Apiao-1/Kaggle_NFLBigDataBowl

https://www.kaggle.com/c/nfl-big-data-bowl-2020

Code Explation

1. data

   original data from kaggle

2. deprecate

   Useless code file, for backup only

3. online

   model or some GridSearch online

   lgbm:

   • model_lgbm_1365: 直接用lgbm的api：https://www.kaggle.com/enzoamp/nfl-lightgbm/code

   • model_lgbm_multipleClassifier: LGBMClassifier + 平滑， https://www.kaggle.com/mrkmakr/lgbm-multiple-classifier

   • model_lgbm_regression_1412 LGBMClassifier + 平滑,V7 效果不好，cv:0.020105, lb:0.02159 ,猜测哪里有bug,(原文：cv:0.013140205432501861，lb:0.01384)

   NN：

   • model_NN_1362

   • online_NN:online test use

   • online_NN2:online test use

   RF:

   • model_RF_1372

   GridSearch

   • online_cv_lgbm

   • online_cv_rf

   Model Ensemble

   • online_ensemble: simple blend

     • to be optimized

   • online_stacking: The second layer predicts the cumlate probability of each category【deprecate:The results are not strictly increasing and do not meet the requirements of the topic】

   • online_stacking_cumlate:The second layer predicts the probability of each category, then cumlate all just like single model does

     • didn't perform well, maybe try attention stacking later

4. reference

   Code reproduced from discussion area

5. test

   test only

6. model_xgb

   to be finished model

7. feature_filter/feature_filter_test

   FE use