Speed ​​It Up Without Rcpp?

Question

I want to speed up the following algorithm. I provide functions to the xts time series, and then I want to analyze the main components for each time point at the previous X points (I am using 500 now), and then use the results of this PCA (5 main components in the following code) to calculate some value. Something like that:

lookback <- 500 for(i in (lookback+1):nrow(x)) { x.now <- x[(i-lookback):i] x.prcomp <- prcomp(x.now) ans[i] <- (some R code on x.prcomp) } 

I suppose this would require me to replicate the reverse lookup rows as columns so that x something like cbind(x,lag(x),lag(x,k=2),lag(x,k=3)...lag(x,k=lookback)) and then ran prcomp on every line? It seems expensive though. Perhaps some apply option? I'm ready to take a peek at Rcpp, but I wanted you to do this before, guys.

Edit: Thanks for the answers. Information about my dataset / algorithm:

• dim (x.xts) currently = 2000x24. But in the end, if it means a promise, he will have to work fast (I will give several data sets).
• func (x.xts) takes ~ 70 seconds. These are 2000-500 prcomp calls with 1500 500x24 frame creation.

I tried using Rprof to find out which one is the most expensive part, but the first time I use Rprof , so I need more experience with this tool to get clear results (thanks for the suggestion).

I think that I will first try to flip this into a loop of type _apply, and then look at parallelization.

Answer 1

On my quad-core desktop, if this does not finish in a reasonable amount of time, I would run the piece using something in the lines (not tested):

 library(snowfall) sfInit(parallel = TRUE, cpus = 4, type = "SOCK") lookback <- 500 sfExport(list = c("lookback", "x")) sfLibrary(xts) output.object <- sfSapply(x = (lookback+1):nrow(x), fun = function(i, my.object = x, lb = lookback) { x.now <- my.object[(i-lb):i] x.prcomp <- prcomp(x.now) ans <- ("some R code on x.prcomp") return(ans) }, simplify = FALSE) # or maybe it TRUE? depends on what ans is