargs):
return self.func(*args, **kwargs)
def call_and_shelve(self, *args, **kwargs):
return NotMemorizedResult(self.func(*args, **kwargs))
def __reduce__(self):
return (self.__class__, (self.func,))
def __repr__(self):
return '%s(func=%s)' % (
self.__class__.__name__,
self.func
)
def clear(self, warn=True):
# Argument "warn" is for compatibility with MemorizedFunc.clear
pass
###############################################################################
# class `MemorizedFunc`
###############################################################################
class MemorizedFunc(Logger):
""" Callable object decorating a function for caching its return value
each time it is called.
All values are cached on the filesystem, in a deep directory
structure. Methods are provided to inspect the cache or clean it.
Attributes
----------
func: callable
The original, undecorated, function.
cachedir: string
Path to the base cache directory of the memory context.
ignore: list or None
List of variable names to ignore when choosing whether to
recompute.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}
The memmapping mode used when loading from cache
numpy arrays. See numpy.load for the meaning of the different
values.
compress: boolean, or integer
Whether to zip the stored data on disk. If an integer is
given, it should be between 1 and 9, and sets the amount
of compression. Note that compressed arrays cannot be
read by memmapping.
verbose: int, optional
The verbosity flag, controls messages that are issued as
the function is evaluated.
"""
#-------------------------------------------------------------------------
# Public interface
#-------------------------------------------------------------------------
def __init__(self, func, cachedir, ignore=None, mmap_mode=None,
compress=False, verbose=1, timestamp=None):
"""
Parameters
----------
func: callable
The function to decorate
cachedir: string
The path of the base directory to use as a data store
ignore: list or None
List of variable names to ignore.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}, optional
The memmapping mode used when loading from cache
numpy arrays. See numpy.load for the meaning of the
arguments.
compress : boolean, or integer
Whether to zip the stored data on disk. If an integer is
given, it should be between 1 and 9, and sets the amount
of compression. Note that compressed arrays cannot be
read by memmapping.
verbose: int, optional
Verbosity flag, controls the debug messages that are issued
as functions are evaluated. The higher, the more verbose
timestamp: float, optional
The reference time from which times in tracing messages
are reported.
"""
Logger.__init__(self)
self.mmap_mode = mmap_mode
self.func = func
if ignore is None:
ignore = []
self.ignore = ignore
self._verbose = verbose
self.cachedir = cachedir
self.compress = compress
if compress and self.mmap_mode is not None:
warnings.warn('Compressed results cannot be memmapped',
stacklevel=2)
if timestamp is None:
timestamp = time.time()
self.timestamp = timestamp
mkdirp(self.cachedir)
try:
functools.update_wrapper(self, func)
except:
" Objects like ufunc don't like that "
if inspect.isfunction(func):
doc = pydoc.TextDoc().document(func)
# Remove blank line
doc = doc.replace('\n', '\n\n', 1)
# Strip backspace-overprints for compatibility with autodoc
doc = re.sub('\x08.', '', doc)
else:
# Pydoc does a poor job on other objects
doc = func.__doc__
self.__doc__ = 'Memoized version of %s' % doc
def _cached_call(self, args, kwargs):
"""Call wrapped function and cache result, or read cache if available.
This function returns the wrapped function output and some metadata.
