Your IP : 216.73.216.85
ש�W~ � @ s� d Z d d l m Z d d l Z d d l Z d d l m Z d d l Z d d l Z d d l Z d d l
Z
d d l m Z d d l
m Z y d d l Z Wn d d l Z Yn Xd d l m Z d d l m Z d d
l m Z m Z d d l m Z m Z d d l m Z d d
l m Z m Z m Z m! Z! d d l" m# Z# d d l$ m% Z% d d l m& Z& d d l m' Z' d e d e d e! i Z( d a) d Z* e j+ � Z, d d � Z- e d/ d d � � Z. e/ e d � r e j0 j1 d d � j2 � p�d Z3 e j4 d e3 � Z5 n d Z5 Gd d � d e6 � Z7 d d � Z8 d! d" � Z9 d# d$ d% � Z: Gd& d' � d' e6 � Z; d( d) d* � Z< d0 d+ d, � Z= Gd- d. � d. e � Z> d S)1z+
Helpers for embarrassingly parallel code.
� )�divisionN)�sqrt)�Integral)�contextmanager� )�mp)�format_outer_frames)�Logger�short_format_time)�TransportableException�
_mk_exception)�memstr_to_bytes)�FallbackToBackend�MultiprocessingBackend�ThreadingBackend�SequentialBackend)� _basestring)�getfullargspec)�AutoBatchingMixin)�ParallelBackendBase�multiprocessing� threading�
sequentialc C s9 t t d d � } | d k r" | St t � } | t f S)z!Return the active default backend�backend_and_jobsN)�getattr�_backend�BACKENDS�DEFAULT_BACKEND�DEFAULT_N_JOBS)�active_backend_and_jobs�active_backend� r! �@/opt/alt/python35/lib/python3.5/site-packages/joblib/parallel.py�get_active_backend8 s
r# c
k s� t | t � r t | | � } t t d d � } z | | f t _ | | f VWd | d k r t t d d � d k r� t ` n | t _ Xd S)a� Change the default backend used by Parallel inside a with block.
If ``backend`` is a string it must match a previously registered
implementation using the ``register_parallel_backend`` function.
Alternatively backend can be passed directly as an instance.
By default all available workers will be used (``n_jobs=-1``) unless the
caller passes an explicit value for the ``n_jobs`` parameter.
This is an alternative to passing a ``backend='backend_name'`` argument to
the ``Parallel`` class constructor. It is particularly useful when calling
into library code that uses joblib internally but does not expose the
backend argument in its own API.
>>> from operator import neg
>>> with parallel_backend('threading'):
... print(Parallel()(delayed(neg)(i + 1) for i in range(5)))
...
[-1, -2, -3, -4, -5]
Warning: this function is experimental and subject to change in a future
version of joblib.
.. versionadded:: 0.10
r N)�
isinstancer r r r r )�backend�n_jobs�backend_params�old_backend_and_jobsr! r! r" �parallel_backendC s r) � get_contextZJOBLIB_START_METHOD� �methodc @ s: e Z d Z d Z d d � Z d d � Z d d � Z d S) � BatchedCallszCWrap a sequence of (func, args, kwargs) tuples as a single callablec C s% t | � | _ t | j � | _ d S)N)�list�items�len�_size)�self�iterator_slicer! r! r" �__init__~ s zBatchedCalls.__init__c C s d d � | j D� S)Nc S s( g | ] \ } } } | | | � � q Sr! r! )�.0�func�args�kwargsr! r! r" �
<listcomp>� s z)BatchedCalls.__call__.<locals>.<listcomp>)r/ )r2 r! r! r" �__call__� s zBatchedCalls.__call__c C s | j S)N)r1 )r2 r! r! r" �__len__� s zBatchedCalls.__len__N)�__name__�
__module__� __qualname__�__doc__r4 r: r; r! r! r! r" r- { s r- c C s t d k r d St j � S)zReturn the number of CPUs.Nr )r � cpu_countr! r! r! r" r@ � s r@ c C sv | s
d S| d k r d S| d k r* d Sd d | d } t | | � } t | d | � } t | � t | � k S) z� Returns False for indices increasingly apart, the distance
depending on the value of verbose.
We use a lag increasing as the square of index
T�
Fr g �?� � r )r �int)�index�verbose�scale�
next_scaler! r! r" �_verbosity_filter� s rI Tc sW | r t j � � � f d d � } y t j � � | � } Wn t k
rR Yn X| S)a! Decorator used to capture the arguments of a function.
