Possible typo in kernels.py:sum_reduction function

[oleksandr-pavlyk]

https://github.com/soda-inria/sklearn-numba-dpex/blob/main/sklearn_numba_dpex/common/kernels.py#L221-L225

The function reads

def sum_reduction(summands):
    for kernel, result in kernels_and_empty_tensors_pairs:
        kernel(summands, result) 
        summands = result
    return result

The returned variable result is outside of its scope of definition (local variable defined in the scope of the loop).

I believe the last line should be return summands instead.

[fcharras]

I can change this in a PR if it's more readable, but in fact in python the scope of the variables defined in a loop is the same scope that other local variables.

e.g you can do

for i in range(10):
    pass
print(i)

output will be 9.

BTW we just merged unit tests for the reduction kernels and added a reduction for 2d tensors on axis 1 😁

(See #46 for some thinking on using custom kernels vs dpnp functions)

[oleksandr-pavlyk]

Thanks for highlighting #46 !

I think return summands would read better indeed, but its just a suggestion.

[ogrisel]

No strong opinion, the only difference is when the list of kernels is empty.