Returns
-------
output: value or tuple
what is returned by wrapped function
argument_hash: string
hash of function arguments
metadata: dict
some metadata about wrapped function call (see _persist_input())
"""
# Compare the function code with the previous to see if the
# function code has changed
output_dir, argument_hash = self._get_output_dir(*args, **kwargs)
metadata = None
output_pickle_path = os.path.join(output_dir, 'output.pkl')
# FIXME: The statements below should be try/excepted
if not (self._check_previous_func_code(stacklevel=4) and
os.path.isfile(output_pickle_path)):
if self._verbose > 10:
_, name = get_func_name(self.func)
self.warn('Computing func %s, argument hash %s in '
'directory %s'
% (name, argument_hash, output_dir))
out, metadata = self.call(*args, **kwargs)
if self.mmap_mode is not None:
# Memmap the output at the first call to be consistent with
# later calls
out = _load_output(output_dir, _get_func_fullname(self.func),
timestamp=self.timestamp,
mmap_mode=self.mmap_mode,
verbose=self._verbose)
else:
try:
t0 = time.time()
out = _load_output(output_dir, _get_func_fullname(self.func),
timestamp=self.timestamp,
metadata=metadata, mmap_mode=self.mmap_mode,
verbose=self._verbose)
if self._verbose > 4:
t = time.time() - t0
_, name = get_func_name(self.func)
msg = '%s cache loaded - %s' % (name, format_time(t))
print(max(0, (80 - len(msg))) * '_' + msg)
except Exception:
# XXX: Should use an exception logger
_, signature = format_signature(self.func, *args, **kwargs)
self.warn('Exception while loading results for '
'{}\n {}'.format(
signature, traceback.format_exc()))
out, metadata = self.call(*args, **kwargs)
argument_hash = None
return (out, argument_hash, metadata)
def call_and_shelve(self, *args, **kwargs):
"""Call wrapped function, cache result and return a reference.
This method returns a reference to the cached result instead of the
result itself. The reference object is small and pickeable, allowing
to send or store it easily. Call .get() on reference object to get
result.
Returns
-------
cached_result: MemorizedResult or NotMemorizedResult
reference to the value returned by the wrapped function. The
class "NotMemorizedResult" is used when there is no cache
activated (e.g. cachedir=None in Memory).
"""
_, argument_hash, metadata = self._cached_call(args, kwargs)
return MemorizedResult(self.cachedir, self.func, argument_hash,
metadata=metadata, verbose=self._verbose - 1,
timestamp=self.timestamp)
def __call__(self, *args, **kwargs):
return self._cached_call(args, kwargs)[0]
def __reduce__(self):
""" We don't store the timestamp when pickling, to avoid the hash
depending from it.
In addition, when unpickling, we run the __init__
"""
return (self.__class__, (self.func, self.cachedir, self.ignore,
self.mmap_mode, self.compress, self._verbose))
#-------------------------------------------------------------------------
# Private interface
#-------------------------------------------------------------------------
def _get_argument_hash(self, *args, **kwargs):
return hashing.hash(filter_args(self.func, self.ignore,
args, kwargs),
coerce_mmap=(self.mmap_mode is not None))
def _get_output_dir(self, *args, **kwargs):
""" Return the directory in which are persisted the result
of the function called with the given arguments.
"""
argument_hash = self._get_argument_hash(*args, **kwargs)
output_dir = os.path.join(self._get_func_dir(self.func),
argument_hash)
return output_dir, argument_hash
get_output_dir = _get_output_dir # backward compatibility
def _get_func_dir(self, mkdir=True):
""" Get the directory corresponding to the cache for the
function.
"""
func_dir = _cache_key_to_dir(self.cachedir, self.func, None)
if mkdir:
mkdirp(func_dir)
return func_dir
def _hash_func(self):
"""Hash a function to key the online cache"""
func_code_h = hash(getattr(self.func, '__code__', None))
return id(self.func), hash(self.func), func_code_h
def _write_func_code(self, filename, func_code, first_line):
""" Write the function code and the filename to a file.
"""
# We store the first line because the filename and the function
# name is not always enough to identify a function: people
# sometimes have several functions named the same way in a
# file. This is bad practice, but joblib should be robust to bad
# practice.
func_code = u'%s %i\n%s' % (FIRST_LINE_TEXT, first_line, func_code)
with io.open(filename, 'w', encoding="UTF-8") as out:
out.write(func_code)
# Also store in the in-memory store of function hashes
is_named_callable = False
if PY3_OR_LATER:
is_named_callable = (hasattr(self.func, '__name__')
and self.func.__name__ != '<lambda>')
else:
is_named_callable = (hasattr(self.func, 'func_name')
and self.func.func_name != '<lambda>')
if is_named_callable:
# Don't do this for lambda functions or strange callable
# objects, as it ends up being too fragile
func_hash = self._hash_func()
try:
_FUNCTION_HASHES[self.func] = func_hash
except TypeError:
# Some callable are not hashable
pass
def _check_previous_func_code(self, stacklevel=2):
"""
stacklevel is the depth a which this function is called, to
issue useful warnings to the user.
