From the course: Advanced Predictive Modeling: Mastering Ensembles and Metamodeling

Unlock the full course today

Join today to access over 22,600 courses taught by industry experts or purchase this course individually.

Stacking

Stacking

- [Instructor] Okay, now we're going to try a very simplistic stacking model. Stacking models do extremely well in caggle.com competitions. But this is really just going to be a conceptual example. This really doesn't have the level of complexity of a competition winning stacking model. So what's stacking? Well, what I have here is three model predictions, in fact the same ones that we used when we simply averaged them. We have our Chevy Malibu and we've got an actual miles per gallon of 18 and we have the predictions from Neural Net, the CART and the linear regression. So what stacking is is let's give these predictions to a fourth model and the job of that fourth model is to decide how to combine and how to weight these. So it won't be a simple average, it'll be a weighted average essentially as determined by this fourth model. Let's take a look. Okay, this is atypical of a stacking model, we wouldn't normally have a CART sitting on top of this. But here we can really see what's…

Contents