On 5 September 2010 05:28, William Stein <wst...@gmail.com> wrote: > scipy.stats has random number generators for about a 100 different > families of distributions. The last time I tried them, they were of > variable quality, in that some of them were (very, very) slow. But > the range of distributions was really impressive. You should look > at the official book about numpy (yes, numpy), which Travis Oliphant > wrote. It has a pretty good list of the distributions in numpy.stats > = scipy.stats.
> Whatever you do, I hope you'll benchmark carefully as you go. A > package that generates random numbers 1000 times slower than MATLAB is > going to be very annoying to use. Non-cryptography people usually > generate random numbers because they want a lot of them. If they are of variable quality, I would suggest that not only benchmarking them is important, but testing them for randomness is too. I know that some implementations that work on one word-size computer perform really poor if the word size is increased. So an algorithm only tested on a 32-bit system might perform very badly in a 64-bit one. Most make use of a CPUs register overflowing at some point. If the word size changes, so the overflow point changes, so the algorithm has changes. Dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
