基於 dataframe 生成 3 個不同的列

Question

我有一個 dataframe：

Date_1      Date_2  individual_count
01/09/2019  02/08/2019  2
01/09/2019  03/08/2019  2
01/09/2019  04/08/2019  2
01/09/2019  05/08/2019  2
.   .   .
01/09/2019  28/08/2019  10
01/09/2019  29/08/2019  11
01/09/2019  30/08/2019  12
01/09/2019  31/08/2019  14

我想生成 3 列，num_days_2、num_days_3、num_days_5、num_days_20

我想以如下方式聚合數據集：

num_days_2 : all individual_count aggregated for date_1 for date_2 = (date_2- 2, date_2- 1)
num_days_3 : all individual_count aggregated for date_1 for date_2 = (date_2- 5, date_2- 3)
num_days_5 : all individual_count aggregated for date_1 for date_2 = (date_2- 6, date_2- 10)
num_days_20 : all individual_count aggregated for date_1 for date_2 = left all dates

例如，對於特定的 date_1 = 01/09/2019：

num_days_2 = sum of individual counts for date_2 = 30/08/2019 - 31/08/2019 
num_days_3 = sum of individual counts for date_2 = 27/08/2019 - 29/08/2019 
num_days_5 = sum of individual counts for date_2 = 26/08/2019 - 22/08/2019 
num_days_20 = sum of individual counts for date_2 = 25/08/2019 - 02/08/2019 

編輯

預期 output：

Date_1      num_days_2  num_days_3  num_days_5  num_days_20
01/09/2019   
02/09/2019
.
.
.
30/09/2019

任何人都可以實現相同的目標。

Answer 1

我創建了一個您可以使用的示例。 您可能需要重命名列，並查看cut的 function 以正確排序垃圾箱。

# Generate example data.
# This is just an way go generate data that can be used to simulate your data.
df = pd.DataFrame(
    data=dict(
      Date_1=pd.Timestamp('today'), # This is Date_1
      Date_2=pd.date_range(end=pd.Timestamp('today'), periods=25), # This is Date_2
      individual_count=range(25) # This is individual_count
    )
)

# Calculate an offset as integer days:
# For each day, calculate the differace in days between day Date1 and Date2
df['offset_timedelta'] = (df.Date_1 - df.Date_2)
# To make bining eaiser convert the datetime delta to ints.
df['offset'] = df['offset_timedelta'].dt.days.astype('int16')


# Create bins for each offset:
# Each row will be grouped into an interval. based on the list [1,2,5,10,1000]
# 1000 is just an upper bound to get "the rest"
df['bins'] = pd.cut(df['offset'], [1,2,5,10,1000], include_lowest=True)

# This groups on day1 and the bin, so that we can sum for each.
grouped = df.groupby(['Date_1','bins'])[['individual_count']].sum()

# The groupby gives and index of 'Date_1','bins'. This converts bins to columns instead of and index.
final = grouped.unstack()

編輯：重命名列以使它們更像原始問題。