Pass `check_pickle=False` when:
- performing a possibly repeated check is too costly and has been done
already once outside of the call to delayed.
- when used in conjunction `Parallel(backend='threading')`.
c s
� | | f S)Nr! )r7 r8 )�functionr! r" �delayed_function� s z!delayed.<locals>.delayed_function)�pickle�dumps� functools�wraps�AttributeError)rJ � check_picklerK r! )rJ r" �delayed� s
rR c @ s. e Z d Z d Z d d � Z d d � Z d S)�BatchCompletionCallBacka_ Callback used by joblib.Parallel's multiprocessing backend.
This callable is executed by the parent process whenever a worker process
has returned the results of a batch of tasks.
It is used for progress reporting, to update estimate of the batch
processing duration and to schedule the next batch of tasks to be
processed.
c C s | | _ | | _ | | _ d S)N)�dispatch_timestamp�
batch_size�parallel)r2 rT rU rV r! r! r" r4 � s z BatchCompletionCallBack.__init__c C sq | j j | j 7_ t j � | j } | j j j | j | � | j j � | j j d k rm | j j � d S)N)
rV �n_completed_tasksrU �timerT r �batch_completed�print_progress�_original_iterator�
dispatch_next)r2 �out�this_batch_durationr! r! r" r: � s
z BatchCompletionCallBack.__call__N)r< r= r> r? r4 r: r! r! r! r" rS � s
rS Fc C s | t | <| r | a d S)a� Register a new Parallel backend factory.
The new backend can then be selected by passing its name as the backend
argument to the Parallel class. Moreover, the default backend can be
overwritten globally by setting make_default=True.
The factory can be any callable that takes no argument and return an
instance of ``ParallelBackendBase``.
Warning: this function is experimental and subject to change in a future
version of joblib.
.. versionadded:: 0.10
N)r r )�name�factory� make_defaultr! r! r" �register_parallel_backend� s
rb c C s t � \ } } | j d | � S)am Determine the number of jobs that can actually run in parallel
n_jobs is the is the number of workers requested by the callers.
Passing n_jobs=-1 means requesting all available workers for instance
matching the number of CPU cores on the worker host(s).
This method should return a guesstimate of the number of workers that can
actually perform work concurrently with the currently enabled default
backend. The primary use case is to make it possible for the caller to know
in how many chunks to slice the work.
In general working on larger data chunks is more efficient (less
scheduling overhead and better use of CPU cache prefetching heuristics)
as long as all the workers have enough work to do.
Warning: this function is experimental and subject to change in a future
version of joblib.
.. versionadded:: 0.10
r& )r# �effective_n_jobs)r& r% �_r! r! r" rc � s rc c @ s� e Z d Z d Z d d d d d d d d d d d
� Z d d � Z d
d � Z d d � Z d d � Z d d � Z d d � Z
d d � Z d d � Z d d � Z
d d � Z d d � Z d! d" � Z d# d$ � Z d S)%�Parallela<' Helper class for readable parallel mapping.
Parameters
-----------
n_jobs: int, default: 1
The maximum number of concurrently running jobs, such as the number
of Python worker processes when backend="multiprocessing"
or the size of the thread-pool when backend="threading".
If -1 all CPUs are used. If 1 is given, no parallel computing code
is used at all, which is useful for debugging. For n_jobs below -1,
(n_cpus + 1 + n_jobs) are used. Thus for n_jobs = -2, all
CPUs but one are used.
backend: str or None, default: 'multiprocessing'
Specify the parallelization backend implementation.
Supported backends are:
- "multiprocessing" used by default, can induce some
communication and memory overhead when exchanging input and
output data with the worker Python processes.
- "threading" is a very low-overhead backend but it suffers
from the Python Global Interpreter Lock if the called function
relies a lot on Python objects. "threading" is mostly useful
when the execution bottleneck is a compiled extension that
explicitly releases the GIL (for instance a Cython loop wrapped
in a "with nogil" block or an expensive call to a library such
as NumPy).
- finally, you can register backends by calling
register_parallel_backend. This will allow you to implement
a backend of your liking.
verbose: int, optional
The verbosity level: if non zero, progress messages are
printed. Above 50, the output is sent to stdout.
The frequency of the messages increases with the verbosity level.
If it more than 10, all iterations are reported.
timeout: float, optional
Timeout limit for each task to complete. If any task takes longer
a TimeOutError will be raised. Only applied when n_jobs != 1
pre_dispatch: {'all', integer, or expression, as in '3*n_jobs'}
The number of batches (of tasks) to be pre-dispatched.
