PROC GLMSELECT - Partitioning by a variable

oggylang — Wed, 16 Jun 2021 17:20:01 GMT

Hi SAS Forum,

I am currently doing some LASSO regression, and have a headache, that I hope someone else has had before, and therefore might be able to sort out. I am doing a LASSO regression, and I want to partition my data. I have made a dummy variable indicating whether the data belongs to training or testing part of the data set, but I am having struggles implementing this into the partition statement. The dummy is treated. The SAS documentation is a bit limited and mainly focuses on partitioning by choosing a share of the data rather than choosing based on a variable.

proc glmselect data = forecastmerge2;
class herkomst;
model PostEmplSumD = Woman Married PriorEmpl c:/
selection = lasso(stop=none choose=cvex);
partition = treated(test=(treated=1) train=(treated=0));
output out=GLMOut p = p_hat;
run;

If you have any questions, please let me know. I tried to add all of the coding which seemed relevant for the question.

Oggylang

Re: PROC GLMSELECT - Partitioning by a variable

Rick_SAS — Fri, 18 Jun 2021 13:19:48 GMT

Here is the documentation that specifies the correct syntax. I think for your data the PARTITION statement would look like this (untested)

partition ROLEVAR=treated(test='1' train='0');

You can also include a character variable named _ROLE_ in the input data that has the values "TRAIN" and "TEST". If the input data contains a _ROLE_ variable, then you can omit the PARTITION statement.

topic Re: PROC GLMSELECT - Partitioning by a variable in Statistical Procedures

PROC GLMSELECT - Partitioning by a variable

Re: PROC GLMSELECT - Partitioning by a variable