Sliding window as deque. Yet another implementation of sliding (rolling) window iterator in Python. This is based on answers of http://stackoverflow.com/questions/6822725/rolling-or-sliding-window-iterator-in-python. It is improved with "overlapping" and two window types: soft and strict.

sliding_window.py

# -*- coding: utf8 -*-

from __future__ import absolute_import, division, print_function, unicode_literals

import collections
import types
import itertools


class SlidingWindow(collections.deque):

    DEFAULT_WINDOW_SIZE = 2

    @classmethod
    def sliding_windows(cls, sequence, window_size=None, overlap_size=None, yield_tail=None, strict_windows=None):
        """
        Generates a sequence of sliding windows over the `sequence` (second) parameter.
        It expects that sequence is `iterable` and returns a `generator object`.
        By default the size of the each window is 2,
        but you can specify it with `window_size` (third) parameter.
        Each window implements `collections.deque` interface.

        How can you use it?
        Let define initial sequence and function for uncovering generator content to tuple list.

            >>> sequence = xrange(8)
            >>>
            >>> list(sequence)
            [0, 1, 2, 3, 4, 5, 6, 7]
            >>>
            >>> def sliding_windows(*args, **kwargs):
            ...     return list(tuple(sw) for sw in SlidingWindow.sliding_windows(*args, **kwargs))
            ...
            >>>

        Default behaviour: Number of input itens equals to number of output ones.
        This is not right sliding windows, but it is very usefull for many applications.

            >>> default = sliding_windows(sequence)
            >>> default
            [(0,), (0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7)]
            >>>
            >>> len(sequence) == len(default)
            True
            >>>
            >>> soft_sw_list_2 = sliding_windows(sequence, window_size=2)
            >>> soft_sw_list_2
            [(0,), (0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7)]
            >>>
            >>> len(sequence) == len(soft_sw_list_2)
            True
            >>>
            >>> soft_sw_list_3 = sliding_windows(sequence, window_size=3)
            >>> soft_sw_list_3
            [(0,), (0, 1), (0, 1, 2), (1, 2, 3), (2, 3, 4), (3, 4, 5), (4, 5, 6), (5, 6, 7)]
            >>>
            >>> len(sequence) == len(soft_sw_list_3)
            True
            >>>
            >>> soft_sw_list_4 = sliding_windows(sequence, window_size=4)
            >>> soft_sw_list_4
            [(0,), (0, 1), (0, 1, 2), (0, 1, 2, 3), (1, 2, 3, 4), (2, 3, 4, 5), (3, 4, 5, 6), (4, 5, 6, 7)]
            >>>
            >>> len(sequence) == len(soft_sw_list_4)
            True
            >>>

        It implements more accurate behaviour with `strict_windows` parameter.

            >>> strict_sw_list_2 = sliding_windows(
            ...     sequence,
            ...     window_size=2,
            ...     strict_windows=True
            ... )
            ...
            >>> strict_sw_list_2
            [(0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7)]
            >>>
            >>> len(sequence)
            8
            >>> len(strict_sw_list_2)
            7
            >>>
            >>> strict_sw_list_3 = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     strict_windows=True
            ... )
            >>> strict_sw_list_3
            [(0, 1, 2), (1, 2, 3), (2, 3, 4), (3, 4, 5), (4, 5, 6), (5, 6, 7)]
            >>>
            >>> len(sequence)
            8
            >>> len(strict_sw_list_3)
            6
            >>>
            >>> strict_sw_list_4 = sliding_windows(
            ...     sequence,
            ...     window_size=4,
            ...     strict_windows=True
            ... )
            >>> strict_sw_list_4
            [(0, 1, 2, 3), (1, 2, 3, 4), (2, 3, 4, 5), (3, 4, 5, 6), (4, 5, 6, 7)]
            >>>
            >>> len(sequence)
            8
            >>> len(strict_sw_list_4)
            5
            >>>

        Also you can overlap windows.
        By default windows overlapped with (window_size - 1) elements.

