Make it possible to initialize array from dictionary values in Python 3

At the moment arrays can’t be initialized from dictionary values in Python 3:

In [1]: import numpy as np

In [2]: d = {'a':1,'b':2,'c':3}

In [3]: np.array(d.values())
Out[3]: array(dict_values([2, 1, 3]), dtype=object)

and one has to instead explicitly convert to a list first:

In [4]: np.array(list(d.values()))
Out[4]: array([2, 1, 3])

It might be worth adding a special case in the initializer for the array to automatically do this conversion, since d.values() is a valid iterator?

[njsmith]

No, in python 3, dict.keys and dict.values (and also dict.items for that matter) don’t return iterators, they return special magical view classes. Which are iterable, of course, but they also have specific documented types.

I too am dubious about np.array accepting arbitrary iterables (because of memory overhead and the difficulty of even recognizing an iterable when you see one), but I’m +0.7 on teaching np.array to treat the dict views the same way we treat lists. On Mar 25, 2015 9:39 AM, “Jaime” [e-mail] wrote:

I don’t think that array has a way of knowing that what it received is a dict.keys() or dict.values() iterator. Any change would require accepting generic iterators, which is not an easy task, e.g. np.fromiter requires an explicit dtype, while np.array can figure it out from the input. So I don’t really see an easy way of making this happen…

— Reply to this email directly or view it on GitHub https://github.com/numpy/numpy/issues/5718#issuecomment-86110018.

[eric-wieser]

What is actually going on here?

This works fine:

class A(object):
    def __getitem__(self, index):
        if index > 3:
            raise IndexError
        return index

    def __len__(self): return 4

np.array(A())  # array([1, 2, 3, 4])

So why does np.array(dict_keys) not?

Edit: Duh, because dict_keys, despite having a __len__, has no __getitem__