[Python] EDA 대신 Pandas Profiling

Example

• examples
• census_report

Default

from pandas_profiling import ProfileReport

pr = ProfileReport(pdf2)
pr.to_file(output_file="feature_profiling.html")

# Disable samples, correlations, missing diagrams and duplicates at once
r = ProfileReport(
    samples=None,
    correlations=None,
    missing_diagrams=None,
    duplicates=None,
    interactions=None,
)

• config_default.yaml
• 내가 필요한 부분만 분석하고 싶으면
  • https://pandas-profiling.github.io/pandas-profiling/docs/master/rtd/pages/advanced_usage.html

Minimal

from pandas_profiling import ProfileReport

pr = ProfileReport(pdf2, minimal=True)
pr.to_file(output_file="feature_profiling.html")

• config_minimal.yaml

Parameters

# Number of workers in thread pool. When set to zero, it is set to the number of CPUs available.
# pool_size, default = 0
pr = ProfileReport(df, minimal=True, pool_size=0) 

• 기본값으로 모든 thread pool을 사용도록 설정되어 있음

화면에 표시

# Widget 형태로 보여서 확인하기가 편함 (Tab으로 각각 속성에 대해서 보여줌)
pr.to_widgets()

참고

• https://github.com/pandas-profiling/pandas-profiling
• https://pandas-profiling.github.io/pandas-profiling/docs/master/rtd/pages/advanced_usage.html