The 2nd Kospo X Dacon Recommendation Algorithm AI Contest(Recruiting)

Recruit | Algorithm | Tabular | Recommendation System | RMSE

 

[Private 5위 3.27893] Optuna + KFold 5 +Catboost + Ensemble

2023.05.21 19:17 2,866 Views language

1) 최초 접근법 (한계)
- AutoML 전체 적용 해봄 (pycaret, H2O -> 대부분 : rmse 3.4 까지 나옴)
2) 피처 엔지니어링 실행
- 데이터 전처리 및 범주화 실행 (Year-Of-Publication, Age, Location)
3) 모델링 최적화
- AutoML (pycaret) 통해서 찾은 최적 모델 catboost 선택 - Optuna 최적화 실행
- StratifiedKFold 5 적용 모델 5개 평균 Rating prediction
- PostProcessing 적용 -1, 11 예측 -> Rating 0, 10 치환
4) 앙상블(Submit)
- rmse 제일 낮은 2개 catboost 모델 6:4 비율로 제출
(최종 : Public LB: 3.26368, Private LB : 3.27893)

PDF
Code