基于分组和条件添加新列：Pandas教程

import pandas as pd
import numpy as np

data = {
    'id': [1, 2, 3, 4, 5, 6, 7],
    'date': ['2019-02-01', '2019-02-10', '2019-02-25', '2019-03-05', '2019-03-16', '2019-04-05', '2019-05-15'],
    'date_difference': [None, 9, 15, 11, 10, 19, 40],
    'number': [1, 0, 1, 0, 0, 0, 0],
    'text': ['A', 'A', 'A', 'A', 'A', 'B', 'B']
}

df = pd.DataFrame(data)
print(df)

import pandas as pd


data = {
    'id': [1, 2, 3, 4, 5, 6, 7],
    'date': ['2019-02-01', '2019-02-10', '2019-02-25', '2019-03-05', '2019-03-16', '2019-04-05', '2019-05-15'],
    'date_difference': [None, 9, 15, 11, 10, 19, 40],
    'number': [1, 0, 1, 0, 0, 0, 0],
    'text': ['A', 'A', 'A', 'A', 'A', 'B', 'B']
}

df = pd.DataFrame(data)

out = df.assign(
    # We assign the following values to the series name "test"
    test=df
    # Group on "text" -- if we grouped on ["text", "number"] we wouldn't see different numbers within the groups.
    .groupby("text")
    # Apply a chain of methods to the group (a pd.DataFrame).
    .apply(
        lambda g: (
            # We sort "date" in descending order as you mention this partially controls the step size.
            g.sort_values(by="date", ascending=False)
            # We shift "number" forward one period with a fill_value of 1 for any newly introduced nulls.
            .number.shift(periods=1, fill_value=1)
            # Cumulatively sum the shifted "number" values
            .cumsum()
        )
        # This will result in the new series, albeit sorted by descending "date".
    )
    # Drop the "text" level of the new multi-index.
    .droplevel("text")
    # The assign method acts as join, rearranging the newly created series to match the index of `df`.
)
print(out)