Time Series (pandas)

02 Jan 2024 in Study / Computer science on Pandas

pandas

• Array elements’ dtypes : Timestamp or Period
  • Basic structure : year-month-day (ex. 2024-01-02)
  • Timestamp : At time point
    • form : y-m-d or y-m-d time
  • Period : time interval
    • form : y or y-m or y-m-d or y-m-d time
• If we set time series (one column) as index of dataframe, it is convenient to deal with these
  • only extracting year series or month series or day series
  • accessing the specific records using time series elements

Make Time Series

• Way 1
  • Existing string column(Series) representing time (y-m-d) → Time Series (dtypes : timestamp)
  • pd.to_datetime(df[column’s name] or any array like list)
    • return Series whose data type is timestamp
  • If we wanna change from timestamp series to period series
    • (timestamp_series).to_period(freq=)
      • return Series whose data type is period
      • freq=’D’ : y-m-d
      • freq=’M’ : y-m
      • freq=’A’ : y
• Way 2
  • Create Time Series directly
  • pd.date_range(start=, end=, periods=, freq=, tz=)
    • ex. start=’2024-01-01’ : Time Series’ first element is ‘2024-01-01’
    • end= : User can determine the end element directly (end=None : periods parameter will determine it)
    • periods=n : The number of elements in Time Series is n
    • freq=
      • ex. freq=’3A’ (year end) : 2024-12-31, 2027-12-31, 2030-12-31, …
      • ex. freq=’2AS’ (year begin) : 2024-01-01, 2026-01-01, 2028-01-01, …
      • ex. freq=’2M’ (month end) : 2024-01-31, 2024-03-31, 2024-05-31, …
      • ex. freq=’3MS’ (month begin) : 2024-01-01, 2024-04-01, 2027-07-01, …
      • ex. freq=’2D’ : 2024-01-01, 2024-01-03, 2024-01-05, …
    • ex. tz=’Asia/Seoul’ : 00:00:00+09:00
  • pd.period_range(start=, end=, periods=, freq=)
    • ex. start=’2024-01-01’ : Time Series’ first element is ‘2024-01-01’ (Its form can be changed because it is period dtypes)
    • end= : User can determine the end element directly (end=None : periods parameter will determine it)
    • periods=n : The number of elements in Time Series is n
    • freq=
      • ex. freq=’3A’ : 2024, 2027, 2030, …
      • ex. freq=’2M’ : 2024-01, 2024-03, 2024-05, …
      • ex. freq=’2D’ : 2024-01-01, 2024-01-03, 2024-01-05, …
      • ex. freq=’2H’ : 2024-01-01 00:00, 2024-01-01 02:00, 2024-01-01 04:00, …

Use Time Series

• We can only extract year series or month series or day series
  • df[‘year’] = df[timeSeriesColumn’s name].dt.year : add new column (‘year’) whose data is year series
  • df[‘month’] = df[timeSeriesColumn’s name].dt.month : add new column (‘month’) whose data is month series
  • df[‘day’] = df[timeSeriesColumn’s name].dt.day : add new column (‘day’) whose data is day series
  • df[‘year_month’] = df[timeSeriesColumn’s name].dt.to_period(freq=’M’) : add new column (‘year_month’) whose data is y-m series
• If we set Time Series column as index of dataframe (set_index)
  • Method for accessing element
    • ex. df.loc[‘2024’] : return dataframe whose index contains year as 2024
    • ex. df.loc[‘2024-01’] : return dataframe whose index’s form is 2024-01
    • ex. df.loc[‘2024-01’ : ‘2024-05’, ‘c1’ : ‘c3’] : return dataframe (index : 2024-01 ~ 2024-05, column : c1 ~ c3)

  • Difference between two dates
    

    standard_date = pd.to_datetime('y-m-d (example)')
    df['difference'] = standard_date - df.index(timeSeries)