Return equivalent of `:` from function for indexing array

Question

I have a large array and a function that returns index lists into the array, i.e.,

import numpy

n = 500
a = numpy.random.rand(n)

def get_idx(k):
    # M         


        

Answer 1

Have a look at slice

def get_x():
    return slice(2)

a=list(range(100))
a[get_x()]

will return [0, 1]

UPDATE

And for your need get_x function should be

def get_x(k, n):
    return slice(n if k > 6 else k)

Update

as @Eric correctly noted it's better to pass None instead of n. So function would be:

def get_x(k):
    return slice(None if k > 6 else k)

Answer 2

NumPy has a helper np.s_[] which can be used to construct slice and Ellipsis objects:

def get_idx(k):
    return np.s_[:] if k > 6 else np.s_[:k]

    # or even np.s_[:None if k > 6 else k]

In general, a[np.s_[ <stuff> ]] is exactly the same as a[ <stuff> ].

Answer 3

You can use Ellipsis (translates to '...')

np.identity(2)[Ellipsis]
# array([[1.0, 0.0], [0.0, 1.0]])

Edit:

I feel I should mention one caveat: If you are sure this will only be used on 1d arrays, it's a perfectly fine and simple solution. However, in a public interface I wouldn't recommend it because it doesn't generalise correctly if a user uses the output of getidx for indexing a multi dimensional array.

For example A[getidx(k), ...] won't be safe anymore and the semantics of A[getidx(k), someslice] will be inconsistent if A has three or more dims.

That said, slice(None) isn't perfect in this respect either, for exammple A[getidx(k), getidx(k)] will behave inconsistently.

arange(n) looks like it's always giving the correct behaviour, but of course it's wasteful in most cases.