Default is '2*n_jobs'. When batch_size="auto" this is reasonable
default and the multiprocessing workers should never starve.
batch_size: int or 'auto', default: 'auto'
The number of atomic tasks to dispatch at once to each
worker. When individual evaluations are very fast, multiprocessing
can be slower than sequential computation because of the overhead.
Batching fast computations together can mitigate this.
The ``'auto'`` strategy keeps track of the time it takes for a batch
to complete, and dynamically adjusts the batch size to keep the time
on the order of half a second, using a heuristic. The initial batch
size is 1.
``batch_size="auto"`` with ``backend="threading"`` will dispatch
batches of a single task at a time as the threading backend has
very little overhead and using larger batch size has not proved to
bring any gain in that case.
temp_folder: str, optional
Folder to be used by the pool for memmaping large arrays
for sharing memory with worker processes. If None, this will try in
order:
- a folder pointed by the JOBLIB_TEMP_FOLDER environment variable,
- /dev/shm if the folder exists and is writable: this is a RAMdisk
filesystem available by default on modern Linux distributions,
- the default system temporary folder that can be overridden
with TMP, TMPDIR or TEMP environment variables, typically /tmp
under Unix operating systems.
Only active when backend="multiprocessing".
max_nbytes int, str, or None, optional, 1M by default
Threshold on the size of arrays passed to the workers that
triggers automated memory mapping in temp_folder. Can be an int
in Bytes, or a human-readable string, e.g., '1M' for 1 megabyte.
Use None to disable memmaping of large arrays.
Only active when backend="multiprocessing".
mmap_mode: {None, 'r+', 'r', 'w+', 'c'}
Memmapping mode for numpy arrays passed to workers.
See 'max_nbytes' parameter documentation for more details.
Notes
-----
This object uses the multiprocessing module to compute in
parallel the application of a function to many different
arguments. The main functionality it brings in addition to
using the raw multiprocessing API are (see examples for details):
* More readable code, in particular since it avoids
constructing list of arguments.
* Easier debugging:
- informative tracebacks even when the error happens on
the client side
- using 'n_jobs=1' enables to turn off parallel computing
for debugging without changing the codepath
- early capture of pickling errors
* An optional progress meter.
* Interruption of multiprocesses jobs with 'Ctrl-C'
* Flexible pickling control for the communication to and from
the worker processes.
* Ability to use shared memory efficiently with worker
processes for large numpy-based datastructures.
Examples
--------
A simple example:
>>> from math import sqrt
>>> from joblib import Parallel, delayed
>>> Parallel(n_jobs=1)(delayed(sqrt)(i**2) for i in range(10))
[0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0]
Reshaping the output when the function has several return
values:
>>> from math import modf
>>> from joblib import Parallel, delayed
>>> r = Parallel(n_jobs=1)(delayed(modf)(i/2.) for i in range(10))
>>> res, i = zip(*r)
>>> res
(0.0, 0.5, 0.0, 0.5, 0.0, 0.5, 0.0, 0.5, 0.0, 0.5)
>>> i
(0.0, 0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 3.0, 4.0, 4.0)
The progress meter: the higher the value of `verbose`, the more
messages::
>>> from time import sleep
>>> from joblib import Parallel, delayed
>>> r = Parallel(n_jobs=2, verbose=5)(delayed(sleep)(.1) for _ in range(10)) #doctest: +SKIP
[Parallel(n_jobs=2)]: Done 1 out of 10 | elapsed: 0.1s remaining: 0.9s
[Parallel(n_jobs=2)]: Done 3 out of 10 | elapsed: 0.2s remaining: 0.5s
[Parallel(n_jobs=2)]: Done 6 out of 10 | elapsed: 0.3s remaining: 0.2s
[Parallel(n_jobs=2)]: Done 9 out of 10 | elapsed: 0.5s remaining: 0.1s
[Parallel(n_jobs=2)]: Done 10 out of 10 | elapsed: 0.5s finished
Traceback example, note how the line of the error is indicated
as well as the values of the parameter passed to the function that
triggered the exception, even though the traceback happens in the
child process::
>>> from heapq import nlargest
>>> from joblib import Parallel, delayed
>>> Parallel(n_jobs=2)(delayed(nlargest)(2, n) for n in (range(4), 'abcde', 3)) #doctest: +SKIP
#...
---------------------------------------------------------------------------
Sub-process traceback:
---------------------------------------------------------------------------
TypeError Mon Nov 12 11:37:46 2012
PID: 12934 Python 2.7.3: /usr/bin/python
...........................................................................
