Generalising itertools.pairwise

"itertools.pairwise is an iterable from the standard module itertools that lets you access overlapping pairs of consecutive elements of the input iterable. That's quite a mouthful, so let me translate: You give pairwise an iterable, like "ABCD", and pairwise gives you pairs back, like ("A", "B"), ("B", "C"), and ("C", "D"). In loops, it is common to unpack the pairs directly to perform some operation on both values."

"If you had to implement pairwise, you might think of something like the code below: def my_pairwise(iterable): for prev_, next_ in zip(iterable, iterable[1:]): yield (prev_, next_) Which directly translates to def my_pairwise(iterable): yield from zip(iterable, iterable[1:]) But there is a problem with this implementation, and that is the slicing operation. pairwise is supposed to work with any iterable and not all iterables are sliceable. For example, files are iterables but are not sliceable."

"pairwise is supposed to work with any iterable and not all iterables are sliceable. For example, files are iterables but are not sliceable. There are a couple of different ways to fix this but my favourite uses collections.deque with its parameter maxlen: from collections import deque from itertools import islice def my_pairwise(data): data = iter(data) window = deque(islice(data, 1), maxlen=2) for value in data: window.append(value) yield tuple(window)"

itertools.pairwise returns overlapping consecutive pairs from any iterable, producing (a,b), (b,c), ... from input. Common usage unpacks pairs to compute operations such as differences between consecutive balances. A naive implementation using zip(iterable, iterable[1:]) or yield from zip(...) relies on slicing and therefore fails for non-sliceable iterables like file objects. A robust implementation converts the input to an iterator and uses collections.deque with islice to build a sliding window of length two, appending new values and yielding tuple(window) for each step. The deque+islice approach generalises pairwise to work with all iterables without slicing.

Read at Mathspp