The most visited post ever on the ‘Og was In{s}a(ne), my report on Radford Neal’s experiments with speeding up R by using different brackets (the second most populat was Ross Ihaka’s comments, “simply start over and build something better”). I just spotted two new entries by Radford on his blog that are bound to rekindle the debate about the speed of R. The latest one shows that matrix multiplication can be made close to ten time faster by changing the way testing for the presence of NaN’s in a matrix is operated. This gain is not as shocking as producing a 25% improvement when replacing x=1/(1+x) with x=1/{1+x}, but a factor 10 is such a major gain…
Archive for Radford Neal
News about speeding R up
Posted in R, Running, Statistics with tags matrix multiplication, NaN, R, Radford Neal, Ross Ihaka, speed on May 24, 2011 by xi'anIn{s}a(ne)!!
Posted in R, Statistics with tags brackets, R, Radford Neal, speed, system.time on September 6, 2010 by xi'anHaving missed the earliest entry by Radford last month, due to disconnection in Yosemite, I was stunned to read his three entries of the past month about R performances being significantly modify when changing brackets with curly brackets! I (obviously!) checked on my own machine and found indeed the changes in system.time uncovered by Radford… The worst is that I have a tendency to use redundant brackets to separate entities in long expressions and thus make them more readable (I know, this is debatable!), so each new parenthesis add to the wasted time… Note that it is the same with curly bracket: any extra curly bracket takes some additional computing time…
> f=function(n) for (i in 1:n) x=1/(1+x)
> g=function(n) for (i in 1:n) x=(1/(1+x))
> h=function(n) for (i in 1:n) x=(1+x)^(-1)
> j=function(n) for (i in 1:n) x={1/{1+x}}
> k=function(n) for (i in 1:n) x=1/{1+x}
> x=1
> system.time(f(10^6))
user system elapsed
1.684 0.020 1.705
> system.time(g(10^6))
user system elapsed
1.964 0.012 1.976
> system.time(h(10^6))
user system elapsed
2.452 0.016 2.470
> system.time(j(10^6))
user system elapsed
1.716 0.008 1.725
> system.time(k(10^6))
user system elapsed
1.532 0.016 1.548
In the latest of his posts, Radford lists a series of 14 patches that speed up R up to 25%… That R can face such absurdities is pretty annoying! Given that I am currently fighting testing an ABC algorithm with MCMC inner runs, which takes forever to run, this is even depressing!!! As Radford stresses, “slow speed is a significant impediment to greater use of R, much more so than lack of some wish list of new features.”
