BUG? : np.ravel does not flatten out certain arrays

For arrays in which the number of elements are not the same along the inner axes, np.ravel and np.flatten return the array as it is. I’m not sure if this is intended behavior or not.

In [23]: a = np.ravel([[1, 2], [3, 4, 5]])

In [24]: a
Out[24]: array([[1, 2], [3, 4, 5]], dtype=object)

I thought of giving this as a shot, but I’m not sure what part of the source I should modify, the ravel function just says asarray(a).ravel() and I was not able to figure out (quickly) where the ravel methods for the ndarray is.

[mattip]

NumPy is not designed to handle arrays of lists of arbitrary length. While you can use the numpy hammer for this bowl of pudding you will end up just making a mess and it will not be faster than using a spoon. My advice would be to avoid numpy altogether:

lst = [[1, 2], [3, 4, 5]]
flat = [val for sublist in lst for val in sublist]

[cody-wan]

I find sum(z,[]) to be the fastest as well. It takes about 1/2 of the time for list(itertools.chain(*z)) and 1/3 of the time for [x for y in z for x in y], using %timeit magic command. The data type in my list of lists is strings only.