R help - Mar 2009 - Accuracy of R and other platforms

Hi,

The paper

@ARTICLE{AlmironSilvaMM:2009,
  author = {Almiron, M. and Almeida, E. S. and Miranda, M.},
  title = {The Reliability of Statistical Functions in Four Software Packages
	Freely used in Numerical Computation},
  journal = {Brazilian Journal of Probability and Statistics},
  year = {in press},
  volume = {Special Issue on Statistical Image and Signal Processing},
  url = {http://www.imstat.org/bjps/}}

is freely available under the "Future Papers" link. It makes a nice
comparison of the numerical properties of R, Ox, Octave and Python.

Alejandro

-- 
Alejandro C. Frery
Macei?, AL - Brazil
http://acfrery.googlepages.com
http://www.researcherid.com/rid/A-8855-2008



Samuel Beckett  - "Birth was the death of him."

Alejandro C. Frery:
> @ARTICLE{AlmironSilvaMM:2009,
> author = {Almiron, M. and Almeida, E. S. and Miranda, M.},
> title = {The Reliability of Statistical Functions in Four Software
> Packages Freely used in Numerical Computation},
> journal = {Brazilian Journal of Probability and Statistics},
> year = {in press},
> volume = {Special Issue on Statistical Image and Signal Processing},
> url = {http://www.imstat.org/bjps/}}
> 
> is freely available under the "Future Papers" link. It makes a
nice
> comparison of the numerical properties of R, Ox, Octave and Python.
Thanks for posting this. I?m happy to see that the results for R were 
generally excellent, and almost always better than for the three other 
software packages.

But there were a few cases where R did not turn out to be the winner.
Rather surprising that Ox was better than R for computing the 
autocorrelation coefficient for two of the datasets, given its terrible 
results for the standard deviation. Anybody have any ideas why?

-- 
Karl Ove Hufthammer

These comparisons are very simplistic.  In most contexts, it would  
make much better sense to measure "accuracy" in standard error units,
rather than in number of digits.

There doubtless are specialist applications where the 15th digit (or  
even the 10th!) are important.  But the check of accuracy really has  
then to be tuned to that application.  Results in cases where the  
"accuracy" beyond (optimistically) the sixth digit is of no  
consequence are unlikely to give much clue on performance in such  
specialist applications.

John Maindonald             email: john.maindonald@anu.edu.au
phone : +61 2 (6125)3473    fax  : +61 2(6125)5549
Centre for Mathematics & Its Applications, Room 1194,
John Dedman Mathematical Sciences Building (Building 27)
Australian National University, Canberra ACT 0200.


On 23/03/2009, at 10:00 PM, r-help-request@r-project.org wrote:
> From: Karl Ove Hufthammer <Karl.Hufthammer@math.uib.no>
> Date: 23 March 2009 2:58:23 AM
> To: r-help@stat.math.ethz.ch
> Subject: Re: [R] Accuracy of R and other platforms
>
>
> Alejandro C. Frery:
>
>> @ARTICLE{AlmironSilvaMM:2009,
>> author = {Almiron, M. and Almeida, E. S. and Miranda, M.},
>> title = {The Reliability of Statistical Functions in Four Software
>> Packages Freely used in Numerical Computation},
>> journal = {Brazilian Journal of Probability and Statistics},
>> year = {in press},
>> volume = {Special Issue on Statistical Image and Signal Processing},
>> url = {http://www.imstat.org/bjps/}}
>>
>> is freely available under the "Future Papers" link. It makes
a nice
>> comparison of the numerical properties of R, Ox, Octave and Python.
>
> Thanks for posting this. I抦 happy to see that the results for R were
> generally excellent, and almost always better than for the three other
> software packages.
>
> But there were a few cases where R did not turn out to be the winner.
> Rather surprising that Ox was better than R for computing the
> autocorrelation coefficient for two of the datasets, given its  
> terrible
> results for the standard deviation. Anybody have any ideas why?
>
> -- 
> Karl Ove Hufthammer

	[[alternative HTML version deleted]]