Rotate PCA's

BerettarBerettar Member Posts: 3 Contributor I
Hi there,

i am currently working on some stuff, where i want to reduce a given number of attributes to a much smaller set. So i started with a pca and a correlation matrix as visualization. So far so good, now i would like to rotate and load the factors with a derivate bigger 1 ( in my case around 20 factors ), so i am able to reduce my given attributes. Maybe i am just blind, but till now, i haven't found a proper setup to do this, any help is highly appreciated.

Greets :)

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,525   Unicorn
    Hi,
    what exactly do you mean by rotate? It seems to me, I'm not familiar with this. Where do the derivates come from?

    Greetings,
      Sebastian
  • BerettarBerettar Member Posts: 3 Contributor I
    Hi Sebastian,

    well this is a well known method ( in german "exploratorische Faktorenanalyse" ). First you start with a datamatrix (  here the dataset ) . Upon which you get a correlation matrix - correlation of each attributes. With the correlation matrix you get a factor charged  matrix. Within PCA ( in rapidminer ) i come up to this point ( a matrix, where all attributes are loaded on the selected factors ). But then you need to rotate the factor charged matrix, so you "load" a attribute as high as possible on one factor. I know that in SPSS there is a way to do this, but i don t have SPSS, so i tried rapidminer, but can t find the rotation of the factor charged matrix.

    Due to the fact that i am on work, i don t have my rapidminer setup visible, but  as far as i remember following setup should fit mine:

    ExcelExampleSet

    PCA

    Correlation Matrix ( Visualization )


    Greetings from Germany ;)

    p.s: derivates are given by PCA in rapidminer

    [edit] Rotation should be something like the VARIMAX or QUARTIMAX algorithm
  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,525   Unicorn
    Hi,
    currently neither the Varimax or Quartimax algorithm is available, but I added it on the agenda. We are going to enlarge the functionality in this area with the next versions.

    Greetings,
      Sebastian
Sign In or Register to comment.