我发现numpy.vectorize允许转换'普通'函数，这些函数期望单个数字作为函数的输入，该函数也可以将输入列表转换为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，我无法让这个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]
此测试失败：
=================================== 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numpy.vectorize应用于实例方法？
您可以直接在实例上的方法上使用np.vectorize：
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vec_f中创建矢量化函数__init__：
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]
或者使用不同的命名方案：
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()
记住我在memoized装饰器中看到的一种技术，我设法通过子类化numpy.vectorize使装饰器也用于实例方法，如下所示：
import numpy as np
import functools
class vectorize(np.vectorize):
def __get__(self, obj, objtype):
return functools.partial(self.__call__, obj)
现在如果我用Dummy而不是f装饰vectorize类'np.vectorize方法，测试通过：
class Dummy(object):
def __init__(self, val=1):
self.val = val
@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]
if __name__ == "__main__":
pytest.main([__file__])
与输出
test_numpy_vectorize.py .
=========================== 1 passed in 0.01 seconds ===========================
[Finished in 0.7s]
这是一个使用实例方法和函数的通用装饰器（请参阅Numpy's documentation for otypes和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
您可以使用它来按如下方式对方法进行矢量化：
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]
关键是要使用signature kwarg。 ->左侧的带括号的值指定输入参数，右侧的值指定输出值。 ()代表一个标量（0维向量）; (n)代表一维向量; (m,n)代表一个二维向量; (m,n,p)代表一个三维向量;在这里，signature="(),()->()"向Numpy指定第一个参数（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（self.val或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 ======================
如果要使用方法的矢量化实现，可以使用excluded参数，如下所示：
class MyClass:
def __init__(self, data):
self.data = data
self.my_vectorized_func = np.vectorize(self.my_func, excluded='self')
def my_func(self, x):
return pow(x, self.data)
有了这个，你可以使用你的方法，如非矢量化的方法：
In[1]: myclass = MyClass(3) # '3' will be the power factor of our function
In[2]: myclass.my_vectorized_func([1, 2, 3, 4, 5])
Out[3]: array([ 1, 8, 27, 64, 125])
来自docs：
向量化输出的数据类型是通过使用输入的第一个元素调用函数来确定的。通过指定otypes参数可以避免这种情况。
函数f(self, x)中的第一个输入是self。也许你可以使这个函数成为staticmethod函数的包装器？