[GRASS-stats] Re: [GRASS-user] Testing i.pca ~ prcomp(), m.eigensystem ~ princomp()

[Edzer Pebesma]

Nikos Alexandris wrote:
> Edzer:
>   
> I do worry about the wiki that states results are "similar" between R and
>> grass, when the differences appear e.g. in the third or second digit of a
>> loading (eigenvector value). Clearly something is is going on, here.
>>     
>
> So you think it's really important? Isn't it a matter of pixel-inclusion
> in the various algorithms/programs?
>   
Well, it's just much larger than I could imaging numerical dispersion
would give; if you mean by pixel-inclusion different methods how missing
values are treated, then I can understand it a bit better, given that
there are substantial amounts of them. The wiki might mention this
possibility. (Although I don't see why grass wouldn't ignore partially
missing valued pixels)

Note that I didn't run the examples, I was just surprised. When I
compare different software implementations of an algorithm, I usually
don't stop worrying until differences happen beyond digit 5.

-- 
Edzer Pebesma
Institute for Geoinformatics (ifgi), University of Münster
Weseler Straße 253, 48151 Münster, Germany. Phone: +49 251
8333081, Fax: +49 251 8339763 http://ifgi.uni-muenster.de/
http://www.springer.com/978-0-387-78170-9 e.pebesma at wwu.de