Aster GLM Error

Aster
Teradata Employee

Aster GLM Error

Hello,

I am running GLM more than one input variable it gives me following error.

 

SELECT * FROM GLM (
ON (SELECT 1)
PARTITION BY 1
InputTable ('scd_voc_csi.csi_dataset_flagged_churned_1000')
OutputTable ('scd_voc_csi.csi_dataset_flagged2_glm_1000')
ColumnNames ('churn','repeat_complaints','network_complaints')
Family ('LOGISTIC')
Link ('LOGIT')
Weight ('1')
Threshold ('0.01')
MaxIterNum ('10'));

 

Executed as Single statement. Failed [34 : HY000] [AsterData][ASTERJDBCDSII](34) ERROR: SQL-MR function GLM failed: [AsterData][ASTERJDBCDSII](34) ERROR: SQL-MR function GLM_REDUCESOLVEANDUPDATE failed: The input data results in a singular matrix and hence there is no solution () ()
Elapsed time = 00:00:00.000

STATEMENT 1: Select Statement failed.

 

If I remove 'network_complaints' then for one input variable it works. Can you please help with above error like where i am doing mistake.

 

Thanks

Abdulrazak

 

Tags (1)
3 REPLIES
Teradata Employee

Re: Aster GLM Error

Teradata Employee

Re: Aster GLM Error

Abdulrazak,

 

That error message means that some of the input predictor columns (repeat_complaints, network_complaints) are not independent of each other. You can confirm this with the correlation function. A singular matrix (which arrises when independent variables are not independent of each other) means that the matrix cannot be inverted which is necessary for regression.

 

Since you only have numeric variables I reccomend using LARS/LASSO instead. These functions handle dependent predictor variables automatically. If you were to stick with GLM the solution would be to use only one of the complaints columns. 

 

Michelle

Teradata Employee

Re: Aster GLM Error

Thanks Michelle for your help. I am also applying the same approach as they are highly correlated. Thanks Abdulrazak