            >>> strict_lapped_default = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     strict_windows=True
            ... )
            ...
            >>> strict_lapped_default
            [(0, 1, 2), (1, 2, 3), (2, 3, 4), (3, 4, 5), (4, 5, 6), (5, 6, 7)]
            >>>
            >>> strict_lapped_3_2 = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=2,
            ...     strict_windows=True
            ... )
            ...
            >>> strict_lapped_3_2
            [(0, 1, 2), (1, 2, 3), (2, 3, 4), (3, 4, 5), (4, 5, 6), (5, 6, 7)]
            >>>
            >>> strict_lapped_3_1 = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=1,
            ...     strict_windows=True
            ... )
            ...
            >>> strict_lapped_3_1
            [(0, 1, 2), (2, 3, 4), (4, 5, 6)]
            >>>

        The same you can do with «soft» sliding windows.
        Note that overlapping do not work on the incomplete windows
        (at the start of window sequence). It needs
        to accumulate the full window before applying overlapping.

            >>> soft_lapped_3_1 = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=1,
            ...     strict_windows=False
            ... )
            ...
            >>> soft_lapped_3_1
            [(0,), (0, 1), (0, 1, 2), (2, 3, 4), (4, 5, 6)]
            >>>

        Look at the end of the `soft_lapped_3_1`. There is no `7`.
        It has happened because 7 — the last member of initial sequence
        and it do not fit with overlapping scheme.
        For this case you can use `yield_tail` parameter.
        There are two ways to generate tail of [(0, 1, 2), (2, 3, 4), (4, 5, 6)]
        One is
            [(0, 1, 2), (2, 3, 4), (4, 5, 6), (5, 6, 7)]
        And other is
            [(0, 1, 2), (2, 3, 4), (4, 5, 6), (7,)]
        We decide that costant size of window is more important.

            >>> strict_lapped_3_1_tail = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=1,
            ...     yield_tail=True,
            ...     strict_windows=True
            ... )
            ...
            >>> strict_lapped_3_1_tail
            [(0, 1, 2), (2, 3, 4), (4, 5, 6), (5, 6, 7)]
            >>>

        The same you can do with «soft» sliding windows.

            >>> strict_lapped_3_1_tail = sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=1,
            ...     yield_tail=True,
            ...     strict_windows=False
            ... )
            ...
            >>> strict_lapped_3_1_tail
            [(0,), (0, 1), (0, 1, 2), (2, 3, 4), (4, 5, 6), (5, 6, 7)]
            >>>

        This is not recommended case of use. It is quite difficult to understand the logic
        of the last example looking only on the result.

        If you set `overlap_size` to zero you will get partitions of initial sequence.
        This is degenerate case of sliding windows.

            >>> sliding_windows(
            ...     sequence,
            ...     window_size=2,
            ...     overlap_size=0,
            ...     strict_windows=True
            ... )
            [(0, 1), (2, 3), (4, 5), (6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=0,
            ...     strict_windows=True
            ... )
            [(0, 1, 2), (3, 4, 5)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=4,
            ...     overlap_size=0,
            ...     strict_windows=True
            ... )
            [(0, 1, 2, 3), (4, 5, 6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence, window_size=5,
            ...     overlap_size=0,
            ...     strict_windows=True
            ... )
            [(0, 1, 2, 3, 4)]
            >>>

        As you guess, you can use `yield_tail` in this case too.

            >>> sliding_windows(
            ...     sequence,
            ...     window_size=2,
            ...     overlap_size=0,
            ...     yield_tail=True,
            ...     strict_windows=True
            ... )
            [(0, 1), (2, 3), (4, 5), (6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=0,
            ...     yield_tail=True,
            ...     strict_windows=True
            ... )
            [(0, 1, 2), (3, 4, 5), (5, 6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=4,
            ...     overlap_size=0,
            ...     yield_tail=True,
            ...     strict_windows=True
            ... )
            [(0, 1, 2, 3), (4, 5, 6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence, window_size=5,
            ...     overlap_size=0,
            ...     yield_tail=True,
            ...     strict_windows=True
            ... )
            [(0, 1, 2, 3, 4), (3, 4, 5, 6, 7)]
            >>>

        With «soft sliding windows», partitions are also «soft».
        As it was shown before, it needs to accumulate the full partition item by item.

