聚集在PANDAS的滞后时间段内

Question

我有一个多索引数据框，其中包含日期和股票代码的索引。这是一个子集：

我想创建几个滞后变量。我已经想出如何创造一天的滞后期：

df['Number of Tweets [t-1]'] = df['Number of Tweets'].unstack().shift(1).stack()

我陷入困境的是创建一个滞后变量，它在第t-1到t-3天，或者t-1到t-7，或者t-1到t-30之间聚合（求和）值。例如，我想要一个名为“推文数量[t-1到t-3之和]”的列。我已经玩过 DateOffset 并查看了 resample 但是还没有比这更进一步。我似乎也无法在Cookbook中找到任何答案或文档中的示例有帮助。我正在转动这个轮子，所以我将不胜感激。

Answer 1

对移位的数据使用pd.rolling_sum。要计算t-3到t-1的滚动总和，请使用窗口长度3并将数据移位1（如果未指定参数，则为默认值）。

from pandas import Timestamp

# Create series
s = pd.Series({(Timestamp('2015-03-30 00:00:00'), 'AAPL'): 2,
 (Timestamp('2015-03-30 00:00:00'), 'IBM'): 3,
 (Timestamp('2015-03-30 00:00:00'), 'TWTR'): 2,
 (Timestamp('2015-03-31 00:00:00'), 'AAPL'): 6,
 (Timestamp('2015-03-31 00:00:00'), 'IBM'): 2,
 (Timestamp('2015-03-31 00:00:00'), 'TWTR'): 7,
 (Timestamp('2015-04-01 00:00:00'), 'AAPL'): 3,
 (Timestamp('2015-04-01 00:00:00'), 'IBM'): 1,
 (Timestamp('2015-04-01 00:00:00'), 'TWTR'): 2,
 (Timestamp('2015-04-02 00:00:00'), 'AAPL'): 6,
 (Timestamp('2015-04-02 00:00:00'), 'IBM'): 8,
 (Timestamp('2015-04-02 00:00:00'), 'TWTR'): 2,
 (Timestamp('2015-04-06 00:00:00'), 'AAPL'): 4,
 (Timestamp('2015-04-06 00:00:00'), 'IBM'): 2,
 (Timestamp('2015-04-06 00:00:00'), 'TWTR'): 6,
 (Timestamp('2015-04-07 00:00:00'), 'AAPL'): 3,
 (Timestamp('2015-04-07 00:00:00'), 'IBM'): 7,
 (Timestamp('2015-04-07 00:00:00'), 'TWTR'): 8})

# View the data more easily:
s.unstack() 
            AAPL  IBM  TWTR
Date                       
2015-03-30     2    3     2
2015-03-31     6    2     7
2015-04-01     3    1     2
2015-04-02     6    8     2
2015-04-06     4    2     6
2015-04-07     3    7     8

# Calculate a rolling sum on date t for dates t-3 through t-1:
result = pd.rolling_sum(s.unstack().shift(), window=3)  # .shift() <=> .shift(1)

>>> result
            AAPL  IBM  TWTR
Date                       
2015-03-30   NaN  NaN   NaN
2015-03-31   NaN  NaN   NaN
2015-04-01   NaN  NaN   NaN
2015-04-02    11    6    11
2015-04-06    15   11    11
2015-04-07    13   11    10

# Restack the data:
>>> result.stack()
2015-04-02  AAPL    11
            IBM      6
            TWTR    11
2015-04-06  AAPL    15
            IBM     11
            TWTR    11
2015-04-07  AAPL    13
            IBM     11
            TWTR    10