如何用Python Datetime合并跨年的第52周与第0周数据？

我有一个存储日度数据的Pandas DataFrame，尝试按周编号分组对部分列求和时发现：当年份并非从周日开始时，跨年的周数据会被拆分为两组，无法正常合并求和。

原实现代码 #

原周编号生成及分组求和代码如下：

df = pd.DataFrame([
  {'Date': '2020-03-01', 'population': 100, 'admission': 10, 'release': 10, 'feature_id': 'abc'},
  {'Date': '2020-03-02', 'population': 100, 'admission': 10, 'release': 10, 'feature_id': 'abc'},
  ...
])

df['week_num'] = df['Date'].dt.strftime('%Y-%U')
by_week = df.groupby(["week_num", "feature_id"])[["population", "admission", "release"]].sum().reset_index()

问题现象 #

分组后的数据出现跨年周被拆分的情况，示例如下：

...,
{
  'week_num': '2020-52',
  'start_of_week': '2020-12-27T00:00:00.000Z',
  'end_of_week': '2021-01-02T00:00:00.000Z',
  'population': 1000,
  'admission': 100,
  'release': 150
},
{
  'week_num': '2021-00',
  'start_of_week': '2020-12-27T00:00:00.000Z',
  'end_of_week': '2021-01-02T00:00:00.000Z',
  'population': 200,
  'admission': 120,
  'release': 90
},
...

实际上，2020-12-27至2021-01-02这一周的数据应合并为一组，即population=1000+200=1200, admission=100+120=220, release=150+90=240。

问题根源 #

问题出在提取周起止日期的Week类实现上，该类使用%U格式（以周日为一周起始的周编号规则），导致跨年周被拆分到两个年份的周编号中：

import datetime as dt

class Week:
    def __init__(self, year, week_num):
        week_start = str(year) + " " + str(week_num) + " 0"
        week_end = str(year) + " " + str(week_num) + " 6"
        self.first_date = dt.datetime.strptime(week_start, '%Y %U %w')
        self.last_date = dt.datetime.strptime(week_end, '%Y %U %w')

...

by_week["start_of_week"] = by_week["iso_week_num"].apply(lambda w: Week(*w.split("-")).first_date)

最终解决方案 #

采用ISO标准周编号格式（%G-%V，以周一为一周起始，跨年周统一归属到包含周四的年份），可以避免该问题，代码如下：

import datetime

df["week_num"] = df["Date"].dt.strftime("%G-%V")
by_week = df.groupby(["week_num", "feature_id"])[["population", "admission", "release"]].sum().reset_index()
by_week["start_of_week"] = by_week["week_num"].apply(lambda d: datetime.strptime(d+"1", "%G-%V%u"))
by_week["end_of_week"] = by_week["week_num"].apply(lambda d: datetime.strptime(d+"7", "%G-%V%u"))

内容的提问来源于stack exchange，提问作者OJT