Deep copy of a np.array of np.array

Question

I have a numpy array of different numpy arrays and I want to make a deep copy of the arrays. I found out the following:

import numpy as np  pairs = [(2, 3), (3, 4), (4, 5)] array_of_arrays = np.array([np.arange(a*b).reshape(a,b) for (a, b) in pairs])  a = array_of_arrays[:] # Does not work b = array_of_arrays[:][:] # Does not work c = np.array(array_of_arrays, copy=True) # Does not work d = np.array([np.array(x, copy=True) for x in array_of_arrays])  array_of_arrays[0][0,0] = 100 print a[0][0,0], b[0][0,0], c[0][0,0], d[0][0,0] 

Is d the best way to do this? Is there a deep copy function I missed? And what is the best way to interact with each element in this array of different sized arrays?

Answer 1

import numpy as np import copy  pairs = [(2, 3), (3, 4), (4, 5)] array_of_arrays = np.array([np.arange(a*b).reshape(a,b) for (a, b) in pairs])  a = copy.deepcopy(array_of_arrays) 

Feel free to read up more about this here.

Oh, here is simplest test case:

a[0][0,0] print a[0][0,0], array_of_arrays[0][0,0] 

Answer 2

Beaten by one minute. Indeed, deepcopy is the answer here.

To your second question abut indexing: I have a feeling that you may be better off with a simple list or a dictionary-type data structure here. np.arrays make sense primarily if each array element is of the same type. Of course you can argue that each element in array_of_arrays is another array, but what is the benefit of having them collected in a numpy array instead of a simple list?

list_of_arrays = [np.arange(a*b).reshape(a,b) for (a, b) in pairs]