I've combined the PCA Compute, PCA Apply, and PCA Inversion nodes every which way I can think of - but I can't get reconstructed data that looks anything like my original data. What's the right way to do this?

Your Input Data Set needs to be connected to PCA Compute and PCA Apply. The model (green) output of the PCA compute is connected to the PCA Apply, too. Check your data in the PCA Apply output - it should have many of the "PCA dimension x" columns (as many as you have numeric input columns unless you change the config parameters).

As the last step you need to connect the PCA Compute model also to the PCA Inversion node, and you also need to connect the output of the PCA Apply to the PCA Inversion (as this is the data that you want to recover). In the config dialog of the PCA Inversion, select all the "PCA dimension x" columns. The output will then contain the original input data (represented as a set of new columns with the (#1) suffix).

Hope that makes sense...

Thanks, Bernd! The part I was missing was to include ONLY the PCA dimensions in the Inversion node config. I was also including the original data (which the Apply node sent along because I didn't check "Replace original data columns" in its config) and getting nonsense.

Hey ,

I also tried the same Workflow, but I got everytime the same error, between the PCA Compute and the PCA Apply Node. Do you have any idea, what is happend?

Thank you

ERROR PCA Apply 2:6 Unable to clone input data at port 1 (Transformation model): Java heap space