https://www.kaggle.com/competitions/uplift-shift-23
This branch uses the standard approach: sklift library with various base estimators and different sets of features:
- An original set of features: files with "all_features" in file names.
- Added new features.
- Ranked new features by importance and tried various sets from the 1st to n-the features.
XGBoost proved to be the best base estimator so far. The last notebook with best results is 12_...
Notebooks:
- EDA
- data_prep: data preparation
- baseline*: first attempts
- adversarial_validation: checking that all data splits are properly stratified
- numbered notebooks (01_..., etc.): various solutions