/usr/lib/python2.7/heapq.pyc in nlargest(n=2, iterable=3, key=None)
419 if n >= size:
420 return sorted(iterable, key=key, reverse=True)[:n]
421
422 # When key is none, use simpler decoration
423 if key is None:
--> 424 it = izip(iterable, count(0,-1)) # decorate
425 result = _nlargest(n, it)
426 return map(itemgetter(0), result) # undecorate
427
428 # General case, slowest method
TypeError: izip argument #1 must support iteration
___________________________________________________________________________
Using pre_dispatch in a producer/consumer situation, where the
data is generated on the fly. Note how the producer is first
called a 3 times before the parallel loop is initiated, and then
called to generate new data on the fly. In this case the total
number of iterations cannot be reported in the progress messages::
>>> from math import sqrt
>>> from joblib import Parallel, delayed
>>> def producer():
... for i in range(6):
... print('Produced %s' % i)
... yield i
>>> out = Parallel(n_jobs=2, verbose=100, pre_dispatch='1.5*n_jobs')(
... delayed(sqrt)(i) for i in producer()) #doctest: +SKIP
Produced 0
Produced 1
Produced 2
[Parallel(n_jobs=2)]: Done 1 jobs | elapsed: 0.0s
Produced 3
[Parallel(n_jobs=2)]: Done 2 jobs | elapsed: 0.0s
Produced 4
[Parallel(n_jobs=2)]: Done 3 jobs | elapsed: 0.0s
Produced 5
[Parallel(n_jobs=2)]: Done 4 jobs | elapsed: 0.0s
[Parallel(n_jobs=2)]: Done 5 out of 6 | elapsed: 0.0s remaining: 0.0s
[Parallel(n_jobs=2)]: Done 6 out of 6 | elapsed: 0.0s finished
r Nr z
2 * n_jobs�auto�1M�rc
C s� t � \ }
} | d k r- | d k r- | } | | _ | | _ | | _ | | _ t | t � rl t | � } t d | d | d | d t d | j d � � | _
t d k r� t | j
d <| d k r� |
} n� t | d � rt | d
� r| | j
d <t
� } nN y t | } Wn4 t k
rIt d | t t j � � f � � Yn X| � } | d k szt | t � r�| d k r�| | _ n t d
| � � | | _ d | _ t � | _ d | _ t j � | _ d S)Nr �
max_nbytes� mmap_mode� temp_folderrF r �2 �context�Pool�Lockz'Invalid backend: %s, expected one of %rrf z8batch_size must be 'auto' or a positive integer, got: %rF) r# r& rF �timeout� pre_dispatchr$ r r
�dict�max�
_backend_args�DEFAULT_MP_CONTEXT�hasattrr r �KeyError�
ValueError�sorted�keysr rU r �_outputr. �_jobs�_managed_backendr ro �_lock)
r2 r& r% rF rp rq rU rk ri rj r �default_n_jobs�backend_factoryr! r! r" r4 � sL
!
zParallel.__init__c C s d | _ | j � | S)NT)r} �_initialize_backend)r2 r! r! r" � __enter__ s
zParallel.__enter__c C s | j � d | _ d S)NF)�_terminate_backendr} )r2 �exc_type� exc_value� tracebackr! r! r" �__exit__ s
zParallel.__exit__c C sh y& | j j d | j d | | j � SWn; t k
rc } z | j | _ | j � SWYd d } ~ Xn Xd S)z?Build a process or thread pool and return the number of workersr& rV N)r � configurer& rt r r% r� )r2 �er! r! r" r� s z Parallel._initialize_backendc C s | j r | j j | j � Sd S)Nr )r rc r& )r2 r! r! r" �_effective_n_jobs" s zParallel._effective_n_jobsc C s | j d k r | j j � d S)N)r � terminate)r2 r! r! r" r� ' s zParallel._terminate_backendc C s� | j r
d S| j t | � 7_ | j d 7_ t j � } t | t | � | � } | j j | d | �} | j j | � d S)z�Queue the batch for computing, with or without multiprocessing
WARNING: this method is not thread-safe: it should be only called
indirectly via dispatch_one_batch.
Nr �callback)
� _aborting�n_dispatched_tasksr0 �n_dispatched_batchesrX rS r � apply_asyncr| �append)r2 �batchrT �cb�jobr! r! r" � _dispatch+ s zParallel._dispatchc C s( | j | j � s$ d | _ d | _ d S)a Dispatch more data for parallel processing
This method is meant to be called concurrently by the multiprocessing
callback. We rely on the thread-safety of dispatch_one_batch to protect
against concurrent consumption of the unprotected iterator.
