How do I convert a float NumPy array into an int NumPy array?
Add a comment
|
4 Answers
Use the astype method.
>>> x = np.array([[1.0, 2.3], [1.3, 2.9]])
>>> x
array([[ 1. , 2.3],
[ 1.3, 2.9]])
>>> x.astype(int)
array([[1, 2],
[1, 2]])
5 Comments
Garrett
Just make sure you don't have
np.infor np.nan in your array, since they have surprising results. For example, np.array([np.inf]).astype(int) outputs array([-9223372036854775808]).
BallpointBen
On my machine,
np.array([np.inf]).astype(int), np.array([-np.inf]).astype(int), and np.array([np.nan]).astype(int) all return the same thing. Why?
BrenBarn
@BallpointBen:
nan and inf are floating-point values and can't be meaningfully converted to int. As the comment before yours notes, there will be surprising behavior, and I don't think the precise behavior is well-defined. If you want to map nan and inf to certain values, you need to do that yourself.
chris
Note that x.astype(int)[0][0] is not of type
int. It's numpy.int32.
Nathan Musoke
Note that although this does convert the array to ints, @fhtuft's answer that may result in less surprises
Some numpy functions for how to control the rounding: rint, floor,trunc, ceil. depending how u wish to round the floats, up, down, or to the nearest int.
>>> x = np.array([[1.0,2.3],[1.3,2.9]])
>>> x
array([[ 1. , 2.3],
[ 1.3, 2.9]])
>>> y = np.trunc(x)
>>> y
array([[ 1., 2.],
[ 1., 2.]])
>>> z = np.ceil(x)
>>> z
array([[ 1., 3.],
[ 2., 3.]])
>>> t = np.floor(x)
>>> t
array([[ 1., 2.],
[ 1., 2.]])
>>> a = np.rint(x)
>>> a
array([[ 1., 2.],
[ 1., 3.]])
To make one of this in to int, or one of the other types in numpy, astype (as answered by BrenBern):
a.astype(int)
array([[1, 2],
[1, 3]])
>>> y.astype(int)
array([[1, 2],
[1, 2]])
5 Comments
Bakuriu
Exactly what I was looking for.
astype is often too generic, and I think it probably is more useful when doing intx - inty conversions. When I want to do float - int conversion being able to choose the kind of rounding is a nice feature.
endolith
So the simplest way to safely convert almost-ints like
7.99999 to ints like 8, is np.rint(arr).astype(int)?
Ryan
any way in numpy to make it uint8?
chris
@Ryan
astype(np.uint8)
Divyanshu Srivastava
Can you also comment on the runtime of these methods ?
you can use np.int_:
>>> x = np.array([[1.0, 2.3], [1.3, 2.9]])
>>> x
array([[ 1. , 2.3],
[ 1.3, 2.9]])
>>> np.int_(x)
array([[1, 2],
[1, 2]])
answered Feb 1, 2016 at 7:33
Comments
If you're not sure your input is going to be a Numpy array, you can use asarray with dtype=int instead of astype:
>>> np.asarray([1,2,3,4], dtype=int)
array([1, 2, 3, 4])
If the input array already has the correct dtype, asarray avoids the array copy while astype does not (unless you specify copy=False):
>>> a = np.array([1,2,3,4])
>>> a is np.asarray(a) # no copy :)
True
>>> a is a.astype(int) # copy :(
False
>>> a is a.astype(int, copy=False) # no copy :)
True
