Self Validating Ensemble ModelsJMP Chief Data Scientist @chris_gotwalt1 with Principal Lecturer from the University of New Hampshire @philramsey introduced the Self Validating Ensemble Models (SVEM) technique for analyzing DoEs. I think Chris says it best "You can get more accurate models with less observations using SVEM". This is a bridge between Machine Learning and DoE. I hope you enjoy it.
Re: Self Validating Ensemble ModelsVery interesting - thanks for sharing! Actually I was about to ask a lot of questions regarding how to design DoEs for that kind of analysis, however Chris raised all of them himself in the final discussion. I am really looking forward to see the SVEMs evolving and I am excited to hear about optimal designs for them. If available, I am definitely willing to try them out!
I do have one question for my understanding of the example. In the explanation Chris said "You fit some statistical model to the weighted data". In the concrete example does that mean when talking about SVEM RSM we use the training weights calculated in each iteration as a “Weight” in the “Fit Model” platform when fitting a RSM?
Re: Self Validating Ensemble Models@Kira_Alhorn, Yes. What he is saying is that with each model in the SVEM you are re-evaluating the paired fractionally Weighted Bootstrap Weight column, so you will have a model with a new weight on each member of the training set. If you are using Fit Model platform this column will be used as the weight variable. For the Generalized regression platform, this will be used as the freq variable.
Hope this helps. Let me know if I need to clarify further.
