返回数组的函数的numpy.vectorize [英] numpy.vectorize of function that returns an array

问题描述

我试图向量化一个具有2个输入并输出形状=(4，)的np.array的函数.该函数如下所示:

I am trying to vectorize a function which has 2 inputs, and outputs a np.array, of shape =(4,). The function looks like this:

def f(a, b):
    return np.array([a+b, a-b, a, b])

我能够使用签名参数对函数进行矢量化处理，但是仅当我使用np.vectorize的excluded自变量排除其中一个参数时，该函数才有效:

I was able to vectorize the function, using the signature parameter, however it only works if I exclude one of the parameters using the excluded argument of np.vectorize:

这有效:

vec = np.vectorize(f, signature='()->(n)', excluded=[1])
x = np.arange(5)
y = 3
vec(x, y)

>> output:
array([[ 3, -3,  0,  3],
       [ 4, -2,  1,  3],
       [ 5, -1,  2,  3],
       [ 6,  0,  3,  3],
       [ 7,  1,  4,  3]])

但是，如果我删除了excluded参数，事情就不会按计划进行.

However if I take out the excluded argument, things don't go as planned.

这不起作用:

vec = np.vectorize(f, signature='()->(n)')
x = np.arange(5)
y = 3
vec(x, y)

>> Error:
TypeError: wrong number of positional arguments: expected 1, got 2

如何使矢量化函数能够接收一个(或两个)输入参数的输入值的数组/列表?

How can I make the vectorized function able to receive an array/list of input values for either one (or both) of the input parameters?

预期的输出将是vec函数，该函数将允许使用多个输入来调用任一输入参数.

The expected output would be a vec function which would enable calling it with multiple inputs for either one of the input parameters.

推荐答案

In [237]: f1 = np.vectorize(f, signature='(),()->(n)')                          
In [238]: f1(np.arange(5),3)                                                    
Out[238]: 
array([[ 3, -3,  0,  3],
       [ 4, -2,  1,  3],
       [ 5, -1,  2,  3],
       [ 6,  0,  3,  3],
       [ 7,  1,  4,  3]])

In [241]: f1(np.arange(5),np.ones((4,5))).shape                                 
Out[241]: (4, 5, 4)
In [242]: f1(np.arange(5),np.ones((1,5))).shape                                 
Out[242]: (1, 5, 4)

---

frompyfunc返回一个对象dtype数组:

---

frompyfunc returns a object dtype array:

In [336]: f2 = np.frompyfunc(f,2,1)                                             
In [337]: f2(np.arange(5), 3)                                                   
Out[337]: 
array([array([ 3, -3,  0,  3]), array([ 4, -2,  1,  3]),
       array([ 5, -1,  2,  3]), array([6, 0, 3, 3]), array([7, 1, 4, 3])],
      dtype=object)
In [338]: _.shape                                                               
Out[338]: (5,)

np.vectorize(没有signature)使用frompyfunc，但添加了自己的dtype转换.

np.vectorize, without the signature, uses frompyfunc, but adds its own dtype conversion.

In [340]: f1(np.arange(5), np.arange(3)) 
ValueError: shape mismatch: objects cannot be broadcast to a single shape

此操作失败的原因与以下添加操作失败的原因相同

This fails for the same reason the following addition fails:

In [341]: np.arange(5)+np.arange(3)                                             
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-341-fb1c4f4372da> in <module>
----> 1 np.arange(5)+np.arange(3)

ValueError: operands could not be broadcast together with shapes (5,) (3,) 

要获得(5,3)结果，我们需要使第一个参数(5,1)成为形状:

To get a (5,3) result we need to make the first argument (5,1) shape:

In [342]: np.arange(5)[:,None]+np.arange(3)                                     
Out[342]: 
array([[0, 1, 2],
       [1, 2, 3],
       [2, 3, 4],
       [3, 4, 5],
       [4, 5, 6]])

这篇关于返回数组的函数的numpy.vectorize的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！