import cProfile

length = int(input())
string = input()
row = list(map(int, string.split()))
halfrow = sum(row)/2
beg = eval('0b11'+ (length - 1) * '0')
end = eval('0b1'+ length * '1')

def func():

    res = halfrow
    for i in range(beg, end):
        summa = 0

        for j in range(length):

            if i >> j & 1 == 1:
                summa += row[j]

        diff = abs(halfrow - summa)

        if diff == 0:
            return 0

        if diff < res:
            res = diff
    return int(res * 2)

cProfile.run('print(func())')


$ cat 1005
20
101 51 51 3 2 2 40 23 895 345 4 38 678 97 12 234 55 8054 26 108

$ cat 1005 | python3 untitled19.py
5289
         524292 function calls in 3.163 seconds

   Ordered by: standard name

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    3.163    3.163 <string>:1(<module>)
        1    3.107    3.107    3.163    3.163 untitled19.py:10(func)
   524287    0.056    0.000    0.056    0.000 {built-in method builtins.abs}
        1    0.000    0.000    3.163    3.163 {built-in method builtins.exec}
        1    0.000    0.000    0.000    0.000 {built-in method builtins.print}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

So, why same method of solutions takes about 0.1 sec for C++ and about 3.1 sec for python?
The most loaded place is bit shifting.