I am trying H2O AutoML for multi-class classification. Everything seems to be working fine except when I am trying to extract the variable importance from metalearner for the StackedEnsemble.
This is what I am getting from code:
# Get the "BestOfFamily" Stacked Ensemble model
se.best <- h2o.getModel(grep("StackedEnsemble_BestOfFamily_AutoML", model_ids, value = TRUE)[1])
#se.best
metalearner <- h2o.getModel(se.best@model$metalearner$name)
#as.data.frame(h2o.varimp(metalearner))
h2o.varimp_plot(metalearner, num_of_features = 20)
Variable importance from metalearner for "StackedEnsemble_BestOfFamily_AutoML":
The problem here is instead of showing the 20 model names that H2O AutoML picked (from "max_models =20") the above plot shows the names of the classes of the multi-class classification.
To me, this looks like a bug because I do not have this problem when I tried regression with H2O AutoML in R.
