Add window function

[tomdl89]

I've had the lazy-only version of this function in quite a few codebases over the years, and recently took the time to write the transducer version. The cljs support is particularly annoying to write because cljs's array-list type doesn't implement a remove method.

This is like a simple alternative to https://github.com/cgrand/xforms/ 's window transducer, which needs f and invf functions to construct the windows. Plus this has a standard lazy-arity.

An example use case:

(def months [{:month "Jan" :income 500}
             {:month "Feb" :income 450}
             {:month "Mar" :income 580}
             {:month "Apr" :income 420}
             {:month "May" :income 680}])

(->> (window 3 months)
     (map (fn [months]
            {:month (:month (last months))
             :last-3-avg (/ (reduce (fn [acc x] (+ (:income x) acc)) 0 months)
                            (count months))})))
;; => ({:month "Jan", :last-3-avg 500}
;;     {:month "Feb", :last-3-avg 475}
;;     {:month "Mar", :last-3-avg 510}
;;     {:month "Apr", :last-3-avg 483}
;;     {:month "May", :last-3-avg 560})

[weavejester]

Thanks for the PR. My initial impression is that this is reasonable, but let me give it a some thought first.

[weavejester]

Should be 1.9.0 I believe.

[tomdl89]

Yes, sorry I assumed 1.9 had been cut. Will correct.

[weavejester]

Is using == over = due to performance reasons? What do the benchmarks come out with when comparing the two?

[tomdl89]

Yeah, performance hunch. Not particularly thought out. Will run in through criterium, but not expecting a meaningful difference.

[tomdl89]

(c/benchmark-round-robin [(dorun (window== 1000 (range 10000))) (dorun (window= 1000 (range 10000)))] c/*default-benchmark-opts*)
;; window== sample mean: 0.29017708853649266
;; window= sample mean:  0.3584622250301741

;; window== mean: 0.3225724956375 
;; window= mean:  0.3290516003666667

my stats isn't strong enough to be sure sample mean is a better indicator than mean, so i'll leave both means for you to look at. either a ~20% difference or a 2% difference. but both in favour of == over = so probably worth keeping, but I'm easy.

[tomdl89]

from my reading, sample mean will tend to emphasise the impact of outliers, thereby highlighting variance in one direction or another. so in this case, larger sample-mean draws attention to the fact a few of the test runs show significantly slower runs for =