apply_along_axis cuts strings

I'm trying to concatenate all elements of a row into a string as follows:

np.apply_along_axis(lambda x: " ".join(map(str, x)), 1, b)

b is

[[111,111,0,0,0], [111,111,111,111,111]]

However, the result of the line is:

['111 111 0 0 0', '111 111 111 1']

It looks like np.apply_along_axis is cutting the second string to be of the same length as the first one. If I put a longer sequence first, the result is correct:

['111 111 111 111 111', '111 111 0 0 0']

So I'm guessing this is a bug?

Summary 2019-04-30 by @seberg

np.apply_along_axis infers the output dtype from the first pass. Which can be worked around for example but the function returning an array of a correct type.

Actions:

• np.apply_along_axis could/should get a dtype kwarg (or similar, compare also np.vectorize).

Start at the np.apply_along_axis entry point and reproduce the supplied example, comparing results when the longer row comes first or second. Determine how the first pass sets the output dtype; done means the example preserves both complete strings, with the chosen dtype-handling behavior reflected in the API.