h2o GLM with regularization

Question

I am using the h2o package in R to fit a GLM via the h2o.glm() fucntion. One reasonable way to assess feature importance in a GLM with the l1 regularization penalty is to monitor the order that parameters enter the linear predictor (i.e. the model) as the l1 penalty weight decreases over each successive lambda. I cannot find in the h2o documentation if it possible to extract this information from a returned model object.

Does anyone know if it is possible to view the fitted model form after each successive lambda?

Thanks,

Answer 1

you can get the full regularization path with h2o.getGLMFullRegularizationPath(my_glm) where my_glm is the glm you trained, just remember to set lambda_search equal to TRUE (i.e. my_glm = h2o.glm(x,y,training_frame, lambda_search = TRUE)