search

Pipe-Runner-Lab / CVPR2020-FGVC7

CVPR2020-FGVC7 Kaggle submission
MIT License
0 stars 0 forks source link

Take ensembling inspiration from there 2 kernels #9

Open Pipe-Runner opened 4 years ago

Pipe-Runner commented 4 years ago

https://www.kaggle.com/chekoduadarsh/ensemble-efficientnet-densenet201-resnet152 https://www.kaggle.com/seefun/ensemble-top-kernels

Pipe-Runner commented 4 years ago

My pipeline already has mean ensemble mechanism. I need to check out few more models to combine results with. Efficientnet + Xception+ ResNet152

Pipe-Runner commented 4 years ago

Ensembling added ( Need to verify performance )

Pipe-Runner commented 4 years ago

I don't have any rule for choosing base/meta models… I test many models with many different **params.

For example you can use XGBoost, CatBoost, and LightGBM as base models and meta models, then, you ensemble the 3 meta models shown above using simple bagging.