When sampling from a multi-variate truncated normal using rtmvnorm from
the tmvnorm-package, I experience extreme performance differences between
two of my computers. On my laptop computer, draws take ~5s, on my desktop
~30s. I need to run MCMCs with repeat calls to rtmvnorm on my desktop. The
code in both cases is exactly the same. rtmvnorm seems not to have any
machine dependent defaults built in.
The function call is:
rtmvnorm(1, mean =as.numeric(A.i %*% (X%*%b.c[r,])), sigma = solve(t(A)
%*% A),
lower = as.numeric(0^Y * -Inf^(1-Y)), upper=
as.numeric(0^(1-Y)*Inf^(Y)),
algorithm = "gibbs", burn.in.samples=ys.burn,
start.value
= ys.old)
The laptop computer is:
Dell Latitude E4300 with Intel Core2 P9600 @ 2.53GHz, 4GB of RAM (not sure
what type), Windows 7 Professional, 64 bit.
The desktop:
Dell Precision T3500 with Intel Xeon W3565 @ 3.20Ghz, 6GB DDR3 RAM,
Windows 7 Professional, 64 bit.
Many thanks for any hints.
Martin Steinwand
[[alternative HTML version deleted]]
