是否可以numpy.vectorize一个实例方法？

Question

我发现，numpy.vectorize允许将希望将单个数字作为输入的“普通”函数转换为函数，该函数也可以将输入列表转换为该函数已映射到每个输入的列表。例如，以下测试通过：

import numpy as np
import pytest


@np.vectorize
def f(x):
    if x == 0:
        return 1
    else:
        return 2


def test_1():
    assert list(f([0, 1, 2])) == [1, 2, 2]

def test_2():
    assert f(0) == 1

if __name__ == "__main__":
    pytest.main([__file__])

但是，我无法使它用于使用实例属性的实例方法。例如：

class Dummy(object):
    def __init__(self, val=1):
        self.val = val

    @np.vectorize
    def f(self, x):
        if x == 0:
            return self.val
        else:
            return 2


def test_3():
    assert list(Dummy().f([0, 1, 2])) == [1, 2, 2]

该测试失败：

=================================== FAILURES ===================================
____________________________________ test_3 ____________________________________

    def test_3():
>       assert list(Dummy().f([0, 1, 2])) == [1, 2, 2]

test_numpy_vectorize.py:31: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/numpy/lib/function_base.py:2739: in __call__
    return self._vectorize_call(func=func, args=vargs)
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/numpy/lib/function_base.py:2809: in _vectorize_call
    ufunc, otypes = self._get_ufunc_and_otypes(func=func, args=args)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <numpy.lib.function_base.vectorize object at 0x106546470>
func = <function Dummy.f at 0x10653a2f0>, args = [array([0, 1, 2])]

    def _get_ufunc_and_otypes(self, func, args):
        """Return (ufunc, otypes)."""
        # frompyfunc will fail if args is empty
        if not args:
            raise ValueError('args can not be empty')

        if self.otypes is not None:
            otypes = self.otypes
            nout = len(otypes)

            # Note logic here: We only *use* self._ufunc if func is self.pyfunc
            # even though we set self._ufunc regardless.
            if func is self.pyfunc and self._ufunc is not None:
                ufunc = self._ufunc
            else:
                ufunc = self._ufunc = frompyfunc(func, len(args), nout)
        else:
            # Get number of outputs and output types by calling the function on
            # the first entries of args.  We also cache the result to prevent
            # the subsequent call when the ufunc is evaluated.
            # Assumes that ufunc first evaluates the 0th elements in the input
            # arrays (the input values are not checked to ensure this)
            args = [asarray(arg) for arg in args]
            if builtins.any(arg.size == 0 for arg in args):
                raise ValueError('cannot call `vectorize` on size 0 inputs '
                                 'unless `otypes` is set')

            inputs = [arg.flat[0] for arg in args]
>           outputs = func(*inputs)
E           TypeError: f() missing 1 required positional argument: 'x'

是否可以将其应用于numpy.vectorize实例方法？

Answer 1

简单的解决方案，无需修改类

您可以np.vectorize直接在实例上的方法上使用：

class Dummy(object):

    def __init__(self, val=1):
        self.val = val

    def f(self, x):
        if x == 0:
            return self.val
        else:
            return 2


vec_f = np.vectorize(Dummy().f) 


def test_3():
    assert list(vec_f([0, 1, 2])) == [1, 2, 2]

test_3()

您还可以在以下代码vec_f中创建矢量化函数__init__：

向实例添加矢量化版本

class Dummy(object):

    def __init__(self, val=1):
        self.val = val
        self.vec_f = np.vectorize(self.f) 

    def f(self, x):
        if x == 0:
            return self.val
        else:
            return 2


def test_3():
    assert list(Dummy().vec_f([0, 1, 2])) == [1, 2, 2]

或使用其他命名方案：

class Dummy(object):

    def __init__(self, val=1):
        self.val = val
        self.f = np.vectorize(self.scalar_f) 

    def scalar_f(self, x):
        if x == 0:
            return self.val
        else:
            return 2


def test_3():
    assert list(Dummy().f([0, 1, 2])) == [1, 2, 2]

test_3()

    test_3()

Answer 2

这是一个与实例方法和函数一起使用的通用装饰器（请参阅Numpy的otypes和文档signature）：

from functools import wraps

import numpy as np

def vectorize(otypes=None, signature=None):
    """Numpy vectorization wrapper that works with instance methods."""
    def decorator(fn):
        vectorized = np.vectorize(fn, otypes=otypes, signature=signature)
        @wraps(fn)
        def wrapper(*args):
            return vectorized(*args)
        return wrapper
    return decorator

您可以使用它来矢量化您的方法，如下所示：

class Dummy(object):
    def __init__(self, val=1):
        self.val = val

    @vectorize(signature="(),()->()")
    def f(self, x):
        if x == 0:
            return self.val
        else:
            return 2


def test_3():
    assert list(Dummy().f([0, 1, 2])) == [1, 2, 2]

关键是要利用 kwarg signature。左侧的括号值->指定输入参数，右侧的值指定输出值。 ()表示标量（0 维向量）；(n)表示一维向量；(m,n)表示一个二维向量；(m,n,p)表示 3 维向量；此处，signature="(),()->()"指定 Numpy 第一个参数 ( self) 是标量，第二个参数 ( x) 也是标量，并且该方法返回一个标量（或者self.val，2取决于x）。

$ pytest /tmp/instance_vectorize.py
======================= test session starts ========================
platform linux -- Python 3.6.5, pytest-3.5.1, py-1.5.3, pluggy-0.6.0
rootdir: /tmp, inifile:
collected 1 item

../../tmp/instance_vectorize.py .                                                                                                                                                     [100%]

==================== 1 passed in 0.08 seconds ======================