As the following code shows, empty array replaces my int values with float ones. How can I prevent this?
import numpy as np
a=np.empty(3)
a[0]=1
a[1]=2
a[2]=3
print a
Output:
[1., 2., 3.]
2 Answers
Specify the dtype when you call np.empty:
a = np.empty(3, dtype='int')
If you do not specify a dtype, the default is float. This is the call signature of np.empty:
empty(shape, dtype=float, order='C')
4 Comments
Cupitor
But why does it cast it to float automatically?!?!
unutbu
Every NumPy array has a dtype. The default is
float.Cupitor
That seems nut always to be true, or maybe I am confused because for example
numpy.asarray() will keep the format of int if the values are integer.
abarnert
@Naji: It's always true. I explained this in my comment. When it knows the type you want while constructing the array, it can use that type. When it doesn't know the type you want until later, it has to guess the type at construction time, and it always guesses
float.Use dtype=int:
>>> a = np.empty(3, dtype=np.int)
>>> a[0]=1
>>> a[1]=2
>>> a[2]=3
>>> a
array([1, 2, 3])
As the default value for dtype is float for numpy.empty, so your assigned values gets converted to float.
empty(...)
empty(shape, dtype=float, order='C')
answered Nov 25, 2013 at 19:40
Ashwini Chaudhary
252k60 gold badges478 silver badges519 bronze badges
emptyand then fill in the values one by one, instead of just doinga = np.array([1, 2, 3])? That would have avoided the problem you're seeing, as the defaultdtypefor an array constructed from a sequence is determined by the values of the sequence, so you would have gottenint; when it doesn't know the values, it doesn't have any information, so it has to default tofloat.forloop for that. The problem is it doesn't do it always. For examplenumpy.asarray()will keep the format!asarrayorarraycan keep the type, because it knows the type;emptycan't keep the type, because it doesn't. (Also, you can easily turn yourforloop into an iterator, e.g., by wrapping it in a generator function, and then pass the iterator to, e.g.,np.fromiter. Or, of course, build alistto pass tonp.arrayinstead.)