functools — 고차 함수와 콜러블 객체에 대한 연산

소스 코드: Lib/functools.py


functools 모듈은 고차 함수를 위한 것입니다: 다른 함수에 작용하거나 다른 함수를 반환하는 함수. 일반적으로, 모든 콜러블 객체는 이 모듈의 목적상 함수로 취급될 수 있습니다.

functools 모듈은 다음 함수를 정의합니다:

@functools.cache(user_function)

단순하고 가벼운 무제한 함수 캐시. 때때로 “memoize”라고도 합니다.

lru_cache(maxsize=None)와 같은 것을 반환하여, 함수 인자의 딕셔너리 조회 주위로 얇은 래퍼를 만듭니다. 이전 값을 제거할 필요가 없어서, 크기 제한이 있는 lru_cache()보다 작고 빠릅니다.

예를 들면:

@cache
def factorial(n):
    return n * factorial(n-1) if n else 1

>>> factorial(10)      # no previously cached result, makes 11 recursive calls
3628800
>>> factorial(5)       # just looks up cached value result
120
>>> factorial(12)      # makes two new recursive calls, the other 10 are cached
479001600

버전 3.9에 추가.

@functools.cached_property(func)

클래스의 메서드를 값이 한 번 계산된 다음 인스턴스 수명 동안 일반 어트리뷰트로 캐시 되는 프로퍼티로 변환합니다. property()와 유사하고, 캐싱이 추가되었습니다. 비싸게 계산되고 그 외에는 사실상 불변인 인스턴스의 프로퍼티에 유용합니다.

예:

class DataSet:

    def __init__(self, sequence_of_numbers):
        self._data = tuple(sequence_of_numbers)

    @cached_property
    def stdev(self):
        return statistics.stdev(self._data)

cached_property()의 메커니즘은 property()와 다소 다릅니다. 일반 프로퍼티는 setter가 정의되지 않은 경우 어트리뷰트 쓰기를 차단합니다. 이와는 달리, cached_property는 쓰기를 허용합니다.

cached_property 데코레이터는 조회 시에만, 같은 이름의 어트리뷰트가 존재하지 않을 때만 실행됩니다. 실행되면, cached_property는 같은 이름의 어트리뷰트에 기록합니다. 후속 어트리뷰트 읽기와 쓰기는 cached_property 메서드보다 우선하며 일반 어트리뷰트처럼 작동합니다.

캐시 된 값은 어트리뷰트를 삭제하여 지울 수 있습니다. 이렇게 하면 cached_property 메서드가 다시 실행됩니다.

이 데코레이터는 PEP 412 키 공유 딕셔너리의 작동을 방해함에 유의하십시오. 이는 인스턴스 딕셔너리가 평소보다 더 많은 공간을 차지할 수 있음을 의미합니다.

또한, 이 데코레이터는 각 인스턴스의 __dict__ 어트리뷰트가 가변 매핑일 것을 요구합니다. 이는 메타 클래스(형 인스턴스의 __dict__ 어트리뷰트가 클래스 이름 공간에 대한 읽기 전용 프락시이기 때문에)와 __dict__를 정의된 슬롯 중 하나로 포함하지 않고 __slots__를 지정하는 것(이러한 클래스는 __dict__ 어트리뷰트를 전혀 제공하지 않기 때문에)과 같은 일부 형에서 작동하지 않음을 의미합니다.

가변 매핑을 사용할 수 없거나 공간 효율적인 키 공유가 필요하면, cache() 위에 property()를 쌓아서 cached_property()와 유사한 효과를 얻을 수 있습니다:

class DataSet:
    def __init__(self, sequence_of_numbers):
        self._data = sequence_of_numbers

    @property
    @cache
    def stdev(self):
        return statistics.stdev(self._data)

버전 3.8에 추가.

functools.cmp_to_key(func)

구식 비교 함수를 키 함수로 변환합니다. (sorted(), min(), max(), heapq.nlargest(), heapq.nsmallest(), itertools.groupby()와 같은) 키 함수를 받아들이는 도구와 함께 사용됩니다. 이 함수는 주로 비교 함수 사용을 지원하는 파이썬 2에서 변환되는 프로그램의 전이 도구로 사용됩니다.