"""
# First check if our function is in the in-memory store.
# Using the in-memory store not only makes things faster, but it
# also renders us robust to variations of the files when the
# in-memory version of the code does not vary
try:
if self.func in _FUNCTION_HASHES:
# We use as an identifier the id of the function and its
# hash. This is more likely to falsely change than have hash
# collisions, thus we are on the safe side.
func_hash = self._hash_func()
if func_hash == _FUNCTION_HASHES[self.func]:
return True
except TypeError:
# Some callables are not hashable
pass
# Here, we go through some effort to be robust to dynamically
# changing code and collision. We cannot inspect.getsource
# because it is not reliable when using IPython's magic "%run".
func_code, source_file, first_line = get_func_code(self.func)
func_dir = self._get_func_dir()
func_code_file = os.path.join(func_dir, 'func_code.py')
try:
with io.open(func_code_file, encoding="UTF-8") as infile:
old_func_code, old_first_line = \
extract_first_line(infile.read())
except IOError:
self._write_func_code(func_code_file, func_code, first_line)
return False
if old_func_code == func_code:
return True
# We have differing code, is this because we are referring to
# different functions, or because the function we are referring to has
# changed?
_, func_name = get_func_name(self.func, resolv_alias=False,
win_characters=False)
if old_first_line == first_line == -1 or func_name == '<lambda>':
if not first_line == -1:
func_description = '%s (%s:%i)' % (func_name,
source_file, first_line)
else:
func_description = func_name
warnings.warn(JobLibCollisionWarning(
"Cannot detect name collisions for function '%s'"
% func_description), stacklevel=stacklevel)
# Fetch the code at the old location and compare it. If it is the
# same than the code store, we have a collision: the code in the
# file has not changed, but the name we have is pointing to a new
# code block.
if not old_first_line == first_line and source_file is not None:
possible_collision = False
if os.path.exists(source_file):
_, func_name = get_func_name(self.func, resolv_alias=False)
num_lines = len(func_code.split('\n'))
with open_py_source(source_file) as f:
on_disk_func_code = f.readlines()[
old_first_line - 1:old_first_line - 1 + num_lines - 1]
on_disk_func_code = ''.join(on_disk_func_code)
possible_collision = (on_disk_func_code.rstrip()
== old_func_code.rstrip())
else:
possible_collision = source_file.startswith('<doctest ')
if possible_collision:
warnings.warn(JobLibCollisionWarning(
'Possible name collisions between functions '
"'%s' (%s:%i) and '%s' (%s:%i)" %
(func_name, source_file, old_first_line,
func_name, source_file, first_line)),
stacklevel=stacklevel)
# The function has changed, wipe the cache directory.
# XXX: Should be using warnings, and giving stacklevel
if self._verbose > 10:
_, func_name = get_func_name(self.func, resolv_alias=False)
self.warn("Function %s (stored in %s) has changed." %
(func_name, func_dir))
self.clear(warn=True)
return False
def clear(self, warn=True):
""" Empty the function's cache.
"""
func_dir = self._get_func_dir(mkdir=False)
if self._verbose > 0 and warn:
self.warn("Clearing cache %s" % func_dir)
if os.path.exists(func_dir):
shutil.rmtree(func_dir, ignore_errors=True)
mkdirp(func_dir)
func_code, _, first_line = get_func_code(self.func)
func_code_file = os.path.join(func_dir, 'func_code.py')
self._write_func_code(func_code_file, func_code, first_line)
def call(self, *args, **kwargs):
""" Force the execution of the function with the given arguments and
persist the output values.
