[GRASS-user] Calculating eigen values and %
varianceexplainedafter PCA analysis
nikos.alexandris at felis.uni-freiburg.de
Sun Mar 1 08:13:07 EST 2009
> > * Present first the variance (=eigenvalues) because it's the
> >first thing you will look at to know "how much variance of the
> >original data is _expressed_ in each new component.
> > * The importance, since it refers to the eigenvalue, it's better
> >to come right after it.
> to me it picks your eye more quickly if it is not buried in the
> shrug. the important thing is that the numbers are correct & not
Yes, the important thing are the numbers. A clear output is also more
"functional", if I may say so.
> > * Present the loadings (eigenvectors) for each new
> > component.
> we are doing that already, right?
Absolutely. It's just the structure of the output what remains. I have
no objections to whatever will be decided as long as the _numbers_ are
there. Nonetheless, from a user's perspective, I presented my ideas
about the output.
> > * Column-wise or row-wise? The results can be either
> > presented column-wise, that is one column for each new component
> > _or_ row-wise, as they are currently printed. I think row-wise just
> > looks better :-)
> maybe, but row-wise is slightly easier to code.
For the interpretation I think the way that the output will look like,
is just a matter to get used to it. In fact, I think row-wise is easier
than column-wise. Anyway, this is of minor importance.
> > "Some" examples... (only 2 for column-wise and
> > all the rest row-wise... playing around).
> fancy tables are hard for the module output because it uses
> and G_message() condenses any whitespace (multiple spaces, tabs,..) to
> single space. thus formatting is lost.
> and i.pca's main output is maps, not eigen data so I guess it makes
> to keep that text optional instead of sending to stdout. Perhaps a New
> flag to print summary report to stdout? (mmph, just cut&paste from
> for map history it's a bit better, but I can't end with a %.
> now 'r.info -h' output looks like:
> Eigen values, (vectors), and [percent importance]:
> PC1 1170.12 ( -0.63 -0.65 -0.43 ) [ 88.07% ]
> PC2 152.49 ( 0.23 0.37 -0.90 ) [ 11.48% ]
> PC3 6.01 ( 0.75 -0.66 -0.08 ) [ 0.45% ]
> module output is same but not as pretty due to G_message() issue.
Well, whatever is practical and achievable. I would be happy to see more
suggestions upon this. But I like it now that it prints out the
Best regards, Nikos
More information about the grass-user