[R] make R program faster

[Duncan Murdoch]

On 13-03-28 7:07 AM, Christof Kluß wrote:
> Hi
>
> there are some good tips in "The R Inferno"
> http://www.burns-stat.com/documents/books/the-r-inferno/
> or connect C++ to R with Rcpp
> http://dirk.eddelbuettel.com/code/rcpp.html
> or byte code compiler (library(compiler))
> or library(data.table)
>
> but do you have an idea to fasten standard R source code, with the
> following Rprof output
>
>                   self.time self.pct total.time total.pct
> [<-.data.frame       1.34    29.13       1.78     38.70
> [.data.frame         0.26     5.65       1.02     22.17
> [[                   0.12     2.61       0.44      9.57
> NextMethod           0.12     2.61       0.12      2.61
> match                0.10     2.17       0.16      3.48
> <Anonymous>          0.10     2.17       0.10      2.17
> c                    0.10     2.17       0.10      2.17
> [[.data.frame        0.08     1.74       0.32      6.96
> [.Date               0.06     1.30       0.10      2.17
> FUN                  0.06     1.30       0.10      2.17
> [<-                  0.04     0.87       1.82     39.57
> [                    0.04     0.87       1.04     22.61
> [<-.Date             0.04     0.87       0.18      3.91
> vapply               0.04     0.87       0.14      3.04
> %in%                 0.02     0.43       0.18      3.91
> +                    0.02     0.43       0.10      2.17
>
>
> It comes from a simulation algorithmus that calculates day wise values
> (values are depenend from the output of the day before). First I create
> a data.frame with NAs. Finally each row contains the daily values.
>
> output <- as.data.frame(matrix(nrow = 365, ncol = 50))
> for (day in (1:365)) {
>     ...
>     r <- list(Date=d,daylength=daylength,TempSum=tempsum, ...)
>     output[day,] <- r
> }
>
> Is there an better (faster) way to do such things in R?

Generally matrices are much faster than dataframes.  Your code spends 
about 35-60% of its time in dataframe indexing methods, so you might be 
able to double the speed by switching to matrices.

The disadvantage is that matrices can only hold one type, so you may 
need multiple matrices to replace one dataframe, and that will make your 
code harder to read.

Duncan Murdoch