Re: how can I do Principal Components Analysis with octave?

[Paul Kienzle]

Up to a possible change in sign in some of the components, the
following, the following appears to be equivalent to princomp:

## Compute principal components of X
## [pc,z,w,Tsq] = princomp(X)
##   pc  the principal components
##   z   the transformed data
##   w   the eigenvalues of the covariance matrix
##   Tsq Hotelling's T^2 statistic for the transformed data
function [pc,z,w,Tsq] = princomp(X)
  C = cov(X);
  [U,D,pc] = svd(C);
  if nargout>1, z = center(X)*pc; end
  if nargout>2, w = diag(D); end
  if nargout>3, Tsq = sumsq(zscore(z),2); end

Do we want alternative PCA functions in octave-forge as well?
We could extend the princomp interface.  Some things that
seem useful to me are giving it a dimension argument, and
choosing a 'good' one if dimension is 0.  If nargout is zero
we could plot the data in the space of the first two principle
components, or a pareto chart of the eigenvalues.  Other
suggestions?

Thanks,

Paul Kienzle
address@hidden

On May 12, 2004, at 4:49 AM, Fredrik Lingvall wrote:

> Try this:
>
> function [pc,sv,n_sv]  = pca(x)
> % [pc,sv,n_sv]  = pca(x)
> %
> % Input:
> %   x - Data stored column-vise .
> %
> % Output:
> %  pc     - Principal components (eigenvectors of the covariance 
> matrix).
> %  sv     - Singular values.
> %  n_sv - Normalized singular values.
>
> C = cov(x);
> [U,D,pc] = svd(C);
> sv = diag(D);
> n_sv = 100*sv/sum(sv);
>
> \Fredrik
>
> > I believe you would need the programs doing the work. They are called
> > functions in Octave or Matlab, something like "PCA.m". I would be 
> > interested
> > myself so I checked using Google what MATLAB has:
> >
> > Here is a list of the functions with a short description of each:
> > PRINCOMP - principal components from raw data matrix
> > PCACOV - pca from covariance matrix
> > PCARES - residuals from pca
> > BARTTEST - Bartlett's test for dimensionality.
> >
> > Next I checked for the first two, namely "PRINCOMP" and "PCACOV" in
> > octave-forge but apparently neither is present. I guess we're out of 
> > luck
> > for the time being unless we have the capability to write the program.
> > Henry
> >
> >
> > on 5/11/04 1:04 AM, rino mailing at address@hidden wrote:
> >
> >
> > > I'd like to do Principal Components Analysis with octave
> > >
> > > What are the command I ave to write?
> > >
> > > How to plot the result?
> > >
> > >
> > >
> > > Thank you in advance for the time you spend to answer me, Mario.
>
>
>
> -------------------------------------------------------------
> Octave is freely available under the terms of the GNU GPL.
>
> Octave's home on the web:  http://www.octave.org
> How to fund new projects:  http://www.octave.org/funding.html
> Subscription information:  http://www.octave.org/archive.html
> -------------------------------------------------------------



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------