비교 함수는 두 개의 인자를 받아들이고, 그들을 비교하여, 작으면 음수, 같으면 0, 크면 양수를 반환하는 콜러블입니다. 키 함수는 하나의 인자를 받아들이고 정렬 키로 사용할 다른 값을 반환하는 콜러블입니다.

예:

sorted(iterable, key=cmp_to_key(locale.strcoll))  # locale-aware sort order

정렬 예제와 간략한 정렬 자습서는 정렬 HOW TO를 참조하십시오.

버전 3.2에 추가.

@functools.lru_cache(user_function)
@functools.lru_cache(maxsize=128, typed=False)

가장 최근의 maxsize 호출까지 저장하는 기억하는(memoizing) 콜러블 함수를 감싸는 데코레이터. 비싸거나 I/O 병목 함수가 같은 인자로 주기적으로 호출될 때 시간을 절약할 수 있습니다.

결과를 캐시 하는 데 딕셔너리가 사용되기 때문에, 함수에 대한 위치와 키워드 인자는 해시 가능해야 합니다.

서도 다른 인자 패턴은 별도의 캐시 항목을 갖는 별개의 호출로 간주할 수 있습니다. 예를 들어, f(a=1, b=2)f(b=2, a=1)은 키워드 인자 순서가 다르며 두 개의 개별 캐시 항목을 가질 수 있습니다.

user_function이 지정되면, 콜러블이어야 합니다. 이는 lru_cache 데코레이터를 사용자 함수에 직접 적용 할 수 있도록 하며, maxsize를 기본값 128로 유지합니다:

@lru_cache
def count_vowels(sentence):
    sentence = sentence.casefold()
    return sum(sentence.count(vowel) for vowel in 'aeiou')

maxsizeNone으로 설정되면, LRU 기능이 비활성화되고 캐시가 제한 없이 커질 수 있습니다.

typed가 참으로 설정되면, 다른 형의 함수 인자가 별도로 캐시 됩니다. 예를 들어, f(3)f(3.0)은 별개의 결과를 가진 별개의 호출로 취급됩니다.

래핑 된 함수는 maxsizetyped의 값을 표시하는 새 dict를 반환하는 cache_parameters() 함수로 인스트루먼트 됩니다. 이것은 정보 제공만을 위한 것입니다. 값을 변경해도 효과가 없습니다.

캐시의 효과를 측정하고 maxsize 매개 변수를 조정하는 것을 돕기 위해, 래핑 된 함수는 hits, misses, maxsizecurrsize를 표시하는 네임드 튜플을 반환하는 cache_info() 함수로 인스트루먼트 됩니다. 다중 스레드 환경에서, hits와 misses는 근사적(approximate)입니다.

데코레이터는 캐시를 지우거나 무효로 하기 위한 cache_clear() 함수도 제공합니다.

원래의 하부 함수는 __wrapped__ 어트리뷰트를 통해 액세스 할 수 있습니다. 이것은 인트로스펙션, 캐시 우회 또는 다른 캐시로 함수를 다시 래핑하는 데 유용합니다.

LRU (least recently used) 캐시는 가장 최근 호출이 향후 호출에 대한 최상의 예측일 때 가장 잘 작동합니다 (예를 들어, 뉴스 서버에서 가장 인기 있는 기사는 매일 바뀌는 경향이 있습니다). 캐시의 크기 제한은 웹 서버와 같은 오래 실행되는 프로세스에서 제한 없이 캐시가 커지지 않도록 합니다.

일반적으로, LRU 캐시는 이전에 계산된 값을 재사용하려고 할 때만 사용해야 합니다. 따라서, 부작용이 있는 함수, 각 호출에서 고유한 가변 객체를 만들어야 하는 함수, time()이나 random()과 같은 비순수(impure) 함수를 캐시 하는 것은 의미가 없습니다.

