numpy slicing that would always return an array

Question

Given a numpy array and a __getitem__-type index, is there an idiomatic way to get the corresponding slice of the array, that would always return an array and never a scalar?

Examples of valid indices include: an int, a slice, an ellipsis or a tuple of the above.

Say I have an array like this:

a = np.array([[1,2],[3,4]])

I am looking for an operation that would be equivalent to a[whatever] in all cases except when a[whatever] returns a scalar (for example, a[1,1]). In those cases I'd like this alternative operation to return a single-element array instead.

It might be worth clarifying that you want a 1D vector returned in place of a scalar, which I think is implied. (since a single-element array can have any number of dimensions). I have, in the past required that any slice of a 2D table always return a 2D table, in which case I used numpy.atleast_2d inside a table class that I wrote myself that mirrored many of array methods. It would probably break all kinds of stuff if you were able to get atleast_*D behavior out of a numpy array. — Paul
– Paul, Commented May 5, 2011 at 18:53
One possible use I can think of is when you always want the result to be a view, so that modifying it changes the original. If that is what is desired, atleast_Nd isn't good enough - probably need to convert the int indices to slices. — kelm
– kelm, Commented May 5, 2011 at 19:03

Joe Kington · Accepted Answer · 2011-05-05 18:24:30Z

6

If you just want to return a single-element array in cases where a scalar would otherwise be returned, why not just use numpy.atleast_1d on the result of the slice?

E.g.:

import numpy as np
x = np.arange(100).reshape(10,10)
print x[0,0]
print np.atleast_1d(x[0,0])
print np.atleast_1d(x[:,:3])

answered May 5, 2011 at 18:24

Joe Kington

287k73 gold badges621 silver badges474 bronze badges

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

3 Comments

NPE Over a year ago

Thanks for this. Is there a neat way to preserve the type of the array (I am using a subclass of ndarray and would like the result to be of the same type as x).

NPE Over a year ago

I should also note that I don't always know the type of x a priori.

Joe Kington Over a year ago

@aix - Unfortunately, there's no np.asanyarray equivalent for np.atleast_1d, as far as I know. @Sven's answer below is probably your best bet if you need to always preserve the type of the array. As it will always return a slice, the type should be preserved.

Sven Marnach · Accepted Answer · 2011-05-05 20:06:07Z

2

Here is a slightly more complex version that always returns a view into the original array (of course provided that you don't do any advanced indexing; this should be guaranteed by your specification of valid indices):

def get(a, item):
    if not isinstance(item, tuple):
        item = (item,)
    if len(item) == a.ndim and all(isinstance(x, int) for x in item):
        return a[item + (None,)]
    else:
        return a[item]

edited May 5, 2011 at 20:06

answered May 5, 2011 at 19:27

Sven Marnach

608k123 gold badges966 silver badges865 bronze badges

Comments

ktdrv · Accepted Answer · 2011-05-05 18:23:48Z

0

Apart from np.array(a[whatever])? Don't think there is a simpler/more idiomatic way than this.

answered May 5, 2011 at 18:23

ktdrv

3,6933 gold badges32 silver badges45 bronze badges

1 Comment

Joe Kington Over a year ago

Actually that doesn't do what the OP wants (e.g. have a look at the shape of the resulting array when the result of the slice is a scalar.)

Collectives™ on Stack Overflow

numpy slicing that would always return an array

3 Answers 3

3 Comments

Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

3 Comments

Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related