Construct an advanced slice from string in Python/Numpy

Question

Suppose I have a CLI and would like to construct a slice from the CLI, where the CLI user just inputs a range string.

r = '1:4:2'
axis = '2'

And I have a way of slicing this as follows:

arr = np.zeros((10,10,10))

sliced_arr = eval("arr["+','.join([':',':',r])+"]")

Question: is there a better way to do this?

@norok2 yeah, in my full code it is used. I didn't bother because this is the minimal representation. It just is there to imply that it is part of a bigger function. — Chris
– Chris, Commented Dec 5, 2019 at 23:08

Divakar · Accepted Answer · 2019-12-05 20:45:22Z

2

Here's one using r and axis as input arguments -

indexer = [slice(None)]*arr.ndim
indexer[int(axis)] = slice(*map(int,r.split(':')))
out = arr[tuple(indexer)]

Generalizing to handle to any generic indexing string notation in r, it would be -

ax = int(axis)
indexer = [slice(None)]*arr.ndim
o = np.array([0,arr.shape[ax],1])
v = np.array(r.split(':'))
v = np.r_[v,['']*(len(o)-len(v))]
indexer[ax] = slice(*np.where(v=='',o,v).astype(int))
out = arr[tuple(indexer)]

edited Dec 5, 2019 at 20:45

answered Dec 5, 2019 at 17:11

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

2 Comments

Chris Over a year ago

one reason I went with eval is in case someone did something clever like "::-1": the eval will always work exactly like brackets. I did look at slice, but it seemed like it would not handle this case well. Is that wrong?

Divakar Over a year ago

@donlan Edited to handle generic string notations in r.

Dani Mesejo · Accepted Answer · 2019-12-05 17:20:01Z

1

You could build a slice object from the string, and then use numpy.s_:

import numpy as np

r = '1:4:2'
arr = np.zeros((10,10,10))

s = slice(*map(int, r.split(":")))
print(s)

sliced_arr = arr[np.s_[:, :, s]]

Output

slice(1, 4, 2)

edited Dec 5, 2019 at 17:20

answered Dec 5, 2019 at 17:08

Dani Mesejo

62.2k6 gold badges56 silver badges86 bronze badges

3 Comments

Mad Physicist Over a year ago

I think OP wants to use that slice only for the one axis

norok2 Over a year ago

Perhaps, but the string passed to eval() is, indeed, 'arr[:,:,1:4:2]' which would not be strictly equivalent to this solution.

Dani Mesejo Over a year ago

@norok2 You are right, modified the answer to match the output of the eval()

norok2 · Accepted Answer · 2019-12-06 16:04:57Z

0

You can use something like this, which also supports negative indexs:

import numpy as np


def str2slicing(
        slice_str: str,
        index: int,
        n_dim: int):
    assert(-n_dim <= index < n_dim)
    index %= n_dim
    return tuple(
        slice(*map(int, slice_str.split(':'))) if i == index else slice(None)
        for i in range(n_dim))


arr = np.random.randint(1, 100, (10, 10, 10))
sliced_arr = arr[str2slicing('1:4:2', int('2'), arr.ndim)]
print(sliced_arr.shape)
# (10, 10, 2)

(Note that this is pretty much an extension / variation of the other answers: the juice is pretty much this slice(*map(int, slice_str.split(':')))).

edited Dec 6, 2019 at 16:04

answered Dec 5, 2019 at 17:26

norok2

27.1k6 gold badges83 silver badges110 bronze badges

Collectives™ on Stack Overflow

Construct an advanced slice from string in Python/Numpy

3 Answers 3

2 Comments

3 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

2 Comments

3 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related