What are some good models to complement XGBOOST via stacking in typical Kaggle datascience competition? I realize XGBoost with well-tuned hyperparamters are generally quite good already.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
It really depends on how other models behave on your particular dataset, but generally, you should check 2 things:
- Predictions from ensembled models are not highly correlated and other models also perform well on your dataset.
- It's better to use a model from another
family. So if you are using XGboost, its better to add some model like SVM than a catboost (because you want a model with different approach in solving the data, rather then a model with similar approach).
I think this article may give you additional insights about the topic https://machinelearningmastery.com/stacking-ensemble-machine-learning-with-python/