정적 웹 콘텐츠를 위한 LRU 캐시의 예:

@lru_cache(maxsize=32)
def get_pep(num):
    'Retrieve text of a Python Enhancement Proposal'
    resource = 'https://www.python.org/dev/peps/pep-%04d/' % num
    try:
        with urllib.request.urlopen(resource) as s:
            return s.read()
    except urllib.error.HTTPError:
        return 'Not Found'

>>> for n in 8, 290, 308, 320, 8, 218, 320, 279, 289, 320, 9991:
...     pep = get_pep(n)
...     print(n, len(pep))

>>> get_pep.cache_info()
CacheInfo(hits=3, misses=8, maxsize=32, currsize=8)

동적 프로그래밍(dynamic programming) 기법을 구현하기 위해 캐시를 사용하여 피보나치 수를 효율적으로 계산하는 예:

@lru_cache(maxsize=None)
def fib(n):
    if n < 2:
        return n
    return fib(n-1) + fib(n-2)

>>> [fib(n) for n in range(16)]
[0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610]

>>> fib.cache_info()
CacheInfo(hits=28, misses=16, maxsize=None, currsize=16)

버전 3.2에 추가.

버전 3.3에서 변경: typed 옵션을 추가했습니다.

버전 3.8에서 변경: user_function 옵션을 추가했습니다.

버전 3.9에 추가: cache_parameters() 함수를 추가했습니다

@functools.total_ordering

하나 이상의 풍부한 비교(rich comparison) 순서 메서드를 정의하는 클래스를 주면, 이 클래스 데코레이터가 나머지를 제공합니다. 가능한 모든 풍부한 비교 연산을 지정하는 데 드는 노력이 단순화됩니다:

클래스는 __lt__(), __le__(), __gt__() 또는 __ge__() 중 하나를 정의해야 합니다. 또한, 클래스는 __eq__() 메서드를 제공해야 합니다.

예를 들면:

@total_ordering
class Student:
    def _is_valid_operand(self, other):
        return (hasattr(other, "lastname") and
                hasattr(other, "firstname"))
    def __eq__(self, other):
        if not self._is_valid_operand(other):
            return NotImplemented
        return ((self.lastname.lower(), self.firstname.lower()) ==
                (other.lastname.lower(), other.firstname.lower()))
    def __lt__(self, other):
        if not self._is_valid_operand(other):
            return NotImplemented
        return ((self.lastname.lower(), self.firstname.lower()) <
                (other.lastname.lower(), other.firstname.lower()))

참고

이 데코레이터를 사용하면 올바르게 동작하는 전 순서(totally ordered) 형을 쉽게 만들 수 있지만, 파생된 비교 메서드에서 실행 속도가 느려지고 스택 트레이스가 더 복잡해지는 대가를 지불합니다. 성능 벤치마킹이 이것이 특정 응용 프로그램의 병목임을 가리키면, 6가지의 풍부한 비교 메서드를 모두 구현하여 속도를 쉽게 높일 수 있습니다.

버전 3.2에 추가.

버전 3.4에서 변경: 인식할 수 없는 형에 대해 하부 비교 함수에서 NotImplemented를 반환하는 것이 이제 지원됩니다.

functools.partial(func, /, *args, **keywords)

호출될 때 위치 인자 args와 키워드 인자 keywords로 호출된 func처럼 동작하는 새 partial 객체를 반환합니다. 더 많은 인자가 호출에 제공되면, args에 추가됩니다. 추가 키워드 인자가 제공되면, keywords를 확장하고 대체합니다. 대략 다음과 동등합니다:

def partial(func, /, *args, **keywords):
    def newfunc(*fargs, **fkeywords):
        newkeywords = {**keywords, **fkeywords}
        return func(*args, *fargs, **newkeywords)
    newfunc.func = func
    newfunc.args = args
    newfunc.keywords = keywords
    return newfunc

partial()은 함수의 인자 및/또는 키워드의 일부를 “고정”하여 서명이 단순화된 새 객체를 생성하는 부분 함수 응용에 사용됩니다. 예를 들어, partial()을 사용하여 base 인자의 기본값이 2이면서 int() 함수 같은 동작을 하는 콜러블을 만들 수 있습니다:

