The objective is to concatenate nested list of arrays (i.e., list_arr). However, some of the sublists within the list_arr is of len zero.

Simply using np.array or np.asarray on the list_arr does not produce the intended result.

import numpy as np

ncondition=2
nnodes=30
nsbj=6


np.random.seed(0)

# Example of nested list list_arr

list_arr=[[[np.concatenate([[idx_sbj],[ncondi],[nepoch] ,np.random.rand(nnodes)]) for nepoch in range(np.random.randint(5))] \
 for ncondi in range(ncondition)] for idx_sbj in range(nsbj)]

The following does not produce the expected concatenate output

test1=np.asarray(list_arr)
test2=np.array(list_arr)
test3= np.vstack(list_arr)

The expected output is an array of shapes (15,33)

CodePudding user response：

OK, my curiosity got the better of me.

Make an object dtype array from the list:

In [280]: arr=np.array(list_arr,object)
In [281]: arr.shape
Out[281]: (6, 2)

All elements of this array are lists, with len:

In [282]: np.frompyfunc(len,1,1)(arr)
Out[282]: 
array([[4, 1],
       [0, 2],
       [0, 2],
       [0, 0],
       [2, 3],
       [1, 0]], dtype=object)

Looking at specific sublists. One has two empty lists

In [283]: list_arr[3]
Out[283]: [[], []]

others have one empty list, either first or second:

In [284]: list_arr[-1]
Out[284]: 
[[array([5.        , 0.        , 0.        , 0.3681024 , 0.3127533 ,
         0.80183615, 0.07044719, 0.68357296, 0.38072924, 0.63393096,
         ...])],
 []]

and some have lists of differing numbers of arrays:

If I add up the numbers in [282] I get 15, so that must be where you get the (15,33). And presumably all the arrays have the same length.

The outer layer of nesting isn't relevant, so we can ravel and remove it.

In [295]: alist = arr.ravel().tolist()

then filter out the empty lists, and apply vstack to the remaining:

In [296]: alist = [np.vstack(x) for x in alist if x]
In [297]: len(alist)
Out[297]: 7

and one more vstack to join those:

In [298]: arr2 = np.vstack(alist)
In [299]: arr2.shape
Out[299]: (15, 33)