Python numpy convert dtype object to dtype float32 in case of a 2d array

Question

Given this code:

arr = np.array(
    [
        [
            np.array([1, 2, 3]),
            np.array([4, 9])
        ],
        [
            np.array([1, 2, 5]),
            np.array([5, 6])
        ],
        [
            np.array([1, 2, 5]),
            np.array([85, 86])
        ]
    ]
)

(input, expected) = arr.T

input[:2]

I am getting this output:

array([array([1, 2, 3]), array([1, 2, 5])], dtype=object)

But I want this to be a 2d array of type dtype float32 .

So if I create it using a different way:

np.array(
    [
        [1, 2, 3],
        [1, 2, 5]
    ], dtype=np.float32
)

This is what I want the output to look like:

array([[1., 2., 3.],
       [1., 2., 5.]], dtype=float32)

How do I convert input into this 2d array, so I won't see dtype=object , but dtype=float32 ?

I've tried this:

np.array(input[:2], dtype=np.float32)

but it prints this error:

ValueError: setting an array element with a sequence.

I've also looked at other functions like view , asarray , all of them resulted in a similar error.

Answer 1

I found it eventually:

This doesn't work:

np.array(input, dtype=np.float32)

But copying it does:

np.array([x for x in input], dtype=np.float32)