Thread Subject: Difference between processpca and princomp

Hi all,

I'm using PCA to reduce the dimensionality of my data to train a SOM. To check I was processing the data correctly, I thought I'd check if processpca and princomp give the same data in PCA space. Thing is they don't, anyone know why? As a test I set up a simple matrix:


>> matrix = [1 4 8; 10 5 8; 18 4 8];

Then used princomp and printed out the data in PC space:


>> [pcs, pc_data, latents] = princomp(matrix);

>> pc_data =

   -8.6674 0.3133 0
    0.3349 -0.6659 0
    8.3325 0.3526 0

Then used processpca and printed out the data in PC space:

>> [pcs2, pca_settings] = processpca(matrix);

>> pcs2 =

  -20.1263 -6.9170 -12.7358
   -4.4646 2.9594 5.4482
   -0.0000 -0.6307 0.3426

I would expect both to be the same, any ideas on why they aren't?

On Sep 12, 7:52 am, "John " <re...@epsam.keele.ac.uk> wrote:
> Hi all,
>
> I'm using PCA to reduce the dimensionality of my data to train a SOM.  To check I was processing the data correctly, I thought I'd check if processpca and princomp give the same data in PCA space.  Thing is they don't, anyone know why?  As a test I set up a simple matrix:
>
> >> matrix = [1 4 8; 10 5 8; 18 4 8];
>
> Then used princomp and printed out the data in PC space:
>
> >> [pcs, pc_data, latents] = princomp(matrix);
> >> pc_data =
>
>    -8.6674    0.3133         0
>     0.3349   -0.6659         0
>     8.3325    0.3526         0
>
> Then used processpca and printed out the data in PC space:
>
> >> [pcs2, pca_settings] = processpca(matrix);
> >> pcs2 =
>
>   -20.1263   -6.9170  -12.7358
>    -4.4646    2.9594    5.4482
>    -0.0000   -0.6307    0.3426
>
> I would expect both to be the same, any ideas on why they aren't?

Read the documentation closely. Does one of them assume the
data has zero mean?

Hope this helps.

Greg