How to get the size of numpy's numeric types in bytes?

Question

I want to get the size in bytes of numerical types, in particular those defined by numpy. This is achieved by this helper function,

def sizeof(dtype):
    a = np.zeros(1, dtype=dtype)
    return int(a.nbytes)

but I was wondering if there would be a built-in, less awkward way of doing it. sys.getsizeof is of no help in this case — it returns 400 bytes for all of numpy's numeric types I have tested.

That's because that's the size of the object.

Mad Physicist
– Mad Physicist

2017-05-30 18:15:49 +00:00
Commented May 30, 2017 at 18:15 — Mad Physicist
– Mad Physicist, Commented May 30, 2017 at 18:15

MSeifert · Accepted Answer · 2017-05-30 18:16:41Z

8

Use dtype.itemsize or ndarray.itemsize:

>>> import numpy as np

>>> arr = np.array([1, 2, 3], dtype=np.int32)
>>> arr.dtype.itemsize
4

>>> arr.itemsize
4

answered May 30, 2017 at 18:16

MSeifert

154k41 gold badges356 silver badges377 bronze badges

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

5 Comments

P-Gn Over a year ago

Do you know why arr.dtype.itemsize works but say np.float32.itemsize does not?

MSeifert Over a year ago

Because it's an instance attribute - not a class attribute. np.float32().itemsize works.

Mad Physicist Over a year ago

So np.float32 is an extension of np.dtype, not an instance of it?

MSeifert Over a year ago

@MadPhysicist np.float32 is not a dtype it's a scalar type (that can be used as argument for the dtype parameter). dtype(np.float32) is the float32 dtype.

P-Gn Over a year ago

Interestingly, dtype(np.float32).itemsize also works.

Collectives™ on Stack Overflow

How to get the size of numpy's numeric types in bytes?

1 Answer 1

5 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

5 Comments

Related