topic Re: how to get dmodx and T2 for predictions in PLS in Discussions

how to get dmodx and T2 for predictions in PLS

mauritz — Sun, 21 Jan 2018 13:51:30 GMT

In the PLS-module it is straightforward to save the prediction formula and the standard error of prediction formula.

Is there also a way to save the DModX / DModY / T2 for the predictions, for additional outlier detection ?

Thanks !

Re: how to get dmodx and T2 for predictions in PLS

Bill_Worley — Mon, 22 Jan 2018 15:18:33 GMT

Look under the red hot spot for your model (NIPALS or SIMPLS) - go to Save Columns and save T Square and Distance. Distance will save DModX and DModY values.

Re: how to get dmodx and T2 for predictions in PLS

mauritz — Mon, 22 Jan 2018 17:31:05 GMT

Hello, thanks, but what I really need is the DModX and the T2 for the added rows of the unknowns. The columns for DModX and T2 do not get updated, unlike the columns for the predicted formula Y and predicted SE Y.
Kind regards !
Mk

Re: how to get dmodx and T2 for predictions in PLS

Bill_Worley — Mon, 22 Jan 2018 19:11:30 GMT

My apologies, I should have checked to see if those were saved as formula columns in the PLS platform.

You can get DModX and T square formulas from the Prinipal Components platform if that helps.

Re: how to get dmodx and T2 for predictions in PLS

Byron_JMP — Mon, 22 Jan 2018 20:47:42 GMT

Something like this?

Names Default To Here( 1 );
dt = Open( "$SAMPLE_DATA/Tiretread.jmp" );
//DModX formlula from the Principal Components Platform
obj1=dt << Principal Components(
      Y( :ABRASION, :MODULUS, :ELONG, :HARDNESS ),
      Estimation Method( "Default" ),"on Correlations");
obj1 << Save DMODX( 3 );
formula1= :DModX<<get formula();
 
//T Square formlula from the Multivariate Platform
obj2=dt<<Multivariate(
      Y( :ABRASION, :MODULUS, :ELONG, :HARDNESS ));
obj2 << T Square(1, Save T Square);

Re: how to get dmodx and T2 for predictions in PLS

mauritz — Sat, 27 Jan 2018 18:22:05 GMT

I would love to be able to do something similar for PLS : the PC’s of PCA will however not suffice : I really would like to see if there are outliers in PLS-factors for Dmodx en T2 !
Mauritz