Re: [Gretl-devel] simulation speed

I discovered that the results of my previous post today were obtained NOT with 32-bit R but with 64-bit R. I was running the script within the R-bundle of the editor TextMate; it uses 64-bit R and I don't know how to change that. Sorry about this. The correct 32-bit results are RNGmethod ctype result user system elapsed 1 Inversion loop 0.05118006 2.081 0.012 2.093 2 Inversion iteration 0.05154388 1.930 0.226 2.157 3 Kinderman-Ramage loop 0.05292662 1.503 0.047 1.549 4 Kinderman-Ramage iteration 0.05189927 1.334 0.228 1.562 5 Box-Muller loop 0.05266636 1.750 0.024 1.773 6 Box-Muller iteration 0.05138073 1.610 0.226 1.836 7 Ahrens-Dieter loop 0.05259550 1.461 0.025 1.486 8 Ahrens-Dieter iteration 0.05209534 1.298 0.229 1.526 This is more in line with my post of the 22nd december. It also clearly show that the 64-bit version of R is quite a bit quicker in this case. There is a package on CRAN: SuppDists which provides the ziggurat method for generating from a normal distribution. Warning: it only runs correctly in a 32-bit environment. Results: RNGmethod ctype result user system elapsed 1 ziggurat loop 0.05240394 0.837 0.014 0.851 2 ziggurat iteration 0.05200224 0.541 0.362 0.902 Faster indeed. But the system overhead in the "iteration" (or vectorized) version is rather large. Berend