"""
start_time = time.time()
output_dir, _ = self._get_output_dir(*args, **kwargs)
if self._verbose > 0:
print(format_call(self.func, args, kwargs))
output = self.func(*args, **kwargs)
self._persist_output(output, output_dir)
duration = time.time() - start_time
metadata = self._persist_input(output_dir, duration, args, kwargs)
if self._verbose > 0:
_, name = get_func_name(self.func)
msg = '%s - %s' % (name, format_time(duration))
print(max(0, (80 - len(msg))) * '_' + msg)
return output, metadata
# Make public
def _persist_output(self, output, dir):
""" Persist the given output tuple in the directory.
"""
try:
filename = os.path.join(dir, 'output.pkl')
mkdirp(dir)
write_func = functools.partial(numpy_pickle.dump,
compress=self.compress)
concurrency_safe_write(output, filename, write_func)
if self._verbose > 10:
print('Persisting in %s' % dir)
except OSError:
" Race condition in the creation of the directory "
def _persist_input(self, output_dir, duration, args, kwargs,
this_duration_limit=0.5):
""" Save a small summary of the call using json format in the
output directory.
output_dir: string
directory where to write metadata.
duration: float
time taken by hashing input arguments, calling the wrapped
function and persisting its output.
args, kwargs: list and dict
input arguments for wrapped function
this_duration_limit: float
Max execution time for this function before issuing a warning.
"""
start_time = time.time()
argument_dict = filter_args(self.func, self.ignore,
args, kwargs)
input_repr = dict((k, repr(v)) for k, v in argument_dict.items())
# This can fail due to race-conditions with multiple
# concurrent joblibs removing the file or the directory
metadata = {"duration": duration, "input_args": input_repr}
try:
mkdirp(output_dir)
filename = os.path.join(output_dir, 'metadata.json')
def write_func(output, dest_filename):
with open(dest_filename, 'w') as f:
json.dump(output, f)
concurrency_safe_write(metadata, filename, write_func)
except Exception:
pass
this_duration = time.time() - start_time
if this_duration > this_duration_limit:
# This persistence should be fast. It will not be if repr() takes
# time and its output is large, because json.dump will have to
# write a large file. This should not be an issue with numpy arrays
# for which repr() always output a short representation, but can
# be with complex dictionaries. Fixing the problem should be a
# matter of replacing repr() above by something smarter.
warnings.warn("Persisting input arguments took %.2fs to run.\n"
"If this happens often in your code, it can cause "
"performance problems \n"
"(results will be correct in all cases). \n"
"The reason for this is probably some large input "
"arguments for a wrapped\n"
" function (e.g. large strings).\n"
"THIS IS A JOBLIB ISSUE. If you can, kindly provide "
"the joblib's team with an\n"
" example so that they can fix the problem."
% this_duration, stacklevel=5)
return metadata
# XXX: Need a method to check if results are available.
#-------------------------------------------------------------------------
# Private `object` interface
#-------------------------------------------------------------------------
def __repr__(self):
return '%s(func=%s, cachedir=%s)' % (
self.__class__.__name__,
self.func,
repr(self.cachedir),
)
###############################################################################
# class `Memory`
###############################################################################
class Memory(Logger):
""" A context object for caching a function's return value each time it
is called with the same input arguments.
All values are cached on the filesystem, in a deep directory
structure.
see :ref:`memory_reference`
"""
#-------------------------------------------------------------------------
# Public interface
#-------------------------------------------------------------------------
def __init__(self, cachedir, mmap_mode=None, compress=False, verbose=1,
bytes_limit=None):
"""
Parameters
----------
cachedir: string or None
The path of the base directory to use as a data store
or None. If None is given, no caching is done and
the Memory object is completely transparent.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}, optional
The memmapping mode used when loading from cache
numpy arrays. See numpy.load for the meaning of the
arguments.
