View#

It is possible to access the array differently by just changing certainmetadata likestride anddtype without changing thedata buffer. This creates a new way of looking at the data and these newarrays are called views. The data buffer remains the same, so any changes madeto a view reflects in the original copy. A view can be forced through thendarray.view method.

Copy#

When a new array is created by duplicating the data buffer as well as themetadata, it is called a copy. Changes made to the copydo not reflect on the original array. Making a copy is slower andmemory-consuming but sometimes necessary. A copy can be forced by usingndarray.copy.

Indexing operations#

Views are created when elements can be addressed with offsets and stridesin the original array. Hence, basic indexing always creates views.For example:

>>>importnumpyasnp>>>x=np.arange(10)>>>xarray([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])>>>y=x[1:3]# creates a view>>>yarray([1, 2])>>>x[1:3]=[10,11]>>>xarray([ 0, 10, 11,  3,  4,  5,  6,  7,  8,  9])>>>yarray([10, 11])

Here,y gets changed whenx is changed because it is a view.

Advanced indexing, on the other hand, always creates copies.For example:

>>>importnumpyasnp>>>x=np.arange(9).reshape(3,3)>>>xarray([[0, 1, 2],       [3, 4, 5],       [6, 7, 8]])>>>y=x[[1,2]]>>>yarray([[3, 4, 5],       [6, 7, 8]])>>>y.baseisNoneTrue

Here,y is a copy, as signified by thebaseattribute. We can also confirm this by assigning new values tox[[1,2]]which in turn will not affecty at all:

>>>x[[1,2]]=[[10,11,12],[13,14,15]]>>>xarray([[ 0,  1,  2],       [10, 11, 12],       [13, 14, 15]])>>>yarray([[3, 4, 5],       [6, 7, 8]])

It must be noted here that during the assignment ofx[[1,2]] no viewor copy is created as the assignment happens in-place.

Other operations#

Thenumpy.reshape function creates a view where possible or a copyotherwise. In most cases, the strides can be modified to reshape thearray with a view. However, in some cases where the array becomesnon-contiguous (perhaps after andarray.transpose operation),the reshaping cannot be done by modifying strides and requires a copy.In these cases, we can raise an error by assigning the new shape to theshape attribute of the array. For example:

>>>importnumpyasnp>>>x=np.ones((2,3))>>>y=x.T# makes the array non-contiguous>>>yarray([[1., 1.],       [1., 1.],       [1., 1.]])>>>z=y.view()>>>z.shape=6Traceback (most recent call last):...AttributeError:Incompatible shape for in-place modification. Use`.reshape()` to make a copy with the desired shape.

Taking the example of another operation,ravel returns a contiguousflattened view of the array wherever possible. On the other hand,ndarray.flatten always returns a flattened copy of the array.However, to guarantee a view in most cases,x.reshape(-1) may be preferable.

How to tell if the array is a view or a copy#

Thebase attribute of the ndarray makes it easyto tell if an array is a view or a copy. The base attribute of a view returnsthe original array while it returnsNone for a copy.

>>>importnumpyasnp>>>x=np.arange(9)>>>xarray([0, 1, 2, 3, 4, 5, 6, 7, 8])>>>y=x.reshape(3,3)>>>yarray([[0, 1, 2],       [3, 4, 5],       [6, 7, 8]])>>>y.base# .reshape() creates a viewarray([0, 1, 2, 3, 4, 5, 6, 7, 8])>>>z=y[[2,1]]>>>zarray([[6, 7, 8],       [3, 4, 5]])>>>z.baseisNone# advanced indexing creates a copyTrue

Note that thebase attribute should not be used to determineif an ndarray object isnew; only if it is a view or a copyof another ndarray.