PROC PLS - Error Invalid group number

[klroesner]

Hi folks, what does the error message mean? With a smaller data set, it works fine. 

NOTE: FC2._PLS18DAT 17M datasets, <100 class-values
57         ods graphics on;
58         proc pls data=FC2._pls18dat nfac=3
59                   plot=(VIP ParmProfiles XLoadingProfiles);
60              class VGP TAET9_NIVEAU WOHNBL;
61              model DB2_WSG_KK = VGP TAET9_NIVEAU WOHNBL;
62         run;

ERROR:  UNABLE TO ACCESS MESSAGE 27648.20771
ERROR: Invalid group number
ERROR:  UNABLE TO ACCESS MESSAGE 29255.294
ERROR: Invalid group number
… * 100
ERROR:  UNABLE TO ACCESS MESSAGE 33789.26305
ERROR: Invalid group number
ERROR: Unable to restore '3864455' from template store!
NOTE: The SAS System stopped processing this step because of errors.

[PaigeMiller]

Hi folks, what does the error message mean? With a smaller data set, it works fine.

Explain exactly what you mean by this. What smaller data set, what are the sizes of the data set and how many variables in the model when it works fine. What larger data set, what are the sizes of the data set and how many variables in the model when it doesn't work fine. How many class variables in the model when it does work? How many class variables in the model when it doesn't work?

[klroesner]

Hi Paige,

Thanks for the quick response. The question originally comes from a colleague of mine and here is his answer:

Class Vars:

• VGP:                      char5, 12 Levels
• WOHNBL:             char3, 16 Levels
• TAET9_NIVEAU:    char1, 3 Levels

The small data set on which it works has 1 million records and on which it does not work 17 million records.

[PaigeMiller]

Followup question

Are there more variables when you run the larger data set, or the same number of variables?

[klroesner]

It is actually only the number of records that are different. All other parameters are the same:

• same number of variables,
• same code and
• same data structures.

[PaigeMiller]

So I have not seen this error before, but obviously it is related to the number of records in your data set. I would discuss this with SAS tech support.

[klroesner]

Okay, Paige .. I'm going to discuss this with Tech Support. Thank you for your help!

[Reeza]

Is it possible one of the values one of the CLASS variables is not in the smaller data set? Is every category in the smaller data set?

PLS shouldn't be using a ton of memory except maybe for graphs so I'd try turning off the graph portion as well to see if that helps. Hopefully Tech Support has a better answer than me 🙂

[klroesner]

Hi Reeza, You are on the right track: It's a graphics problem. Without graphics output, there is no error message. Each of the three graphics options (VIP ParmProfiles XLoadingProfiles) leads to the same error message.

[Rick_SAS]

The procedure us trying to access an error message but cannot find it. Presumably, there is no error for the smaller data and so the procedure does not need to access the error.

Please give tech support your version of SAS by submitting

%put &SYSVLONG4;

and copy/paste from the SAS log.

[klroesner · Accepted solution]

We got a workaround from the SAS TechSupport: In SAS9.4, it is likely that the 17 million observations overtax the observation scatterplot in PLS's correlation loadings plot, which is created by default. With plot(only)= this plot can be switched off.

         proc pls data=FC2._pls18dat nfac=3   plot(only)=(VIP ParmProfiles XLoadingProfiles);

              class VGP TAET9_NIVEAU WOHNBL;
              model DB2_WSG_KK = VGP TAET9_NIVEAU WOHNBL;
         run;