kookmin-sw / 2019-cap1-2019_4

Flex Ads to implement advertising system in offline market
MIT License
10 stars 7 forks source link

Evaluation기록하기!! #50

Closed Jisoo-Min closed 5 years ago

Jisoo-Min commented 5 years ago

evaluation 할때마다 수치 기록하기

aaa0265 commented 5 years ago

First

위 5개의 feature를 통한 evaluation 결과 private : 0.24738 public : 0.24899

aaa0265 commented 5 years ago

second

위 7개 feature를 통한 evaluation 결과 private : 0.34293 public : 0.34564

aaa0265 commented 5 years ago

Third

두번째 eval에 한 가지(timezone)를 추가한 것 private : 0.34128 public : 0. 34404 -> timezone이 추가되면서 결과가 조금 내려감.

aaa0265 commented 5 years ago

스크린 캡쳐로 설명 대체 01

aaa0265 commented 5 years ago

스크린샷, 2019-05-14 16-07-22

현재까지 최근 구매 기록을 살피는 함수와 order_ratio_bychane()를 사용하여 만든 feature가 있을 때가 제일 높은 기록을 가지고 있음.

aaa0265 commented 5 years ago

model 생성시 최대 depth을 줄이면 정말 작은 값이지만 증가하는 것을 볼 수 있다. 비교 -> 첫번째와 다섯번째 값을 통해 알 수 있음

aaa0265 commented 5 years ago

reordered_xxx , order_ratio_bychance 함수와 model depth 5로 줄인 것이 0.36302로 제일 높아졌다.

Jisoo-Min commented 5 years ago

기록 그래프로 공유하고 이슈 닫기~

aaa0265 commented 5 years ago

그래프 17번 제출 score 그래프 입니다.