compress: boolean, or integer
Whether to zip the stored data on disk. If an integer is
given, it should be between 1 and 9, and sets the amount
of compression. Note that compressed arrays cannot be
read by memmapping.
verbose: int, optional
Verbosity flag, controls the debug messages that are issued
as functions are evaluated.
bytes_limit: int, optional
Limit in bytes of the size of the cache
"""
# XXX: Bad explanation of the None value of cachedir
Logger.__init__(self)
self._verbose = verbose
self.mmap_mode = mmap_mode
self.timestamp = time.time()
self.compress = compress
self.bytes_limit = bytes_limit
if compress and mmap_mode is not None:
warnings.warn('Compressed results cannot be memmapped',
stacklevel=2)
if cachedir is None:
self.cachedir = None
else:
self.cachedir = os.path.join(cachedir, 'joblib')
mkdirp(self.cachedir)
def cache(self, func=None, ignore=None, verbose=None,
mmap_mode=False):
""" Decorates the given function func to only compute its return
value for input arguments not cached on disk.
Parameters
----------
func: callable, optional
The function to be decorated
ignore: list of strings
A list of arguments name to ignore in the hashing
verbose: integer, optional
The verbosity mode of the function. By default that
of the memory object is used.
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}, optional
The memmapping mode used when loading from cache
numpy arrays. See numpy.load for the meaning of the
arguments. By default that of the memory object is used.
Returns
-------
decorated_func: MemorizedFunc object
The returned object is a MemorizedFunc object, that is
callable (behaves like a function), but offers extra
methods for cache lookup and management. See the
documentation for :class:`joblib.memory.MemorizedFunc`.
"""
if func is None:
# Partial application, to be able to specify extra keyword
# arguments in decorators
return functools.partial(self.cache, ignore=ignore,
verbose=verbose, mmap_mode=mmap_mode)
if self.cachedir is None:
return NotMemorizedFunc(func)
if verbose is None:
verbose = self._verbose
if mmap_mode is False:
mmap_mode = self.mmap_mode
if isinstance(func, MemorizedFunc):
func = func.func
return MemorizedFunc(func, cachedir=self.cachedir,
mmap_mode=mmap_mode,
ignore=ignore,
compress=self.compress,
verbose=verbose,
timestamp=self.timestamp)
def clear(self, warn=True):
""" Erase the complete cache directory.
"""
if warn:
self.warn('Flushing completely the cache')
if self.cachedir is not None:
rm_subdirs(self.cachedir)
def reduce_size(self):
"""Remove cache folders to make cache size fit in ``bytes_limit``."""
if self.cachedir is not None and self.bytes_limit is not None:
cache_items_to_delete = _get_cache_items_to_delete(
self.cachedir, self.bytes_limit)
for cache_item in cache_items_to_delete:
if self._verbose > 10:
print('Deleting cache item {}'.format(cache_item))
try:
shutil.rmtree(cache_item.path, ignore_errors=True)
except OSError:
# Even with ignore_errors=True can shutil.rmtree
# can raise OSErrror with [Errno 116] Stale file
# handle if another process has deleted the folder
# already.
pass
def eval(self, func, *args, **kwargs):
""" Eval function func with arguments `*args` and `**kwargs`,
in the context of the memory.
This method works similarly to the builtin `apply`, except
that the function is called only if the cache is not
up to date.
"""
if self.cachedir is None:
return func(*args, **kwargs)
return self.cache(func)(*args, **kwargs)
#-------------------------------------------------------------------------
# Private `object` interface
#-------------------------------------------------------------------------
def __repr__(self):
return '%s(cachedir=%s)' % (
self.__class__.__name__,
repr(self.cachedir),
)
def __reduce__(self):
""" We don't store the timestamp when pickling, to avoid the hash
depending from it.
In addition, when unpickling, we run the __init__
"""
# We need to remove 'joblib' from the end of cachedir
cachedir = self.cachedir[:-7] if self.cachedir is not None else None
return (self.__class__, (cachedir,
self.mmap_mode, self.compress, self._verbose))
|