>>> from functools import partial
>>> basetwo = partial(int, base=2)
>>> basetwo.__doc__ = 'Convert base 2 string to an int.'
>>> basetwo('10010')
18
class functools.partialmethod(func, /, *args, **keywords)

직접 호출하기보다는 메서드 정의로 사용되도록 설계된 것을 제외하고는 partial과 같이 동작하는 새 partialmethod 디스크립터를 반환합니다.

func디스크립터나 콜러블이어야 합니다 (일반 함수처럼 둘 모두인 객체는 디스크립터로 처리됩니다).

func가 디스크립터(가령 일반 파이썬 함수, classmethod(), staticmethod(), abstractmethod() 또는 partialmethod의 다른 인스턴스)이면, __get__에 대한 호출은 하부 디스크립터에 위임되고, 적절한 partial 객체가 결과로 반환됩니다.

func가 디스크립터가 아닌 콜러블이면, 적절한 연결된 메서드가 동적으로 만들어집니다. 이것은 메서드로 사용될 때 일반 파이썬 함수처럼 작동합니다: partialmethod 생성자에 제공된 argskeywords보다도 전에 self 인자가 첫 번째 위치 인자로 삽입됩니다.

예:

>>> class Cell:
...     def __init__(self):
...         self._alive = False
...     @property
...     def alive(self):
...         return self._alive
...     def set_state(self, state):
...         self._alive = bool(state)
...     set_alive = partialmethod(set_state, True)
...     set_dead = partialmethod(set_state, False)
...
>>> c = Cell()
>>> c.alive
False
>>> c.set_alive()
>>> c.alive
True

버전 3.4에 추가.

functools.reduce(function, iterable[, initializer])

두 인자의 function을 왼쪽에서 오른쪽으로 iterable의 항목에 누적적으로 적용해서, 이터러블을 단일 값으로 줄입니다. 예를 들어, reduce(lambda x, y: x+y, [1, 2, 3, 4, 5])((((1+2)+3)+4)+5)를 계산합니다. 왼쪽 인자 x는 누적값이고 오른쪽 인자 yiterable에서 온 갱신 값입니다. 선택적 initializer가 있으면, 계산에서 이터러블의 항목 앞에 배치되고, 이터러블이 비어있을 때 기본값의 역할을 합니다. initializer가 제공되지 않고 iterable에 하나의 항목만 포함되면, 첫 번째 항목이 반환됩니다.

대략 다음과 동등합니다:

def reduce(function, iterable, initializer=None):
    it = iter(iterable)
    if initializer is None:
        value = next(it)
    else:
        value = initializer
    for element in it:
        value = function(value, element)
    return value

모든 중간값을 산출하는 이터레이터는 itertools.accumulate()를 참조하십시오.

@functools.singledispatch

함수를 싱글 디스패치 제네릭 함수로 변환합니다.

To define a generic function, decorate it with the @singledispatch decorator. When defining a function using @singledispatch, note that the dispatch happens on the type of the first argument:

>>> from functools import singledispatch
>>> @singledispatch
... def fun(arg, verbose=False):
...     if verbose:
...         print("Let me just say,", end=" ")
...     print(arg)

To add overloaded implementations to the function, use the register() attribute of the generic function, which can be used as a decorator. For functions annotated with types, the decorator will infer the type of the first argument automatically:

>>> @fun.register
... def _(arg: int, verbose=False):
...     if verbose:
...         print("Strength in numbers, eh?", end=" ")
...     print(arg)
...
>>> @fun.register
... def _(arg: list, verbose=False):
...     if verbose:
...         print("Enumerate this:")
...     for i, elem in enumerate(arg):
...         print(i, elem)

형 어노테이션을 사용하지 않는 코드의 경우, 적절한 형 인자를 데코레이터 자체에 명시적으로 전달할 수 있습니다:

>>> @fun.register(complex)
... def _(arg, verbose=False):
...     if verbose:
...         print("Better than complicated.", end=" ")
...     print(arg.real, arg.imag)
...

