Using Josh's nice example, with data.table's built-in 'by' (optimised grouping) yields a 6 times speedup (100 seconds down to 15 on my netbook).
> system.time(all.2b <- lapply(si, function(.indx) { coef(lm(y ~ + x, data=d[.indx,])) })) user system elapsed 144.501 0.300 145.525 > system.time(all.2c <- lapply(si, function(.indx) { minimal.lm(y + = d[.indx, y], x = d[.indx, list(int, x)]) })) user system elapsed 100.819 0.084 101.552 > system.time(all.2d <- d[,minimal.lm2(y=y, x=cbind(int, x)),by=key]) user system elapsed 15.269 0.012 15.323 # 6 times faster > head(all.2c) $`1` coef se x1 0.5152438 0.6277254 x2 0.5621320 0.5754560 $`2` coef se x1 0.2228235 0.312918 x2 0.3312261 0.261529 $`3` coef se x1 -0.1972439 0.4674000 x2 -0.1674313 0.4479957 $`4` coef se x1 -0.13915746 0.2729158 x2 -0.03409833 0.2212416 $`5` coef se x1 0.007969786 0.2389103 x2 -0.083776526 0.2046823 $`6` coef se x1 -0.58576454 0.5677619 x2 -0.07249539 0.5009013 > head(all.2d) key coef V2 [1,] 1 0.5152438 0.6277254 [2,] 1 0.5621320 0.5754560 [3,] 2 0.2228235 0.3129180 [4,] 2 0.3312261 0.2615290 [5,] 3 -0.1972439 0.4674000 [6,] 3 -0.1674313 0.4479957 > minimal.lm2 # slightly modified version of Josh's function(y, x) { obj <- lm.fit(x = x, y = y) resvar <- sum(obj$residuals^2)/obj$df.residual p <- obj$rank R <- .Call("La_chol2inv", x = obj$qr$qr[1L:p, 1L:p, drop = FALSE], size = p, PACKAGE = "base") m <- min(dim(R)) d <- c(R)[1L + 0L:(m - 1L) * (dim(R)[1L] + 1L)] se <- sqrt(d * resvar) list(coef = obj$coefficients, se) } > -- View this message in context: http://r.789695.n4.nabble.com/SLOW-split-function-tp3892349p3900851.html Sent from the R help mailing list archive at Nabble.com. ______________________________________________ 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.