I have a stacked learner where the output layer is a regr.ranger with params list(rf.quantreg = TRUE, rf.keep.inbag = TRUE). Is it possible to predict quantiles with GraphLearners like this?
I know that for a pure lrn('regr.ranger'), once trained, I can simply reference the ranger model directly and use that for quantile prediction:
predict(my_learner$model, data = my_test_data, type = "quantiles", quantiles = c(0.025, 0.975))
But for the stacked learner, I have other learners mediating between the features and regr.ranger, so it seems to me that I have to go via mlr3.
My GraphLearner consists of some feature coding, a regr.kknn learner and a regr.glm learner plus some extras. Then rf (a regr.ranger) sits at the output level:
