0

I recently was reading about ensemble models, and I want to try constructing one with my data. I want to randomly combine my two models with the lowest RMSLE scores into one similarly constructed model.

My first model uses a boosted regression tree, has 1000 rows and 2 columns (ID and prediction).

My second model is an elastic net regression, also with 1000 rows and 2 columns.

How can I randomly combine the predictions of each model without repeating or missing any ID values? Also, if anyone has other tips about using this type of modeling, I'd greatly appreciate them!

Jamie Warren
  • 79
  • 8
  • 1
    You should use the predictions from your two models as the input for the ensable model. Meaning that prediction of model one and model two for each case are the explanatory variables - the traget stays the same. So you would have to combine the results of both models acording to the cases first to then proceed with training of the the ensamble model. https://www.r-bloggers.com/2020/05/how-to-build-stacked-ensemble-models-in-r/ – DPH Mar 07 '21 at 02:41
  • 1
    A nice framework for ensemble models: https://stacks.tidymodels.org/articles/basics.html – Jon Spring Mar 07 '21 at 03:17
  • @DPH Fantastic. That makes much more sense. I appreciate the link! – Jamie Warren Mar 07 '21 at 22:24

0 Answers0