Interpretation of L1 and L2 values in GLM options

[xloffree]

Hi,

What exactly do the L1 and L2 values refer to in the GLM model? Is it the lambda value or the lambda value multiplied by the alpha (or 1-alpha)? I got a result where the elastic net model selected zero variables, which was a very unexpected result. Consequently, I am suspecting that I am misinterpreting what the L1 and L2 values are referring to in PiML.

Thank you

[CnBDM-Su]

Hi @xloffree , With different L1 and L2, the GLM model will be changed.

• {l1_regularzation=0 & l2_regularzation=0}: Ordinary least squares Linear Regression.
• {l1_regularzation>0 & l2_regularzation=0}: Lasso Regression.
• {l1_regularzation=0 & l2_regularzation>0}: Ridge Regression.
• {l1_regularzation>0 & l2_regularzation>0}: Elasticnet Regression.

For Lasso, alpha= l1, For Ridge, alpha=l2, For Elasticnet, alpha=l1+l2; l1_ratio= l1/ alpha

[xloffree]

Hi,

Thank you for the response.

When using PiML, if I have a lambda value of 0.1 and an alpha value of 0.5, what value should I enter in for L1 and L2 when using elastic net? Would it be 0.05?

Thank you