transducer_bench.clj

(def dataset (interleave (range 10000000) (range)))

(defn workflow [ds]
  (->> ds
       (dedupe)
       (map #(* % %))
       (filter #(= 0 (rem % 111)))
       (take 10)))

(with-progress-reporting (bench (workflow dataset)))

;;           Execution time mean : 173.265424 ns
;;  Execution time std-deviation : 6.851563 ns
;; Execution time lower quantile : 164.763159 ns ( 2.5%)
;; Execution time upper quantile : 187.295385 ns (97.5%)
;;                 Overhead used : 1.964161 ns

(def workflow2
  (comp
   (dedupe)
   (map #(* % %))
   (filter #(= 0 (rem % 111)))
   (take 10)))

(with-progress-reporting (bench (sequence workflow2 dataset)))

;; Evaluation count : 273778620 in 60 samples of 4562977 calls.
;;              Execution time mean : 215.319869 ns
;;     Execution time std-deviation : 3.715614 ns
;;    Execution time lower quantile : 209.969636 ns ( 2.5%)
;;    Execution time upper quantile : 221.657620 ns (97.5%)
;;                    Overhead used : 1.964161 ns

;; Updated version, transducer is faster in this test.
;; It seems that only when the number of items great than a threshold, is transducer the better choice.


(def dataset (interleave (range 10000) (range)))

(defn workflow [ds]
  (->> ds
       (dedupe)
       (map #(* % %))
       (filter #(= 0 (rem % 111)))
       (reduce +)))

(with-progress-reporting (bench (workflow dataset)))

;; Evaluation count : 34260 in 60 samples of 571 calls.
;;              Execution time mean : 1.794328 ms
;;     Execution time std-deviation : 20.238946 µs
;;    Execution time lower quantile : 1.761130 ms ( 2.5%)
;;    Execution time upper quantile : 1.833225 ms (97.5%)
;;                    Overhead used : 1.964161 ns

(def workflow2
  (comp
   (dedupe)
   (map #(* % %))
   (filter #(= 0 (rem % 111)))))

(with-progress-reporting (bench (transduce workflow2 + dataset)))

;; Evaluation count : 51900 in 60 samples of 865 calls.
;;              Execution time mean : 1.143402 ms
;;     Execution time std-deviation : 24.978019 µs
;;    Execution time lower quantile : 1.117554 ms ( 2.5%)
;;    Execution time upper quantile : 1.215618 ms (97.5%)
;;                    Overhead used : 1.964161 ns