generator to list conversion yields different results when it is done in or outside the a list comprehension

Question

I had a code snippet behaving weirdly; I made another to illustrate the problem. I think the two snippets should yield identical results but don't :

alphas = np.arange(5)
args = zip(np.arange(5), np.arange(5))

output = [np.mean([ a+b for a, b in args]) for _ in alphas]

yielding

[4.0, nan, nan, nan, nan]

and

alphas = np.arange(5)
args = zip(np.arange(5), np.arange(5))
args = list(args)  # <- this is the only difference

output = [np.mean([ a+b for a, b in args]) for _ in alphas]

yielding

[4.0, 4.0, 4.0, 4.0, 4.0]

The only difference is that the generator is converted to a list inside the list comprehension. Does anyone understand the underlying mechanism that makes the results different ?

`zip` returns an iterator, and you're trying to iterate over it multiple times (once for each `_ in alphas`). — Aran-Fey, Apr 04 '18 at 15:01
Thank you for your answer and for your edit. Sorry for swapping the outputs. — Ghazi Felhi, Apr 05 '18 at 10:19

score 2 · Accepted Answer · answered Apr 04 '18 at 15:03

2

A generator produces the output sequence only once. Your first expression uses the entire sequence on the out for's first iteration. When _ advances from 0 to 1 and cycles around to iterate through args again, there's nothing left to yield, and you have an empty return. This results in nan for the mean value.

In your second example, args is a static list; you can iterate through that as many times as you like.

answered Apr 04 '18 at 15:03

Prune

76,765
14
60
81

Oh I see ! Thank you for the explanation :) – Ghazi Felhi Apr 05 '18 at 10:09
And thank you for the edit. I apologize for the mistakes I made. – Ghazi Felhi Apr 05 '18 at 10:17

generator to list conversion yields different results when it is done in or outside the a list comprehension

1 Answers1