How do I create a sliding window with a 50% overlap with a numpy array?

Question

Say I have an array like so:

array([1, 2, 3, 4, 5, 5, 7, 8, 9, 10])

I want to change it to

array([1, 2, 3, 4]), array([3, 4, 5, 6]), array([5, 6, 7, 8]), array([7, 8, 9, 10])

Here the window size would be 4, and step size would be 2.

How can I do this in a Python function?

Looks like basic numpy-usage. What did you try? You obviously need to create new arrays, so did you try a list-comprehension, where the start-index is controlley by python's range() with step? — sascha, Aug 17 '17 at 08:53
Do you want four separate arrays rather than a single 4x4 array? The latter could be more more convenient and is just an application of [`as_strided`](https://docs.scipy.org/doc/numpy/reference/generated/numpy.lib.stride_tricks.as_strided.html). — Alex Riley, Aug 17 '17 at 09:00
For some more general example of as_strided for this use-case: [skimage's view_as_windows](https://github.com/scikit-image/scikit-image/blob/master/skimage/util/shape.py#L107). — sascha, Aug 17 '17 at 09:04

Daniel F · Accepted Answer · 2017-08-17T09:22:35.227

15

You want numpy.lib.stride_tricks.as_strided

def window(a, w = 4, o = 2, copy = False):
    sh = (a.size - w + 1, w)
    st = a.strides * 2
    view = np.lib.stride_tricks.as_strided(a, strides = st, shape = sh)[0::o]
    if copy:
        return view.copy()
    else:
        return view

Out[]: 
array([[ 1,  2,  3,  4],
       [ 3,  4,  5,  6],
       [ 5,  6,  7,  8],
       [ 7,  8,  9, 10]])

Be sure to set copy = True it if you want to write to the windowed array, as otherwise it is a memory-sharing view.

edited Aug 17 '17 at 09:22

answered Aug 17 '17 at 09:04

Daniel F

13,620
2
29
55

Where would I add `copy()`? – Aug 17 '17 at 09:08
Edited answer to clarify, also fixed the offsets since that only worked for your toy problem before – Daniel F Aug 17 '17 at 09:23
`Expected type 'Optional[Iterable[int]]`, got 'int' instead` for `strides=st` – Aug 17 '17 at 09:26
What was your input `a`? – Daniel F Aug 17 '17 at 09:30
I inputed `a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])` – Aug 17 '17 at 09:32
Works on my end. Did you copy the whole function? – Daniel F Aug 17 '17 at 09:35
I copied the whole function. It's just a warning though so it may work. – Aug 17 '17 at 09:38
Never mind, it works! Thanks! – Aug 17 '17 at 09:40
don't forget to mark the question answered then – Daniel F Aug 17 '17 at 09:42
So, for `a` I'm passing in a `DataFrame.values` and the shape of the `Data Frame` is (540, 3). I get the error: `ValueError: mismatch in length of strides and shape` How do I resolve this? – Aug 17 '17 at 13:44
Thanks a lot @DanielF for you suggestions. – Ranjan Pal May 02 '21 at 16:06

How do I create a sliding window with a 50% overlap with a numpy array?

1 Answers1