Wesley: > Firstly I would like to standardise the PCA, therefore I would like > each input to contribute equally to the result. As such I use the > correlation matrix as opposed to the covariance. By using this method > I do not need to centre the data, yes?
I think we have to make it clear that _data centering_ and _standardising_ and are two different things. # data centering: is about whether the variables should be shifted to be zero centered (copy-pasted from R's help for prcomp() function).# # standardising: //Ehmm... thinking...// if I got the PCA concept right, using the correlation matrix is a kind of normalising with which the variance of each input variable/feature/dimension is set to 1 just before the analysis is applied.# It's up to you to test data centering or not, and/or standardising or not. For several applications it has been shown that standardising improves results in several ways. (!?). > Secondly, using the by hand method r.covar -r -> m.eigensystem -> > r.mapcalculator, in particular when applying the eigen vectors to the > input imagery I can disregard the signs and take them as absolute > values, yes? No. My apologies for not being clear before. You can ignore the signs when you just try to interpret "how much" each original dimension has affected a component. When it comes to produce a component by hand (as you describe it above) you _certainly_ need to consider the signs. > Finally the size of these values indicates their relative contribution > to the component, so for eg. if band 1 has an eigen vecor of 0.8 and > band 2 has a value of 0.1, band 1 contributes more to the pc than band > 2, yes? Correct (given that you are talking about the _same_ component). > I will run some tests this afternoon and continue next week and report > back. Let me know if my knowledge above is correct. Let _me_ know, if you have the time, if my statements are correct! :-) Cheers, Nikos _______________________________________________ grass-user mailing list grass-user@lists.osgeo.org http://lists.osgeo.org/mailman/listinfo/grass-user