            >>> sliding_windows(
            ...     sequence,
            ...     window_size=2,
            ...     overlap_size=0,
            ...     strict_windows=False
            ... )
            [(0,), (0, 1), (2, 3), (4, 5), (6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=3,
            ...     overlap_size=0,
            ...     strict_windows=False
            ... )
            [(0,), (0, 1), (0, 1, 2), (3, 4, 5)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=4,
            ...     overlap_size=0,
            ...     strict_windows=False
            ... )
            [(0,), (0, 1), (0, 1, 2), (0, 1, 2, 3), (4, 5, 6, 7)]
            >>>
            >>> sliding_windows(
            ...     sequence,
            ...     window_size=5,
            ...     overlap_size=0,
            ...     strict_windows=False
            ... )
            [(0,), (0, 1), (0, 1, 2), (0, 1, 2, 3), (0, 1, 2, 3, 4)]
            >>>

        If you deal with finite sequences and set the window size greater than sequence size
        you will get some strange effects. It is also not recommended case of use.

            >>> finite_sequence = (0, 1, 2, 3)
            >>>
            >>> sliding_windows(
            ...     finite_sequence,
            ...     window_size=10,
            ...     strict_windows=False
            ... )
            [(0,), (0, 1), (0, 1, 2), (0, 1, 2, 3)]
            >>>
            >>> sliding_windows(
            ...     finite_sequence,
            ...     window_size=10,
            ...     strict_windows=True
            ... )
            []
            >>>
            >>> sliding_windows(
            ...     finite_sequence,
            ...     window_size=10,
            ...     strict_windows=True,
            ...     yield_tail=True,
            ... )
            [(0, 1, 2, 3)]
            >>>

        For fool-tolerance you cannot set `window_size` and `overlap_size` to the same number,
        because it doesn't make sense.

            >>> sliding_windows(sequence, window_size=3, overlap_size=3)
            Traceback (most recent call last):
                ...
            AssertionError: it does not make sense: overlap greater then window
            >>>

        In this case you should get:
            [(0,), (0, 1), (0, 1, 2), (0, 1, 2), (0, 1, 2), (0, 1, 2), ... an infinite number of times ... ]
        You can achieve this much more simpler without sliding windows =).
        You will get the same with overlap size greater then window size.

            >>> sliding_windows(sequence, window_size=3, overlap_size=100500)
            Traceback (most recent call last):
                ...
            AssertionError: it does not make sense: overlap greater then window
            >>>

        The size of window should be a positive number.

            >>> sliding_windows(sequence, window_size=0)
            ...
            Traceback (most recent call last):
                ...
            AssertionError: it does not make sense: window with zero or negative size
            >>>

        The overlap size should be be a positive number or null.

            >>> sliding_windows(sequence, overlap_size=-1)
            ...
            Traceback (most recent call last):
                ...
            AssertionError: it does not make sense: overlap with negative size
            >>>

        Also it controls types of arguments:

            >>> sliding_windows(42)
            Traceback (most recent call last):
                ...
            AssertionError: sequence is iterable
            >>>
            >>> sliding_windows(sequence, 2.1)
            Traceback (most recent call last):
                ...
            AssertionError: window_size is positive integer or None (for default)
            >>>
            >>> sliding_windows(sequence, 2, 1.1)
            Traceback (most recent call last):
                ...
            AssertionError: overlap_size is integer or None (for default)
            >>>
            >>> sliding_windows(sequence, 2, yield_tail=0)
            Traceback (most recent call last):
                ...
            AssertionError: yield_tail is boolean or None (for default)
            >>>
            >>> sliding_windows(sequence, 2, strict_windows=0)
            Traceback (most recent call last):
                ...
            AssertionError: strict_windows is boolean or None (for default)
            >>>

        Remember: these assertion errors will not work within python optimisation.

        To sum up all examples, this is formal specification.

            :param collections.Iterable sequence:
                initial sequence of any element.
            :param integer window_size:
                size of each sliding window;
                window_size have to be greater than zero.
            :param integer overlap_size: number
                number of overlapping items in sliding windows;
                by default overlap_size = (window_size - 1)
            :param boolean strict_windows:
                flag to form sliding windows with the same size each other;
                otherwise at first it generates windows during accumulating items;
                by default `strict_windows` is False.
            :param boolean yield_tail:
                flag to generate the rest of sequence
                that do not match to overlapping scheme.
                by default `yield_tail` is False.

