Prevent numpy object array from casting elements

Question

I am trying to construct a numpy array of pandas MultiIndexes. The array has dtype='object' and each element of the array should be a pd.MultiIndex.

Here's an example:

a  = pd.MultiIndex.from_product([pd.RangeIndex(3), pd.RangeIndex(2)])
b  = pd.MultiIndex.from_product([pd.RangeIndex(3), pd.RangeIndex(2)])
array = np.array([a,b], dtype='object')

The output is:

array([[(0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1)],
       [(0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1)]], dtype=object)

What i would like the output to be:

array([MultiIndex([(0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1)]),
       MultiIndex([(0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1)])], dtype=object)

I would assume that specifying dtype='object' when creating the array would prevent numpy from casting the input to some other python type, but apparently it is casting my multiindexes into list. How do i stop it from casting?

Thanks!

why not just use a list [a,b]? What does object array do for you? — hpaulj
– hpaulj, Commented Apr 8, 2022 at 9:49
I need the array because i need to use a lot of numpy-specific tools on it, like flatten and reshape. This array is used to construct another 'real' numpy array. If i could use a normal list I would, believe me. — pnjun
– pnjun, Commented Apr 8, 2022 at 9:52
That will depend on the input data. Could be as easy as (1,). But it might also be (30,30,20) — pnjun
– pnjun, Commented Apr 8, 2022 at 10:18
I wanted the shape of the actual array you show, I was puzzled by the tuples in what looks like a (2,6) array. Why wasn't it (2,6,2). Turns out a.values is a 1d object dtype array of tuples. If they differed in length it would keep them separate, but since they match it can make a 2d array instad. That's normal np.array behavior. — hpaulj
– hpaulj, Commented Apr 8, 2022 at 16:15

Ynjxsjmh · Accepted Answer · 2022-04-08 12:45:57Z

2

You can use np.empty to create an empty array of the desired shape and then copy the data to the array:

lst = [a, b]
array = np.empty(len(lst), dtype=object)
array[:] = lst[:]

[MultiIndex([(0, 0),
             (0, 1),
             (1, 0),
             (1, 1),
             (2, 0),
             (2, 1)],
            )         MultiIndex([(0, 0),
                                  (0, 1),
                                  (1, 0),
                                  (1, 1),
                                  (2, 0),
                                  (2, 1)],
                                 )        ]

answered Apr 8, 2022 at 12:45

Ynjxsjmh

30.3k7 gold badges43 silver badges64 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Prevent numpy object array from casting elements

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related