[R] R binaries, platform independent and Design of Experiments

Nam-Ky Nguyen nkn at turing.une.edu.au
Sun Sep 4 01:49:21 CEST 2005

> On 02/09/05, Nam-Ky Nguyen <nkn at turing.une.edu.au> wrote:
>> > b) You do NOT want to do numerical computations on software available
>> in
>> > Java byte code.
>> You do not want to do heavy numerical computations with R either. Most
>> statistical calculation using R requires a fraction of a second and I
>> cannot see a real difference between say 0.05 second and 0.07 second.
>> NKN.
> It is my understanding that the problem with Java is that it wasn't
> written with serious numerical computation in mind - as far as I know
> only in the latest version have Sun started to be address this issue.
> The byte code for the java virtual machine has a flawed numerical
> model which is not fully compliant with the IEEE754 standard - this
> has nothing to do with speed of computation.  Furthermore the integer
> model is very restrictive when you want to work on random numbers
> using bit-twiddling.

I am not sure that this is an appropriate platform to compare C++ (or C#)
and Java. Besides, there are already several articles on this topic on the
web. I like the BYOL idea. Like BYOB (Bring Your Own Beer), BYO Language
will allow you to use the language of your choice for software development
(Please see http://www.mono-project.com/Languages/).

I take this oppotunity to thank Berwin, Brian, Peter and Uwe for their
helpful emails.

More information about the R-help mailing list