Notice
Recent Posts
Recent Comments
Link
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | ||||
4 | 5 | 6 | 7 | 8 | 9 | 10 |
11 | 12 | 13 | 14 | 15 | 16 | 17 |
18 | 19 | 20 | 21 | 22 | 23 | 24 |
25 | 26 | 27 | 28 | 29 | 30 | 31 |
Tags
- sklearn
- 알고리즘
- R
- 파이썬
- level 1
- 머신러닝
- 빅분기
- Kaggle
- level 2
- pandas
- 실습
- 코딩테스트
- Numpy
- matplotlib
- SQL
- Python
- 프로그래머스
- python3
- seaborn
- Oracel
- 튜닝
- 오라클
- 카카오
- 실기
- oracle
- 데이터 분석
- 빅데이터 분석 기사
Archives
- Today
- Total
라일락 꽃이 피는 날
Bootstrapping (부트스트래핑) 본문
728x90
Bootstrapping (부트스트래핑)
# A그룹: gate_30 / B그룹: gate_40
# 각각의 A,B 그룹에 대해 bootstrapp된 means값의 리스트 생성
boot_1d = []
for i in range(1000):
boot_mean = df.sample(frac = 1,replace = True).groupby('version')['retention_1'].mean()
boot_1d.append(boot_mean)
# list를 DataFrame으로 변환
boot_1d = pd.DataFrame(boot_1d)
# A Kernel Density Estimate plot of the bootstrap distributions
boot_1d.plot(kind='density')

위의 두 분포는 A,B 그룹에 대해 1 day retention(retention_1)이 가질 수 있는 부트 스트랩 불확실성을 표현한다.
# 두 A,B 그룹간의 % 차이 평균 컬럼 추가
boot_1d['diff'] = (boot_1d.gate_30 - boot_1d.gate_40)/boot_1d.gate_40*100
# bootstrap % 차이를 시각화
ax = boot_1d['diff'].plot(kind='density')
ax.set_title('% difference in 1-day retention between the two AB-groups')
# 게이트가 레벨30에 있을 때 1-day retention이 클 확률 계산
(boot_1d['diff'] > 0).mean() # 0.96

위 도표에서 가장 가능성이 높은 % 차이는 약 1%-2%이며 분포의 95%는 0%이상이므로 레벨 30의 게이트를 선호한다.
부트 스트랩 분석에 따르면 게이트가 레벨 30에있을 때 1일 유지율(retention_1)이 더 높을 가능성이 있다.
# 각각의 A,B그룹에 대해 bootstrapp된 means 값의 리스트 생성
boot_7d = []
for i in range(500):
boot_mean = df.sample(frac=1,replace=True).groupby('version')['retention_7'].mean()
boot_7d.append(boot_mean)
# list를 DataFrame으로 변환
boot_7d = pd.DataFrame(boot_7d)
# 두 A,B 그룹간의 % 차이 평균 컬럼 추가
boot_7d['diff'] = (boot_7d.gate_30 - boot_7d.gate_40)/boot_7d.gate_40*100
# bootstrap % 차이를 시각화
ax = boot_7d['diff'].plot(kind='density')
ax.set_title('% difference in 7-day retention between the two AB-groups')
# 게이트가 레벨30에 있을 때 7-day retention이 더 클 확률 계산
(boot_7d['diff'] > 0).mean() # 1.0

부트 스트랩 결과는 게이트가 레벨 40에 있을 때보다 레벨 30에 있을 때 7일 retention(retention_7)이 더 높다는 것을 나타낸다.
결론은 retention을 늘리기 위해서 게이트를 레벨 30에서 레벨 40으로 이동해서는 안된다는 것이다.
728x90
'데이터 분석 > Python' 카테고리의 다른 글
Chi-Square (카이제곱 검정) (0) | 2021.06.03 |
---|---|
T-test (T-검정) (0) | 2021.06.03 |
텍스트 마이닝 (0) | 2021.05.27 |
Kmeans 군집 분류 (0) | 2021.05.27 |
회귀분석 (0) | 2021.05.27 |