numpy 广播机制

1.  +. -. *. /都会激发广播机制

>>> s
array([[1],
       [2],
       [3]])
>>> ss
array([[1, 2],
       [3, 4],
       [5, 6]])
>>> s
array([[1],
       [2],
       [3]])
>>> ss + s
array([[2, 3],
       [5, 6],
       [8, 9]])
>>> ss - s
array([[0, 1],
       [1, 2],
       [2, 3]])
>>> ss * s
array([[ 1,  2],
       [ 6,  8],
       [15, 18]])
>>> ss / s
array([[1.        , 2.        ],
       [1.5       , 2.        ],
       [1.66666667, 2.        ]])

• 标量vs向量/矩阵
• 会自动扩展至纬度相等(此处标量->(3, 1))

>>> s
array([[1],
       [2],
       [3]])
>>> s + 1
array([[2],
       [3],
       [4]])
>>> s - 1
array([[0],
       [1],
       [2]])
>>> s * 2
array([[2],
       [4],
       [6]])
>>> s / 2
array([[0.5],
       [1. ],
       [1.5]])

• 向量vs矩阵
• 会自动扩展至纬度相等(此处向量->(3, 2))
• 必须要有一个维度是相等的才会发生广播，否则报错

>>> s
array([[1],
       [2],
       [3]])
>>> ss
array([[1, 2],
       [3, 4],
       [5, 6]])
>>> s
array([[1],
       [2],
       [3]])
>>> ss + s
array([[2, 3],
       [5, 6],
       [8, 9]])
>>> ss - s
array([[0, 1],
       [1, 2],
       [2, 3]])
>>> ss * s
array([[ 1,  2],
       [ 6,  8],
       [15, 18]])
>>> ss / s
array([[1.        , 2.        ],
       [1.5       , 2.        ],
       [1.66666667, 2.        ]])