PLS model develpment and predict

[Jonison]

Hello, I am building a PLS model using training dataset, and also have a dataset to do prediction (prediction dataset).

The training dataset is finalized and then the model configuration has been fixed.

The prediction dataset is updated everyday, and this is reason why the datasets are managed separately.

The issue is how can I code PLS procedure for this task.

The model code I am using is:

PROC PLS CV=ONE DATA=&EM_IMPORT_DATA METHOD=PLS NOCENTER NOSCALE NFAC=6 MISSING=AVG;
 MODEL &Y1 = &X1 &X2 &X3 &X4 &X5 &X6 &X7 &X8 &X9 &X10 &X11 &X12 &X13;
QUIT;

It is a complicated model and &Y1 have five variables, similar setting for &X1 and etc (using let to define &X1 ....&Y1).

The prediction dataset has same structure, but how can set the prediction for prediction dataset?  (prefer not to merge the training dataset and prediction dataset)

Many thanks

[Rick_SAS · Accepted solution]

Some regression procedures support the STORE statement, which creates an item store for future scoring by using the PLM procedure.

Unfortunately, the PLM procedure does not support the STORE statement, so I think your only choice is to use the missing response trick, which unfortunately requires concatenating the new data with the original. The Getting Started example in the PLS doc has an example of scoring.

[Jonison]

Thanks for your kind reply. I think this would be the solution to current version.