            :returns: <generator object sliding_windows at ... >
                Do not forget about it. If you want to use the result of this functions several times
                you should apply itertools.tee or convert this generator to concrete objects.

        """
        assert isinstance(sequence, collections.Iterable), \
            'sequence is iterable'
        assert isinstance(window_size, (int, long, types.NoneType)), \
            'window_size is positive integer or None (for default)'
        assert isinstance(overlap_size, (int, long, types.NoneType)), \
            'overlap_size is integer or None (for default)'
        assert isinstance(yield_tail, (bool, types.NoneType)), \
            'yield_tail is boolean or None (for default)'
        assert isinstance(strict_windows, (bool, types.NoneType)), \
            'strict_windows is boolean or None (for default)'

        if window_size is None:
            window_size = cls.DEFAULT_WINDOW_SIZE
        if overlap_size is None:
            overlap_size = window_size - 1

        assert window_size > 0, \
            'it does not make sense: window with zero or negative size'
        assert 0 <= overlap_size, \
            'it does not make sense: overlap with negative size'
        assert overlap_size < window_size, \
            'it does not make sense: overlap greater then window'

        win = SlidingWindow([], maxlen=window_size)
        append = win.append
        yield_condition = None

        skip_window_limit = window_size - overlap_size
        skip_window_counter = 0

        for index, item in enumerate(sequence):
            skip_window_counter += 1
            append(item)
            skip_window_condition = (skip_window_counter < skip_window_limit)
            head_condition = index < window_size
            yield_condition = head_condition or (not skip_window_condition)
            if strict_windows:
                head_condition = index >= (window_size - 1)
                yield_condition = head_condition and (not skip_window_condition)
            if yield_condition:
                yield win
                skip_window_counter = 0
        else:
            if yield_tail and not yield_condition:
                yield win

    def __getitem__(self, index):
        if isinstance(index, slice):
            return type(self)(itertools.islice(self, index.start, index.stop, index.step))
        return super(SlidingWindow, self).__getitem__(index)

if __name__ == '__main__':

    sequence = xrange(17)

    def sliding_windows(*args, **kwargs):
        return list(tuple(sw) for sw in SlidingWindow.sliding_windows(*args, **kwargs))

    print (
        sliding_windows(sequence, window_size=100, overlap_size=0, strict_windows=False)
    )

See also https://github.com/w495/python-video-shot-detector/blob/master/shot_detector/utils/collections/sliding_window.py

Patch for Python 3 compatibility: diff -u sliding_windows.py sliding_windows_py3.py

--- sliding_windows.py	2018-05-04 07:19:53.000000000 -0600
+++ sliding_windows_py3.py	2018-05-04 07:22:21.000000000 -0600
@@ -388,15 +388,16 @@
                 Do not forget about it. If you want to use the result of this functions several times
                 you should apply itertools.tee or convert this generator to concrete objects.
         """
+        NoneType = type(None)
         assert isinstance(sequence, collections.Iterable), \
             'sequence is iterable'
-        assert isinstance(window_size, (int, long, types.NoneType)), \
+        assert isinstance(window_size, (int, NoneType)), \
             'window_size is positive integer or None (for default)'
-        assert isinstance(overlap_size, (int, long, types.NoneType)), \
+        assert isinstance(overlap_size, (int, NoneType)), \
             'overlap_size is integer or None (for default)'
-        assert isinstance(yield_tail, (bool, types.NoneType)), \
+        assert isinstance(yield_tail, (bool, NoneType)), \
             'yield_tail is boolean or None (for default)'
-        assert isinstance(strict_windows, (bool, types.NoneType)), \
+        assert isinstance(strict_windows, (bool, NoneType)), \
             'strict_windows is boolean or None (for default)'

         if window_size is None:
@@ -441,7 +442,7 @@

 if __name__ == '__main__':

-    sequence = xrange(17)
+    sequence = range(17)

     def sliding_windows(*args, **kwargs):
         return list(tuple(sw) for sw in SlidingWindow.sliding_windows(*args, **kwargs))

This gist is quite useful. Could you please add a license statement at the top, so that we know if we're allowed to use it?