如何获取两个不同的numpy.ndarray子类的__matmul__以返回特定的子类？

Question

我有两个np.ndarray子类。 Tuple @ Matrix返回Tuple，但是Matrix @ Tuple返回Matrix。我如何让它返回Tuple？

import numpy as np

class Tuple(np.ndarray):
    def __new__(cls, input_array, info=None):
        return np.asarray(input_array).view(cls)

class Matrix(np.ndarray):
    def __new__(cls, input_array, info=None):
        return np.asarray(input_array).view(cls)

def scaling(x, y, z):
    m = Matrix(np.identity(4))
    m[0, 0] = x
    m[1, 1] = y
    m[2, 2] = z
    return m

示例：

>>> Tuple([1,2,3,4]) @ scaling(2,2,2)
Tuple([2., 4., 6., 4.])

>>> scaling(2,2,2) @ Tuple([1,2,3,4])
Matrix([2., 4., 6., 4.])   # XXXX I'd like this to be a Tuple

PS：Matrix @ Matrix应该返回Matrix

Answer 1

您可以重载__matmul__方法以返回Tuple-并且如果您想成为Tuple（如果任何变量是Tuple和{{ 1}}否则，我认为这会起作用：

Matrix

Answer 2

我在复制np.matrix示例中犯了一个错误。

class Tuple(np.ndarray): 
    __array_priority__ = 10 
    def __new__(cls, input_array, info=None): 
        return np.asarray(input_array).view(cls) 
class Matrix(np.ndarray):
    __array_priority__ = 5.0 
    def __new__(cls, input_array, info=None): 
        return np.asarray(input_array).view(cls)

In [2]: def scaling(x, y, z):  
   ...:      ...:     m = Matrix(np.identity(4))  
   ...:      ...:     m[0, 0] = x  
   ...:      ...:     m[1, 1] = y  
   ...:      ...:     m[2, 2] = z  
   ...:      ...:     return m  
   ...:                                                                                                                                  
In [3]: Tuple([1,2,3,4]) @ scaling(2,2,2)                                                                                                
Out[3]: Tuple([2., 4., 6., 4.])
In [4]: scaling(2,2,2) @ Tuple([1,2,3,4])                                                                                                
Out[4]: Tuple([2., 4., 6., 4.])

===

从np.matrix定义中获取线索：numpy.matrixlib.defmatrix.py

添加一个__array_priority__属性：

In [382]: class Tuple(np.ndarray): 
     ...:     def __new__(cls, input_array, info=None): 
     ...:         __array_priority = 10 
     ...:         return np.asarray(input_array).view(cls) 
     ...:  
     ...: class Matrix(np.ndarray): 
     ...:     def __new__(cls, input_array, info=None): 
     ...:         __array_priority = 5 
     ...:         return np.asarray(input_array).view(cls) 
     ...:                                                                                            
In [383]:                                                                                            
In [383]: def scaling(x, y, z): 
     ...:     m = Matrix(np.identity(4)) 
     ...:     m[0, 0] = x 
     ...:     m[1, 1] = y 
     ...:     m[2, 2] = z 
     ...:     return m 
     ...:                                                                                            
In [384]: Tuple([1,2,3,4]) @ scaling(2,2,2)                                                          
Out[384]: Tuple([2., 4., 6., 4.])
In [385]: scaling(2,2,2) @ Tuple([1,2,3,4])                                                          
Out[385]: Matrix([2., 4., 6., 4.])

Answer 3

解决此问题的一种方法是在__matmul__中实现自定义Matrix，在__rmatmul__中实现Tuple：

import numpy as np

class Tuple(np.ndarray):
    def __new__(cls, input_array, info=None):
        return np.asarray(input_array).view(cls)

    def __rmatmul__(self, other):
        return super().__matmul__(other)

class Matrix(np.ndarray):
    def __new__(cls, input_array, info=None):
        return np.asarray(input_array).view(cls)

    def __matmul__(self, other):
        if not isinstance(other, Matrix):
            return NotImplemented
        return super().__matmul__(other)

def scaling(x, y, z):
    m = Matrix(np.identity(4))
    m[0, 0] = x
    m[1, 1] = y
    m[2, 2] = z
    return m

scaling(2,2,2) @ scaling(2,2,2)
# Matrix([[4., 0., 0., 0.],
#         [0., 4., 0., 0.],
#         [0., 0., 4., 0.],
#         [0., 0., 0., 1.]])
Tuple([1,2,3,4]) @ scaling(2,2,2)
# Tuple([2., 4., 6., 4.])
scaling(2,2,2) @ Tuple([1,2,3,4])
# Tuple([2., 4., 6., 4.])

Answer 4

只需重载__matmul__类的Matrix即可返回元组

class Matrix(np.ndarray):
    def __new__(cls, input_array, info=None):
        return np.asarray(input_array).view(cls)

    def __matmul__(self, other):
        return other @ self