FN)�dispatch_one_batchr[ �
_iterating)r2 r! r! r" r\ > s zParallel.dispatch_nextc C s~ | j d k r! | j j � } n | j } | j �D t t j | | � � } t | � d k rb d S| j | � d SWd QRXd S)aT Prefetch the tasks for the next batch and dispatch them.
The effective size of the batch is computed here.
If there are no more jobs to dispatch, return False, else return True.
The iterator consumption and dispatching is protected by the same
lock so calling this function should be thread safe.
rf r FTN) rU r �compute_batch_sizer~ r- � itertools�islicer0 r� )r2 �iteratorrU �tasksr! r! r" r� J s
zParallel.dispatch_one_batchc C sY | j s
d S| j d k r+ t j j } n t j j } | | } | d | | f � d S)z=Display the message on stout or stderr depending on verbosityNrl z [%s]: %s
)rF �sys�stderr�write�stdout)r2 �msg�msg_args�writerr! r! r" �_printc s
zParallel._printc C s | j s
d St j � | j } | j d k rj t | j | j � rH d S| j d | j t | � f � n� | j } | j } | d k s� | | d | j
} | | j d } | d | k } | s� | | r� d S| | | j | d } | j d | | t | � t | � f � d S)zvDisplay the process of the parallel execution only a fraction
of time, controlled by self.verbose.
Nz!Done %3i tasks | elapsed: %sr r g �?z/Done %3i out of %3i | elapsed: %s remaining: %s) rF rX � _start_timer[ rI r� r� rW r
r� �_pre_dispatch_amount)r2 � elapsed_timerE � total_tasks�cursor� frequency� is_last_item�remaining_timer! r! r" rZ p s2
zParallel.print_progressc C s� t � | _ x�| j s- t | j � d k r�t | j � d k rR t j d � q | j � | j j d � } Wd QRXyT d t | j
� j k r� | j j | j
d | j
� � n | j j | j
� � Wq t k
r�} z� d | _ t | t � r=t d d d d � } d | | j f } t | j � d } | | � } | j } | d k rzt | d
� rz| j } | j d | � | � WYd d } ~ Xq Xq Wd S) Nr g{�G�z�?rp Trm rA � stack_startr z�Multiprocessing exception:
%s
---------------------------------------------------------------------------
Sub-process traceback:
---------------------------------------------------------------------------
%s�abort_everything� ensure_ready)r. r{ r� r0 r| rX �sleepr~ �popr �getr7 �extendrp �
BaseExceptionr� r$ r r �messager �etyper rv r} r� )r2 r� � exception� this_report�report�exception_typer% r� r! r! r" �retrieve� s2 !
" zParallel.retrievec C s� | j r t d � � d | _ | j s6 | j � } n | j � } t | � } | j } | d k so | d k r� d | _ d | _ nI | | _ t
| d � r� t | � } t | � | _ } t
j | | � } t j � | _ d | _ d | _ d | _ z� x% | j | � rd | _ q� Wd | _ | d k s:| d k rCd | _ | j � t j � | j } | j d t | j � t | j � t | � f � Wd | j s�| j � t � | _ X| j } d | _ | S) Nz)This Parallel instance is already runningF�allr r �endswithTz*Done %3i out of %3i | elapsed: %s finished) r| rx r� r} r� r� �iterrq r[ r� rv �evalrD r� r� rX r� r� r� rW r� r� r� r� r0 r{ r
r� r. )r2 �iterabler& r� rq r� �outputr! r! r" r: � sJ
zParallel.__call__c C s d | j j | j f S)Nz
%s(n_jobs=%s))� __class__r< r& )r2 r! r! r" �__repr__ s zParallel.__repr__)r< r= r> r? r4 r� r� r� r� r� r� r\ r� r� rZ r� r: r� r! r! r! r" re s" � 8
+ 6 =re ���r� )?r? �
__future__r �osr� �mathr rN rX r r� �numbersr �
contextlibr �cPicklerL �_multiprocessing_helpersr � format_stackr �loggerr r
�
my_exceptionsr r �diskr
�_parallel_backendsr r r r �_compatr � func_inspectr r r r r r �localr r# r) rv �environr� �stripr, r* ru �objectr- r@ rI rR rS rb rc re r! r! r! r" �<module> sZ " 0!