Size of data type using NumPy

Question

In NumPy, I can get the size (in bytes) of a particular data type by:

datatype(...).itemsize

or:

datatype(...).nbytes

For example:

np.float32(5).itemsize #4
np.float32(5).nbytes   #4

I have two questions. First, is there a way to get this information without creating an instance of the datatype? Second, what’s the difference between itemsize and nbytes?

Asked By: mgilson

||

Source

Answer 1

You need an instance of the dtype to get the itemsize, but you shouldn’t need an instance of the ndarray. (As will become clear in a second, nbytes is a property of the array, not the dtype.)

E.g.

print np.dtype(float).itemsize
print np.dtype(np.float32).itemsize
print np.dtype('|S10').itemsize

As far as the difference between itemsize and nbytes, nbytes is just x.itemsize * x.size.

E.g.

In [16]: print np.arange(100).itemsize
8

In [17]: print np.arange(100).nbytes
800

Answered By: Joe Kington

Answer 2

Looking at the NumPy C source file, this is the comment:

size : int
    Number of elements in the array.
itemsize : int
    The memory use of each array element in bytes.
nbytes : int
    The total number of bytes required to store the array data,
    i.e., ``itemsize * size``.

So in NumPy:

>>> x = np.zeros((3, 5, 2), dtype=np.float64)
>>> x.itemsize
8

So .nbytes is a shortcut for:

>>> np.prod(x.shape)*x.itemsize
240
>>> x.nbytes
240

So, to get a base size of a NumPy array without creating an instance of it, you can do this (assuming a 3x5x2 array of doubles for example):

>>> np.float64(1).itemsize * np.prod([3,5,2])
240

However, important note from the NumPy help file:

|  nbytes
|      Total bytes consumed by the elements of the array.
|
|      Notes
|      -----
|      Does not include memory consumed by non-element attributes of the
|      array object.

Answered By: dawg

Size of data type using NumPy

Question:

Answers: