[GRASS-stats] Re: [GRASS-user] Testing i.pca (continued...)

[Roger Bivand]

On Wed, 29 Apr 2009, Nikos Alexandris wrote:

>
> Nikos:
>>   I tested... tested... & tested... (almost all tests below) and I just
>> can't understand WHY i.pca does NOT perform data centering when using
>> MODIS data (raw, rescaled at 0~255 or 0~254).
>
> The attached html file (exported from Tomboy note-taker) is (I suppose)
> easier to read :-)

Not easier to read, please simply post things on a website!


In your previous mail which you deleted here, you asked whether the code 
of i.pca shows if the input data is centred. Yes, it is centred, the 
function calc_mu(), as its name suggests, does this. The mean values are 
then passed to the function calculating the covariance matrix: 
calc_covariance(), where the means are subtracted from the input values of 
bands before accumulating the covariance matrix. Is is possible that where 
you do not see any effect of center=TRUE in prcomp that the means are all 
zero? You will find reading imagery/i.pca/main.c interesting and not hard.

Roger

>
> Cheers, Nikos
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no