使类可转换为ndarray

问题描述

除了通过子类化(例如，从list继承)之外，如何使python对象隐式转换为ndarray?

Other than by subclassing (from list for example) how do I make a python object implicitly convertable to ndarray?

示例:

import numpy
arg=[0,1,2]
numpy.dot(arg,arg) # OK, arg is converted to ndarray

#Looks somewhat array like, albeit without support for slices
class A(object):
    def __init__(self, x=0,y=1,z=2):
        (self.x,self.y,self.z)=(x,y,z)
    def __getitem__(self, idx):
        if idx==0:
            return self.x
        elif idx==1:
            return self.y
        elif idx==2:
            return self.z
        else:
            raise IndexError()
    def __setitem__(self, idx, val):
        if idx==0:
            self.x=val
        elif idx==1:
            self.y=val
        elif idx==2:
            self.z=val
        else:
            raise IndexError()
    def __iter__(self):
        for v in (self.x,self.y,self.z):
            yield v
     # Is there a set of functions I can add here to allow
     # numpy to convert instances of A into ndarrays?
arg=A()
numpy.dot(arg,arg) # does not work

错误:

>>> scipy.dot(a,a) # I use scipy by default
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/home/dave/tmp/<ipython-input-9-f73d996ba2b6> in <module>()
----> 1 scipy.dot(a,a)

TypeError: unsupported operand type(s) for *: 'A' and 'A'

它正在调用array(arg)，但是会产生一个类似于[arg,]的数组，它是shape==()，所以dot尝试将A实例相乘在一起.

It's calling array(arg) but that yields an array like [arg,], it's shape==() so dot tries to multiply the A instances together.

可以(实际上是预期的)转换为ndarray时需要复制数据.

It's ok (in fact, expected) that the conversion to ndarray will require copying the data.

推荐答案

__len__似乎是关键功能:只需添加

__len__ seems to be the key feature: just adding a

def __len__(self):
    return 3

使类在numpy.dot中工作.

这篇关于使类可转换为ndarray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！