In [207]: B=np.array([[[1],[2],[23]],[[2],[4],[21]],[[6],[45],[61]],[[1],[34],[231]]])
In [208]: B
Out[208]:
array([[[ 1],
[ 2],
[ 23]],
[[ 2],
[ 4],
[ 21]],
[[ 6],
[ 45],
[ 61]],
[[ 1],
[ 34],
[231]]])
In [209]: B.shape
Out[209]: (4, 3, 1)
reshape保持顺序,只是重新排列维度的大小:
In [210]: B.reshape(3,1,4)
Out[210]:
array([[[ 1, 2, 23, 2]],
[[ 4, 21, 6, 45]],
[[ 61, 1, 34, 231]]])
请注意,您可以按照创建B 时使用的相同顺序读取1,2,23,2,...。
transpose 是不同的操作:
In [211]: B.transpose(1,2,0)
Out[211]:
array([[[ 1, 2, 6, 1]],
[[ 2, 4, 45, 34]],
[[ 23, 21, 61, 231]]])
In [212]: _.shape
Out[212]: (3, 1, 4)
In [213]: __.ravel()
Out[213]: array([ 1, 2, 6, 1, 2, 4, 45, 34, 23, 21, 61, 231])
1,2,23,... 订单仍然存在 - 如果您仔细阅读这些行。但是拆线的顺序变了。
In [216]: B.transpose(1,2,0).ravel(order='F')
Out[216]: array([ 1, 2, 23, 2, 4, 21, 6, 45, 61, 1, 34, 231])
In [217]: B[1,:,:]
Out[217]:
array([[ 2],
[ 4],
[21]])
In [218]: B.transpose(1,2,0)[:,:,1]
Out[218]:
array([[ 2],
[ 4],
[21]])