[Numpy] 数组和内存的浅复制和深复制（View vs Copy)_numpy中slice是view还是copy-优快云博客

本文介绍了Numpy中数组切片的View（浅复制）和Copy（深复制）的区别，重点关注切片操作何时产生视图何时产生复制。当索引对象为非元组序列、numpy数组或包含序列的元组时，会产生复制；其他情况则创建视图，以节省内存。同时，文章提到了as_strided方法始终创建视图，以及如何通过flags.owndata检查复制类型。内容还涉及LaTeX数学表达式、甘特图、UML图表和Flowchart流程图的使用。

Numpy数组和内存 View vs Copy 或称浅复制或深复制

一句话概括最重要的区别：

在对numpy切片（slicing）的时候，如果切片的表达式内含有带括号的非tuple对象，如数列，或者另一个numpy数列，或者tuple对象（但至少包含一个含有序列的对象或者一个numpy数列），则进行复制操作。其他情况下进行View操作，只是访问了原有numpy数列的内存。

（这个view是指浅复制，和view方法没有关系。）

as_strided方法任何时候都只是View操作。
flags.owndata可以查看view or copy。

例子和英文解释摘自课堂笔记。

numpy uses pass-by-reference semantics so that slice operations are views into the array without implicit copying, which is consistent with python’s semantics. this is particularly helpful with large arrays that already strain available memory. in numpy terminology, slicing creates views (no copying) and advanced indexing creates copies. let’s start with advanced indexing.
if the indexing object (i.e., the item between the brackets) is a non-tuple sequence object, another numpy array (of type integer or boolean), or a tuple with at least one sequence object or numpy array, then indexing creates copies. for the above example, to accomplish the same array extension in numpy, you have to do something like the following

>>> x = np.ones((3,3))
>>> x
array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]])
>>> x[:,[0,1,2,2]] # notice duplicated last dimension
array([[1., 1., 1., 1.],
       [1., 1., 1., 1.],
       [1., 1., 1., 1.]])
>>> y=x[:,[0,1,2,2]] # same as above, but do assign it to y
# y=x[:,(0,1,2,2)] has the same results.

>>> x[0,0]=999 # change element in x
>>> x                         # changed
array([[999.,   1.,   1.],
       [  1.,   1.,   1.],
       [  1.,   1.,   1.]])
>>> y                         # not changed!
array([[1., 1., 1., 1.],
       [1., 1., 1., 1.],
       [1., 1., 1., 1.]])

>>> x = np.ones((3,3))
>>> y = x[:2,:2] # view of upper left piece
>>> x[0,0] = 999 # change value
>>> x  # see the change?
array([[999.,   1.,   1.],
       [  1.,   1.,   1.],
       [  1.,   1.,   1.]])
>>> y
array([[999.,   1.],
       [  1.,   1.]])

>>> x = np.arange(5) # create array
>>> x
array([0, 1, 2, 3, 4])
>>> y=x[[0,1,2]] # index by integer list to force copy
>>> y
array([0, 1, 2])
>>> z=x[:3]      # slice creates view
>>> z            # note y and z have same entries
array([0, 1, 2])
>>> x[0]=999     # change element of x
>>> x
array([999,   1,   2,   3,   4])
>>> y            # note y is unaffected,
array([0, 1, 2])
>>> z            # but z is (it's a view).
array([999,   1,   2])
in this example, y is a copy, not a view, because it was created using advanced indexing


>>> from numpy.lib.stride_tricks import as_strided
>>> x = np.arange(16).astype(np.int32)
>>> y=as_strided(x,(7,4),(8,4)) # overlapped entries
>>> y
array([[ 0,  1,  2,  3],
       [ 2,  3,  4,  5],
       [ 4,  5,  6,  7],
       [ 6,  7,  8,  9],
       [ 8,  9, 10, 11],
       [10, 11, 12, 13],
       [12, 13, 14, 15]], dtype=int32)

>>> x[::2]=99 # assign every other value
>>> x
array([99,  1, 99,  3, 99,  5, 99,  7, 99,  9, 99, 11, 99, 13, 99, 15],
      dtype=int32)
>>> y # the changes appear because y is a view
array([[99,  1, 99,  3],
       [99,  3, 99,  5],
       [99,  5, 99,  7],
       [99,  7, 99,  9],
       [99,  9, 99, 11],
       [99, 11, 99, 13],
       [99, 13, 99, 15]], dtype=int32)
       
>>> n = 8 # number of elements
>>> x = np.arange(n) # create array
>>> k = 5 # desired number of rows
>>> y = as_strided(x,(k,n-k+1),(x.itemsize,)*2)
>>> y
array([[0, 1, 2, 3],
       [1, 2, 3, 4],
       [2, 3, 4, 5],
       [3, 4, 5, 6],
       [4, 5, 6, 7]])

# Use flags.owndata to see wether it is a view or copy.