To enable registering lambdas and pre-existing functions, the register() attribute can also be used in a functional form:

>>> def nothing(arg, verbose=False):
...     print("Nothing.")
...
>>> fun.register(type(None), nothing)

The register() attribute returns the undecorated function. This enables decorator stacking, pickling, and the creation of unit tests for each variant independently:

>>> @fun.register(float)
... @fun.register(Decimal)
... def fun_num(arg, verbose=False):
...     if verbose:
...         print("Half of your number:", end=" ")
...     print(arg / 2)
...
>>> fun_num is fun
False

호출되면, 제네릭 함수는 첫 번째 인자의 형에 따라 디스패치 합니다:

>>> fun("Hello, world.")
Hello, world.
>>> fun("test.", verbose=True)
Let me just say, test.
>>> fun(42, verbose=True)
Strength in numbers, eh? 42
>>> fun(['spam', 'spam', 'eggs', 'spam'], verbose=True)
Enumerate this:
0 spam
1 spam
2 eggs
3 spam
>>> fun(None)
Nothing.
>>> fun(1.23)
0.615

Where there is no registered implementation for a specific type, its method resolution order is used to find a more generic implementation. The original function decorated with @singledispatch is registered for the base object type, which means it is used if no better implementation is found.

If an implementation is registered to an abstract base class, virtual subclasses of the base class will be dispatched to that implementation:

>>> from collections.abc import Mapping
>>> @fun.register
... def _(arg: Mapping, verbose=False):
...     if verbose:
...         print("Keys & Values")
...     for key, value in arg.items():
...         print(key, "=>", value)
...
>>> fun({"a": "b"})
a => b

To check which implementation the generic function will choose for a given type, use the dispatch() attribute:

>>> fun.dispatch(float)
<function fun_num at 0x1035a2840>
>>> fun.dispatch(dict)    # note: default implementation
<function fun at 0x103fe0000>

등록된 모든 구현에 액세스하려면, 읽기 전용 registry 어트리뷰트를 사용하십시오:

>>> fun.registry.keys()
dict_keys([<class 'NoneType'>, <class 'int'>, <class 'object'>,
          <class 'decimal.Decimal'>, <class 'list'>,
          <class 'float'>])
>>> fun.registry[float]
<function fun_num at 0x1035a2840>
>>> fun.registry[object]
<function fun at 0x103fe0000>

버전 3.4에 추가.

버전 3.7에서 변경: The register() attribute now supports using type annotations.

class functools.singledispatchmethod(func)

메서드를 싱글 디스패치 제네릭 함수로 변환합니다.

To define a generic method, decorate it with the @singledispatchmethod decorator. When defining a function using @singledispatchmethod, note that the dispatch happens on the type of the first non-self or non-cls argument:

class Negator:
    @singledispatchmethod
    def neg(self, arg):
        raise NotImplementedError("Cannot negate a")

    @neg.register
    def _(self, arg: int):
        return -arg

    @neg.register
    def _(self, arg: bool):
        return not arg

@singledispatchmethod supports nesting with other decorators such as @classmethod. Note that to allow for dispatcher.register, singledispatchmethod must be the outer most decorator. Here is the Negator class with the neg methods bound to the class, rather than an instance of the class:

class Negator:
    @singledispatchmethod
    @classmethod
    def neg(cls, arg):
        raise NotImplementedError("Cannot negate a")

    @neg.register
    @classmethod
    def _(cls, arg: int):
        return -arg

    @neg.register
    @classmethod
    def _(cls, arg: bool):
        return not arg

The same pattern can be used for other similar decorators: @staticmethod, @abstractmethod, and others.

버전 3.8에 추가.

functools.update_wrapper(wrapper, wrapped, assigned=WRAPPER_ASSIGNMENTS, updated=WRAPPER_UPDATES)

wrapped 함수처럼 보이도록 wrapper 함수를 갱신합니다. 선택적 인자는 원래 함수의 어떤 어트리뷰트가 wrapper 함수의 일치하는 어트리뷰트에 직접 대입되고 wrapper 함수의 어떤 어트리뷰트가 원래 함수의 해당 어트리뷰트로 갱신되는지 지정하는 튜플입니다. 이 인자들의 기본값은 모듈 수준 상수 WRAPPER_ASSIGNMENTS(wrapper 함수의 __module__, __name__, __qualname__, __annotations____doc__ 독스트링에 대입합니다)와 WRAPPER_UPDATES(wrapper 함수의 __dict__, 즉 인스턴스 딕셔너리를 갱신합니다)입니다.

