将 numpy arrays 的列表转换为 numpy 数组，其中元素为 numpy arrays

Question

Python code to convert a list of numpy arrays into a numpy array having numpy arrays as its elements?

I have a list of arrays as below:

my_list

Output:

[array([1,2,3]),
    array(['a', 'b','c]),
    array(['text1', 'text2', 'text3'])
    ]

using np.asarray() converts the elements to list like the one shown below:

my_array = np.asarray(my_list)

my_array looks like:

array([[1,2,3],['a', 'b','c], ['text1', 'text2', 'text3']])

The kind of output that I want:

array([array([1,2,3]),array(['a', 'b','c]),array(['text1', 'text2', 'text3'])])

Can someone please help me with the code?

Answer 1

In [173]: alist = [np.array([1,2,3]),
     ...:     np.array(['a', 'b','c']),
     ...:     np.array(['text1', 'text2', 'text3'])
     ...:     ]
In [174]: alist
Out[174]: 
[array([1, 2, 3]),
 array(['a', 'b', 'c'], dtype='<U1'),
 array(['text1', 'text2', 'text3'], dtype='<U5')]
In [175]: np.array(alist)
Out[175]: 
array([['1', '2', '3'],
       ['a', 'b', 'c'],
       ['text1', 'text2', 'text3']], dtype='<U21')

np.array tries to make as high of a multidimensional array as it can. The 3 arrays of the same size it can make a (3,3).

To avoid that we have to create a target array of the desired shape and dtype, and assign the values from the list:

In [178]: arr = np.empty(3, object)
In [179]: arr[:] = alist
In [180]: arr
Out[180]: 
array([array([1, 2, 3]), array(['a', 'b', 'c'], dtype='<U1'),
       array(['text1', 'text2', 'text3'], dtype='<U5')], dtype=object)

If the list contains arrays (or other objects) of a different size, np.array falls back to making an object dtype array of those items. But pay attention to the warning, and keep in mind that this is a fallback option, and as you found, not the default behavior.

In [181]: alist.append([1,2])
In [182]: np.array(alist)
<ipython-input-182-7512d762195a>:1: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.
  np.array(alist)
Out[182]: 
array([array([1, 2, 3]), array(['a', 'b', 'c'], dtype='<U1'),
       array(['text1', 'text2', 'text3'], dtype='<U5'), list([1, 2])],
      dtype=object)

And to repeat my comment - why do you want an array of arrays? What advantage does that have compared to the original list?