在 pandas 中自定义一周的开始日期

Question

将一周的开始自定义为星期四，将一周的结束自定义为星期三以提取周数

我尝试过使用

df['Date'].dt.to_period('THU-F')

但是 df['Date'].dt.to_period('W-THU') 工作正常时它不起作用

Answer 1

这似乎是一个直截了当的问题，但它有点棘手。要获得周数，从不同的一天开始一周，您首先必须获得该年所有可能的周。然后您必须进行累积计数才能获得周数。最后，您可以将周数与您的数据结合起来。这应该可以解决问题：

import random
import pandas
import datetime


def create_week_numbers(since: datetime.datetime = datetime.datetime(year=2019, month=1, day=1), until: datetime.datetime = datetime.datetime.now()) -> pandas.DataFrame:
    """ Create a dataframe that contains "year", "week" (start-end dates), and "week_number" for all
    possible dates between "since" and "until".

    Args:
        since (datetime.datetime, optional): The start date. Defaults to datetime.datetime(year=2019, month=1, day=1).
        until (datetime.datetime, optional): The end date. Defaults to datetime.datetime.now().

    Returns:
        pandas.DataFrame: A dataframe with all possible weeks and week numbers
    """

    # Generate every possible date since a start date until now
    dates = [since]
    while dates[-1] < until:
        dates.append(dates[-1] + datetime.timedelta(days=1))

    # Convert all the dates into a dataframe
    all_dates = pandas.DataFrame(dates, columns=["date"])

    # Add a column for the year, and add a column for the week (using the to_period method)
    all_dates["year"] = all_dates["date"].dt.year
    all_dates["week"] = all_dates["date"].dt.to_period("W-WED")  # https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#anchored-offsets

    # Ignore the individual days and keep only unique weeks per year
    weeks_per_year = all_dates[["week", "year"]].drop_duplicates()

    # Add the week number by incrementally counting the weeks by year
    weeks_per_year["week_number"] = weeks_per_year.groupby(["year"]).cumcount()

    # Return the year, the span of th week, and the week number in the year
    return weeks_per_year[["year", "week", "week_number"]]


# Create a list of all week numbers
week_number_lookup = create_week_numbers()

# Generate some random test dataframes
df = pandas.DataFrame([{
    "date": datetime.datetime.fromtimestamp(random.randint(1579947057, 1606297560)),
} for _ in range(10000)])

# Add the week to the dataframe
df["week"] = df["date"].dt.to_period("W-WED")  # Week ends on Wednesday

# Merge with the lookup to find the week number
df.merge(on="week", right=week_number_lookup)

输出：

                    date                   week  year  week_number
0    2020-03-17 04:19:48  2020-03-12/2020-03-18  2020           11
1    2020-03-12 04:40:42  2020-03-12/2020-03-18  2020           11
2    2020-03-16 03:17:10  2020-03-12/2020-03-18  2020           11
3    2020-03-14 08:03:52  2020-03-12/2020-03-18  2020           11
4    2020-03-13 05:27:02  2020-03-12/2020-03-18  2020           11
...                  ...                    ...   ...          ...
9995 2020-05-12 00:55:55  2020-05-07/2020-05-13  2020           19
9996 2020-05-09 06:07:28  2020-05-07/2020-05-13  2020           19
9997 2020-05-08 10:01:15  2020-05-07/2020-05-13  2020           19
9998 2020-05-12 12:52:55  2020-05-07/2020-05-13  2020           19
9999 2020-05-07 15:46:41  2020-05-07/2020-05-13  2020           19

[10000 rows x 4 columns]

仅供参考，您原来的df['Date'].dt.to_period('THU-F') 不起作用的原因是它不是有效值。看看你在这里的选项：https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#anchored-offsets