(This article was first published on R snippets, and kindly contributed to R-bloggers)
Recently there is a lot of noise about Julia. I have decided to test its speed in simulation tasks on my toy Cont model. I thought I had vectorized my GNU R code pretty well, but Julia is much faster.The model was described in my earlier posts so let us go down to a comparison:
library(e1071)
cont.run <- function(burn.in, reps, n, d, l, s) {
tr <- rep(0, n)
r <- rep(0, reps)
for (i in 1:reps) {
sig <- rnorm(1, 0, d)
mul <- 1
if (sig < 0) {
sig <- -sig
mul <- -1
}
r[i] <- mul * sum(sig > tr) / (l * n)
tr[runif(n) < s] <- abs(r[i])
}
return(kurtosis(r[burn.in:reps]))
}
system.time(replicate(10,
cont.run(1000, 10000, 1000, 0.005, 10.0, 0.01)))It's execution time is a bit below 10 seconds on my laptop.
An equivalent Julia code is the following:
using Distributions
function cont_run(burn_in, reps, n, d, l, s)
tr = Array(Float64, n)
r = Array(Float64, reps)
for i in 1:reps
aris = 0
sig = randn() * d
mul = 1
if sig < 0
sig = -sig
mul = -1
end
for k in 1:n
if sig > tr[k]
aris += 1
end
end
ari = aris / (l * n)
r[i] = mul * ari
for j in 1:n
if rand() < s
tr[j] = ari
end
end
end
kurtosis(r[burn_in:reps])
end
n = 10
t_start = time()
k = Array(Float64, n)
for i in 1:n
k[i] = cont_run(1000, 10000, 1000, 0.005, 10.0, 0.01)
end
println(time() -t_start)And on my machine it takes a bit less than 0.7 seconds to run.
So we get over tenfold speedup. This is a significant difference for simulation experiments.
I will have to dig more into Julia in the future.
To leave a comment for the author, please follow the link and comment on his blog: R snippets.
R-bloggers.com offers daily e-mail updates about R news and tutorials on topics such as: visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more...