Create a Numpy scalar from dtype

Question

I'm trying to create a numpy scalar of a specified dtype. I know I could do, say, x = numpy.int16(3), but I don't know the dtype in advance.

If I were to want an array then

dtype = int
x = numpy.array(3, dtype=dtype)

would do it, so I had high hopes for

x = numpy.generic(3, dtype=dtype)

but one cannot create an instance of numpy.generic.

Any ideas?

I'm not sure what the problem is. Why doesn't x = dtype(3) do what you want? — DSM
– DSM, Commented Dec 27, 2013 at 17:36
@DSM: Strictly speaking, that doesn't do it. In the OP's example that would create int(3) instead of numpy.int32(3). Whether that's an issue or not, I don't know. — NPE
– NPE, Commented Dec 27, 2013 at 17:49
@NPE: I'm assuming that by "of a specified dtype", the OP has put the wanted dtype in dtype; int will give int, np.int16 will give np.int16. I must be missing something, though, because the OP knows that you can call the type object (in the first line) and also that you can bind the type objects to dtype. — DSM
– DSM, Commented Dec 27, 2013 at 17:53
@DSM, sorry for confusion, I have code like a = np.eye(2, dtype=int), b = np.zeros(2, dtype=a.dtype), so a.dtype can be passed as a dtype argument, but a.dtype(3) gives a not-callable error. The .type method does what I need ("make dtypes callable"). — n00b
– n00b, Commented Dec 27, 2013 at 18:09

gsakkis · Accepted Answer · 2021-03-02 20:15:41Z

11

As commented, the accepted answer is not generally correct for all dtypes, for example timedeltas:

In [6]: d = np.dtype("timedelta64[10m]")

In [7]: d.type(3)
Out[7]: numpy.timedelta64(3)

To always get the correct answer use the following:

In [8]: np.array(3, d)[()]
Out[8]: numpy.timedelta64(3,'10m')

answered Mar 2, 2021 at 20:15

gsakkis

1,6891 gold badge17 silver badges27 bronze badges

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

1 Comment

Rémy Hosseinkhan Boucher Over a year ago

too difficult to read:/

NPE · Accepted Answer · 2013-12-27 17:36:56Z

10

The following will create a scalar of x's dtype:

In [18]: val = x.dtype.type(3)

In [19]: val
Out[19]: 3

In [20]: type(val)
Out[20]: numpy.int32

answered Dec 27, 2013 at 17:36

NPE

503k114 gold badges970 silver badges1k bronze badges

2 Comments

n00b Over a year ago

Thanks - I didn't know about the .type method for dtypes

coderforlife Over a year ago

This however does not work for all dtypes nor does it preserve endian-ness for any dtype (always turns it into native endian). I believe the only way to preserve endian-ness and work with more dtypes (but probably still not all) is to create an array and get the element(s) from it.

Mr Tsjolder from codidact · Accepted Answer · 2017-09-24 18:36:24Z

5

In case someone would like to get the dtype from a string, you can use

import numpy as np
val = np.dtype('int32').type(3.0)

answered Sep 24, 2017 at 18:36

Mr Tsjolder from codidact

2,9161 gold badge29 silver badges46 bronze badges

Collectives™ on Stack Overflow

Create a Numpy scalar from dtype

3 Answers 3

1 Comment

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

1 Comment

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related