如何正确连接numpy的2-d ndarray中的字符串元素？

Question

In below code, I expect arr2 return same list of string as lst2 , but it doesn't. Why lst2 and arr2 are different? Is there any NumPythonic way to make arr2 return same list of string as lst2 ?

code:

import numpy as np

lst = [['MI', '', 'P'], 
       ['B', 'N', 'SUFS'],
       ['KOS', 'XJRXA', 'JJHW'],
       ['ARI', 'TPKI', ''],
       ['VR', 'EYR', '']]

arr = np.array(lst)

arr2 = np.apply_along_axis(lambda x: "".join(x), 1, arr)
lst2 = list(map(lambda x: "".join(x), lst))

print('lst:', lst)
print('arr:', arr.tolist())
print('lst2:', lst2)
print('arr2:', arr2.tolist())

output:

lst: [['MI', '', 'P'], ['B', 'N', 'SUFS'], ['KOS', 'XJRXA', 'JJHW'], ['ARI', 'TPKI', ''], ['VR', 'EYR', '']]
arr: [['MI', '', 'P'], ['B', 'N', 'SUFS'], ['KOS', 'XJRXA', 'JJHW'], ['ARI', 'TPKI', ''], ['VR', 'EYR', '']]
lst2: ['MIP', 'BNSUFS', 'KOSXJRXAJJHW', 'ARITPKI', 'VREYR']
arr2: ['MIP', 'BNS', 'KOS', 'ARI', 'VRE']

Answer 1

Pandas will do it easily:

pd.DataFrame(arr).sum(axis=1)

The reason np.apply_along_axis() gives you trouble is that it infers the length of the strings in the result from the first row. Since you end up with MIP in the first row, all the rows have a capacity of 3 characters which is not what you want.

There is a NumPy bug report for apply_along_axis() with more information: https://github.com/numpy/numpy/issues/8352

Answer 2

Thanks. I found the answer from https://github.com/numpy/numpy/issues/8352#issuecomment-488133970 .

import numpy as np

lst = [['MI', '', 'P'], ['B', 'N', 'SUFS'], ['KOS', 'XJRXA', 'JJHW'], ['ARI', 'TPKI', ''], ['VR', 'EYR', '']]
arr = np.array(lst)

arr2 = np.apply_along_axis(lambda x: np.asarray("".join(x), dtype=object), 1, arr)
# https://github.com/numpy/numpy/issues/8352#issuecomment-488133970
lst2 = list(map(lambda x: "".join(x), lst))

print('lst:', lst)
print('arr:', arr.tolist())
print('lst2:', lst2)
print('arr2:', arr2.tolist())