Here is a way -- you are dividing by zero in the scaling: > x sp wg n v.realtime v.cputime v.tcputime v.idletime v.nswtch 9 0 1 1 28.61300 28.61 28.6039 0.0000000 407 1563 0 1 2 15.20270 30.38 28.5981 0.9805230 483 3128 0 1 4 12.50930 50.00 28.6053 10.7877000 489 4682 0 1 6 12.10260 72.55 28.6193 22.0203000 488 6241 0 1 8 12.11510 96.80 28.6448 34.1126000 494 121 0 10 1 28.61530 28.60 28.6056 0.0000000 1996 1684 0 10 2 14.33530 28.66 28.6003 0.0296437 1908 3241 0 10 4 7.27129 29.05 28.6002 0.2313020 2110 4801 0 10 6 4.91522 29.42 28.6002 0.4259310 2134 6367 0 10 8 3.79337 30.25 28.6032 0.8424120 2245 > x.s <- split(x, x[,1:2]) > x.result <- do.call(rbind, lapply(x.s, function(.sec){ + # create matrix of data to scale + .mat <- as.matrix(.sec[, -(1:3)]) + # need to transpose due to recycle of arguments + .mat <- t(t(.mat) / .mat[1,]) + .sec[, -(1:3)] <- as.data.frame(.mat) + .sec + })) > x.result sp wg n v.realtime v.cputime v.tcputime v.idletime v.nswtch 0.1.9 0 1 1 1.0000000 1.000000 1.0000000 NaN 1.0000000 0.1.1563 0 1 2 0.5313214 1.061866 0.9997972 Inf 1.1867322 0.1.3128 0 1 4 0.4371894 1.747641 1.0000489 Inf 1.2014742 0.1.4682 0 1 6 0.4229756 2.535827 1.0005384 Inf 1.1990172 0.1.6241 0 1 8 0.4234124 3.383432 1.0014299 Inf 1.2137592 0.10.121 0 10 1 1.0000000 1.000000 1.0000000 NaN 1.0000000 0.10.1684 0 10 2 0.5009663 1.002098 0.9998147 Inf 0.9559118 0.10.3241 0 10 4 0.2541050 1.015734 0.9998112 Inf 1.0571142 0.10.4801 0 10 6 0.1717689 1.028671 0.9998112 Inf 1.0691383 0.10.6367 0 10 8 0.1325644 1.057692 0.9999161 Inf 1.1247495 > > On Mon, May 4, 2009 at 3:41 PM, Zeljko Vrba wrote: > I have a data-set that is structured as follows: > > sp wg n v.realtime v.cputime v.tcputime v.idletime v.nswtch > 9 0 1 1 28.61300 28.61 28.6039 0.00000e+00 407 > 1563 0 1 2 15.20270 30.38 28.5981 9.80523e-01 483 > 3128 0 1 4 12.50930 50.00 28.6053 1.07877e+01 489 > 4682 0 1 6 12.10260 72.55 28.6193 2.20203e+01 488 > 6241 0 1 8 12.11510 96.80 28.6448 3.41126e+01 494 > 121 0 10 1 28.61530 28.60 28.6056 0.00000e+00 1996 > 1684 0 10 2 14.33530 28.66 28.6003 2.96437e-02 1908 > 3241 0 10 4 7.27129 29.05 28.6002 2.31302e-01 2110 > 4801 0 10 6 4.91522 29.42 28.6002 4.25931e-01 2134 > 6367 0 10 8 3.79337 30.25 28.6032 8.42412e-01 2245 > > [And a lot of other variables, and sp also changes] > > Now, I want to divide each of the v.* variables for every (sp, wg, n) > combination by the values in the row defined by (sp, wg, 1), thus giving > a relative change to the case n==1 for the same combination of sp and wg. > I have written the following function, but I'm wondering whether there's > an easier/more efficient way.. Perhaps by organizing the data structure > differently, and if so -- how? > > speedup.1 <- function(df) > { > groups <- split(df, df[,c("sp","wg")]) > ret <- list() > > for(g in groups) { > ref <- g[1,] > for(v in 4:15) # sp,wg,n are factors, so / doesn't work.. > g[[v]] <- ref[[v]] / g[[v]] > ret <- rbind(ret, g) > } > ret > } > > ______________________________________________ > R-help@r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. > -- Jim Holtman Cincinnati, OH +1 513 646 9390 What is the problem that you are trying to solve? [[alternative HTML version deleted]]