내부 검사와 기타 목적(예를 들어 lru_cache()와 같은 캐싱 데코레이터 우회)을 위해 원래 함수에 액세스 할 수 있도록, 이 함수는 래핑 되는 함수를 가리키는 __wrapped__ 어트리뷰트를 wrapper에 자동으로 추가합니다.

이 함수의 주요 용도는 데코레이트 된 함수를 래핑하고 wrapper를 반환하는 데코레이터 함수에서 사용하는 것입니다. wrapper 함수가 갱신되지 않으면, 반환된 함수의 메타 데이터는 원래 함수 정의가 아닌 wrapper 정의를 반영하게 되어 일반적으로 도움이 되지 않습니다.

update_wrapper()는 함수 이외의 콜러블과 함께 사용할 수 있습니다. 래핑 되는 객체에서 누락된 assignedupdated로 이름 지정된 어트리뷰트는 무시됩니다 (즉, 이 함수는 wrapper 함수에서 그 어트리뷰트를 설정하려고 시도하지 않습니다). wrapper 함수 자체에 updated에 이름 지정된 어트리뷰트가 없으면 여전히 AttributeError가 발생합니다.

버전 3.2에 추가: __wrapped__ 어트리뷰트 자동 추가.

버전 3.2에 추가: 기본적으로 __annotations__ 어트리뷰트의 복사.

버전 3.2에서 변경: 누락된 어트리뷰트는 더는 AttributeError를 발생시키지 않습니다.

버전 3.4에서 변경: __wrapped__ 어트리뷰트는 이제 해당 함수가 __wrapped__ 어트리뷰트를 정의한 경우에도 항상 래핑 된 함수를 참조합니다. (bpo-17482를 참조하십시오)

@functools.wraps(wrapped, assigned=WRAPPER_ASSIGNMENTS, updated=WRAPPER_UPDATES)

래퍼 함수를 정의할 때 함수 데코레이터로 update_wrapper()를 호출하기 위한 편의 함수입니다. partial(update_wrapper, wrapped=wrapped, assigned=assigned, updated=updated)와 동등합니다. 예를 들면:

>>> from functools import wraps
>>> def my_decorator(f):
...     @wraps(f)
...     def wrapper(*args, **kwds):
...         print('Calling decorated function')
...         return f(*args, **kwds)
...     return wrapper
...
>>> @my_decorator
... def example():
...     """Docstring"""
...     print('Called example function')
...
>>> example()
Calling decorated function
Called example function
>>> example.__name__
'example'
>>> example.__doc__
'Docstring'

이 데코레이터 팩토리를 사용하지 않으면, example 함수의 이름은 'wrapper'가 되고, 원래 example()의 독스트링은 잃어버리게 됩니다.

partial 객체

partial 객체는 partial()이 만든 콜러블 객체입니다. 세 가지 읽기 전용 어트리뷰트가 있습니다:

partial.func

콜러블 객체나 함수. partial 객체에 대한 호출은 새로운 인자와 키워드와 함께 func로 전달됩니다.

partial.args

partial 객체 호출에 제공되는 위치 인자 앞에 추가될 가장 왼쪽 위치 인자들입니다.

partial.keywords

partial 객체가 호출될 때 제공될 키워드 인자들입니다.

partial 객체는 콜러블이고, 약한 참조 가능하며, 어트리뷰트를 가질 수 있다는 점에서 function 객체와 같습니다. 몇 가지 중요한 차이점이 있습니다. 예를 들어, __name____doc__ 어트리뷰트는 자동으로 만들어지지 않습니다. 또한, 클래스에 정의된 partial 객체는 정적 메서드처럼 동작하며 인스턴스 어트리뷰트 조회 중에 연결된 메서드로 변환되지 않습니다.