search

fanghaolei / K_sa

Kagge shelter animal competition
GNU General Public License v3.0
6 stars 0 forks source link

A question related to model stacking #1

Open mshih2 opened 8 years ago

mshih2 commented 8 years ago

In the descriptions, you said the final output from your model is a result of stacking XGB and RF models, but in the code "trainModel" it is not obvious that how you stack them together. It is possible for you to explain the way you stacked the models? Thanks!

Mei-Cheng

ghost commented 8 years ago

I did not push the stacking code into the repo. I think for the initial ones, I simply averaged the predictions from several models. I had some more complicated stacked models built up, but I got frustrated by the causality issues in this data so I discontinued the project.

You might find this guide being helpful: http://mlwave.com/kaggle-ensembling-guide/.

Thanks!

mshih2 commented 8 years ago

Thank you for your reply! Actually before you applied me, I tried to use Bayesian Optimization to determine the weights of the models. The method didn't give me a better result than the result you got, but it did improve my result from 0.709xxx->0.703xxx, you can also try this if you want!