data-science 2015-10-30 | Slack Archive

aaelony16:10:19

A friend of mine wanted to compare benchmark times for summing a 1e9 collection of ones, between R and Clojure, just for fun. In R, it is

x <- rep(1,1e9)
system.time( { s<- sum(x) })
R: 1.278 user, 0 system 1.279 elapsed

roughly. The best clojure equivalent I came up with so far was

(require '[hiphip.int :as hiphip]) 
(def x (let [v (vec (range 1e9))] (hiphip/amake [i 1e9] 1)))

user=> (time (hiphip/asum x))
"Elapsed time: 2854.957051 msecs"
1000000000

but perhaps someone knows an even better way. Would core.matrix be faster? We realize of course that benchmarks are often silly, but in the spirit of fun, can we think up a better way?

squest16:10:10

I’m not familiar with R, are those numbers in ms?

aaelony16:10:45

seconds (https://stat.ethz.ch/R-manual/R-devel/library/base/html/proc.time.html)

mikera18:10:37

with core.matrix:

mikera18:10:39

(set-current-implementation :vectorz)

mikera18:10:50

(def x (broadcast (scalar-array 1) [1e9]))

mikera18:10:47

(time (esum x)) "Elapsed time: 0.08469 msecs"