如何使用多索引对时间序列进行滚动窗口计数？

Question

我有这个数据框：

ID      Date                          Received      
000     2018-01-01 00:00:00+00:00     True
        2018-01-01 06:24:44+00:00     True
        2018-01-03 16:24:45+00:00     False
        2018-01-13 20:00:00+00:00     True
        2018-01-13 23:00:00+00:00     True
        2018-01-25 22:30:55+00:00     True 
        2018-01-26 00:30:55+00:00     False

111     2018-01-01 12:00:00+00:00     True
        2018-01-02 15:00:45+00:00     True
        2018-01-04 00:00:00+00:00     True

有没有办法在 7 天内对 Received 列中的 True 值的数量进行滚动窗口计数并按 ID 分组？我尝试使用 df.rolling('7D').count() 但它返回错误。

我一直在寻找这样的东西：

ID      Date range                                                Count     
000     2018-01-01 00:00:00+00:00 - 2018-01-07 00:00:00+00:00     2
        2018-01-08 20:00:00+00:00 - 2018-01-14 00:00:00+00:00     2
        2018-01-15 22:30:55+00:00 - 2018-01-21 20:00:00+00:00     0
        2018-01-22 22:30:55+00:00 - 2018-01-28 20:00:00+00:00     1

111     2018-01-01 00:00:00+00:00 - 2018-01-07 00:00:00+00:00     3

Answer 1

你可以试试pd.Grouper并指定频率：

df.groupby(["ID", pd.Grouper(key='Date', freq='1W')])["Received"].sum()

完整答案：

# Count the number of True per week per ID
out = df.groupby(["ID", pd.Grouper(key='Date', freq='1W')])["Received"] \
        .sum() \
        .to_frame() \
        .reset_index() \
        .rename(columns={"Received": "Count"})
print(out)
#     ID                      Date  Count
# 0  000 2018-01-07 00:00:00+00:00    2.0
# 1  000 2018-01-14 00:00:00+00:00    2.0
# 2  000 2018-01-28 00:00:00+00:00    1.0
# 3  111 2018-01-07 00:00:00+00:00    3.0


# Fill missing date ranges
def fill_date_range(df):
    dates = pd.date_range(df.Date.min(), df.Date.max(), freq="1W")
    return df.set_index("Date") \
             .reindex(dates)[['Count']] \
             .fillna(0)

# Fill missing date range             
out = out.groupby(by="ID").apply(fill_date_range) \
        .reset_index()  \
        .rename(columns={"level_1": "Date"})
print(out)
#     ID                      Date  Count
# 0  000 2018-01-07 00:00:00+00:00    2.0
# 1  000 2018-01-14 00:00:00+00:00    2.0
# 2  000 2018-01-21 00:00:00+00:00    0.0
# 3  000 2018-01-28 00:00:00+00:00    1.0
# 4  111 2018-01-07 00:00:00+00:00    3.0

# Add date range interval as string
format = '%Y-%m-%d %H:%M:%S'
out["Date_expected"] = out.Date.dt.strftime(format) + " - " + (out.Date + pd.Timedelta(weeks=-1)).dt.strftime(format)
print(out)
#     ID                      Date  Count                              Date_expected
# 0  000 2018-01-07 00:00:00+00:00    2.0  2018-01-07 00:00:00 - 2017-12-31 00:00:00
# 1  000 2018-01-14 00:00:00+00:00    2.0  2018-01-14 00:00:00 - 2018-01-07 00:00:00
# 2  000 2018-01-21 00:00:00+00:00    0.0  2018-01-21 00:00:00 - 2018-01-14 00:00:00
# 3  000 2018-01-28 00:00:00+00:00    1.0  2018-01-28 00:00:00 - 2018-01-21 00:00:00
# 4  111 2018-01-07 00:00:00+00:00    3.0  2018-01-07 00:00:00 - 2017-12-31 00:00:00