import pandas as pd
import numpy as np


# make a dataset

df = pd.DataFrame({''sales'': [''2020-01'',''2020-02'',''2020-03'',''2020-04'',''2020-05'',''2020-06''],''2020-01'': [24,42,18,68,24,30],''2020-02'': [24,''2020-03'': [64,70,88,57],''2020-04'': [22,11,44,3,5,78],''2020-05'': [11,35,74,12,69,51]})

print(df)

# datawork(L2)

for i in range(0,df.shape[0]):
    
    if i==0:
        df.loc[i,''L2'']=df.loc[i,''2020-01'']
    
    else:
        if i!=df.shape[0]-1:
            df.loc[i,''L2'']=df.iloc[i-1:i+1,i:i+2].sum().sum()
            
        if i==df.shape[0]-1:
             df.loc[i,i-1:i+1].sum().sum()

print(df)

#     sales  2020-01  2020-02  2020-03  2020-04  2020-05     L2
#0  2020-01       24       24       64       22       11   24.0
#1  2020-02       42       42       24       11       35  132.0
#2  2020-03       18       18       70       44       74  154.0
#3  2020-04       68       68       70        3       12  187.0
#4  2020-05       24       24       88        5       69   89.0
#5  2020-06       30       30       57       78       51  203.0

Values=f.values[:,1:]
L2=[]
RANGE=Values.shape[0]
for a in range(RANGE):
    if a==0:
        result=Values[a,a]
    else:
        if Values[a-1:a+1,a-1:a+1].shape==(2,1):
            result=np.sum(Values[a-1:a+1,a-2:a])
        
        else:
            result=np.sum(Values[a-1:a+1,a-1:a+1])
         
    L2.append(result)
print(L2)
L2 output:-->[24,132,154,187,89,203]

f["L2"]=L2

f:

我尝试了另一种方法。 这个方法使用了reshape long（在python中：melt），但是我在python中应用了reshape long两次，因为df中销售和其他列的时间频率是每月而不是每天，所以我再一次reshape long使int列对应于每月日期。 （我用Stata的次数比python多，在Stata中，我只能做一次reshape很长时间，因为它有每月的时间频率，而且reshape任务比pandas，python要容易得多）

有兴趣的可以看看

# 00.module

import pandas as pd
import numpy as np
from order import order # https://stackoverflow.com/a/68464246/16478699

# 0.make a dataset

df = pd.DataFrame({''sales'': [''2020-01'',51]}
                  )
df.to_stata(''dataset.dta'',version=119,write_index=False)
print(df)

# 1.reshape long(in python: melt)

t = list(df.columns)
t.remove(''sales'')
df_long = df.melt(id_vars=''sales'',value_vars=t,var_name=''var'',value_name=''val'')
df_long[''id''] = list(range(1,df_long.shape[0] + 1))  # make id for another resape long
print(df_long)

# 2.another reshape long(in python: melt,reason: make int(col name: tid) corresponding to monthly date of sales and monthly columns in df)

df_long2 = df_long.melt(id_vars=[''id'',''val''],value_vars=[''sales'',''var''])
df_long2[''tid''] = df_long2[''value''].apply(lambda x: 1 + list(df_long2.value.unique()).index(x))
print(df_long2)

# 3.back to wide form with tid(in python: pd.pivot)

df_wide = pd.pivot(df_long2,index=[''id'',columns=''variable'',values=[''value'',''tid''])
df_wide.columns = df_wide.columns.map(lambda x: x[1] if x[0] == ''value'' else f''{x[0]}_{x[1]}'')  # change multiindex columns name into just normal columns name
df_wide = df_wide.reset_index()
print(df_wide)

# 4.make values of L2

for i in df_wide.tid_sales.unique():

    if list(df_wide.tid_sales.unique()).index(i) + 1 == len(df_wide.tid_sales.unique()):
        df_wide.loc[df_wide[''tid_sales''] == i,''L2''] = df_wide.loc[(((df_wide[''tid_sales''] == i) | (
                    df_wide[''tid_sales''] == i - 1)) & ((df_wide[''tid_var''] == i - 1) | (
                    df_wide[''tid_var''] == i - 2))),''val''].sum()

    else:
        df_wide.loc[df_wide[''tid_sales''] == i,''L2''] = df_wide.loc[(((df_wide[''tid_sales''] == i) | (
                    df_wide[''tid_sales''] == i - 1)) & ((df_wide[''tid_var''] == i) | (
                    df_wide[''tid_var''] == i - 1))),''val''].sum()

print(df_wide)

# 5.back to shape of df with L2(reshape wide,in python: pd.pivot)

df_final = df_wide.drop(columns=df.filter(regex=''^tid''))  # no more columns starting with tid needed
df_final = pd.pivot(df_final,index=[''sales'',''L2''],columns=''var'',values=''val'').reset_index()
df_final = order(df_final,''L2'',f_or_l=''last'')  # order function is made by me
print(df_final)