Geek Answers Handbook

Numbering link without creating an expensive copy

Let's say that I have a function that requires NumPy ndarraywith two axes, for example, a data matrix of rows and columns. If a "column" is cut from such an array, this function should also work, so for convenience it should do some internal ones X[:, np.newaxis]. However, I do not want to create a new array object for this, as in some cases this can be expensive.

I am wondering if there is a good way to do this. For example, will the following code be safe (I mean, will global arrays be immutable, like Python lists)?

X1 = np.array([[1,2,3], [4,5,6], [7,8,9]])
X2 = np.array([1,4,7])

def some_func(X):
    if len(X.shape) == 1:
        X = X[:, np.newaxis]
    return X[:,0].sum()       

some_func(X2)
some_func(X1[:, 0])
some_func(X1)

I ask because I heard that NumPy arrays are sometimes copied in some cases, however I cannot find a good resource about this. Any ideas?

+4
3

. :

>>> A = np.ones((50000000,))
>>> B = A[:,np.newaxis]
>>> B.flags
  C_CONTIGUOUS : False
  F_CONTIGUOUS : False
  OWNDATA : False
  WRITEABLE : True
  ALIGNED : True
  UPDATEIFCOPY : False

OWNDATA : False - A.

http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html. , , (, A[[1,2,4]]), (, A[[True, False, True]]). .

+5

- - ndarray, .

+1

, M x N, M - . , . .

, IF -

def some_func2(X):
    return X.reshape(X.shape[0],-1)[:,0].sum()

, , np.may_share_memory -

In [515]: X1
Out[515]: 
array([[1, 2, 3],
       [4, 5, 6],
       [7, 8, 9]])

In [516]: np.may_share_memory(X1,X1.reshape(X1.shape[0],-1)[:,0])
Out[516]: True

In [517]: X2
Out[517]: array([1, 4, 7])

In [518]: np.may_share_memory(X2,X2.reshape(X2.shape[0],-1)[:,0])
Out[518]: True

A True np.may_share_memory , , .

+1

More articles:


All Articles