[R] Row-by-row regression on matrix

[Sundar Dorai-Raj]

murali.menon at uk.abnamro.com said the following on 9/19/2007 9:50 AM:
> Folks,
> 
> I have a 3000 x 4 matrix (y), which I need to regress row-by-row against a 
> 4-vector (x) to create a
> matrix lm.y of intercepts and slopes. To illustrate:
> 
> y <- matrix(rnorm(12000), ncol = 4)
> x <- c(1/12, 3/12, 6/12, 1)
> 
> system.time(lm.y <- t(apply(y, 1, function(z) lm(z ~ x)$coefficient)))
> [1] 44.72 18.00 69.52    NA    NA
> 
> Takes more than a minute to do (and I need to do many similar regressions 
> a day). 
> 
> Is there a more efficient way of handling this?
> 
> I'm running R 2.4.1 on Windows XP Service Pack 2 on a Intel Xeon dual-core 
> 2.66GHz with 3GB RAM.
> 
> Thanks very much,
> 
> Murali
> 

Yes. Try,

set.seed(1)
y <- matrix(rnorm(12000), ncol = 4)
x <- c(1/12, 3/12, 6/12, 1)
system.time(lm.y <- t(apply(y, 1, function(z) lm(z ~ x)$coefficient)))

##   user  system elapsed
##   1.10    0.00   38.69

system.time(lm.y2 <- t(coef(lm(t(y) ~ x))))

##   user  system elapsed
##   0.02    0.00    0.04

all.equal(lm.y, lm.y2)
## [1] TRUE

HTH,

--sundar