Python 级联分组/转换操作

假设我有一组组和子组，带有日期和值

最后我需要的是评估一个滚动平均值，使用窗口2，按组按月计算。使用过去2个月评估当前月份的值

如果我将数据帧减少两个连续的groupby，我可以实现这一点：

第一个用于计算每个groupby组的值之和的值 组|日期、总数 第二种方法是通过 按组分组|月、变换、滚动 但这减少了我的数据

我需要的是使用转换操作来完成这一切，这样我就可以在原始数据帧上以列的形式获得结果

让我们看看这个虚拟数据：

values = [100, 100, 200, 200, 300, 300]
dates = ['2017-01-01', '2017-02-01',
         '2018-01-01', '2018-02-01',
         '2019-01-01', '2019-02-01']

df1 = pd.DataFrame({'date': dates, 'value': values})
df1['subgroup'] = 'subgroup1'

df2 = df1.copy()
df2['subgroup'] = 'subgroup2'
df2['value'] = df2.value *2

df_g1 = pd.concat([df1, df2], axis=0)

df_g1['group'] = 'group1'

df_g2 = df_g1.copy()
df_g2['group'] = 'group2'
df_g2['value'] = df_g2.value *2

df = pd.concat([df_g1, df_g2], axis=0)
df['date'] = pd.to_datetime(df.date)

现在执行第一个groupby操作：

df_total_by_group = df.groupby(['group', 'date'], as_index=False)[['value']].sum()
df_total_by_group['month'] = df_total_by_group['date'].dt.month

现在滚动平均值：

def rolling_mean(serie):
    return serie.shift(1).rolling(2, min_periods=1).mean()

df_total_by_group['month_rolling_mean_by_group'] = (df_total_by_group
                                                    .groupby(['group', 'month'])['value']
                                                    .transform(rolling_mean)
                                                   )

# display results
df_total_by_group.sort_values(by=['group', 'month'])

我得到了正确的结果， 但我需要它们作为原始数据帧中的列

我在这里迷路了。有什么建议吗？

与列列表一起使用-此处缺少，因为两个数据帧的所有公共列都按Interscion合并：

df = df.merge(df_total_by_group[['group','date','month_rolling_mean_by_group']], how='left')

所以它的工作原理是一样的：

df = df.merge(df_total_by_group[['group','date','month_rolling_mean_by_group']], 
              how='left',
              on=['group','date'])

print (df)
         date  value   subgroup   group  month_rolling_mean_by_group
0  2017-01-01    100  subgroup1  group1                          NaN
1  2017-01-01    200  subgroup2  group1                          NaN
2  2017-02-01    100  subgroup1  group1                          NaN
3  2017-02-01    200  subgroup2  group1                          NaN
4  2018-01-01    200  subgroup1  group1                        300.0
5  2018-01-01    400  subgroup2  group1                        300.0
6  2018-02-01    200  subgroup1  group1                        300.0
7  2018-02-01    400  subgroup2  group1                        300.0
8  2019-01-01    300  subgroup1  group1                        450.0
9  2019-01-01    600  subgroup2  group1                        450.0
10 2019-02-01    300  subgroup1  group1                        450.0
11 2019-02-01    600  subgroup2  group1                        450.0
12 2017-01-01    200  subgroup1  group2                          NaN
13 2017-01-01    400  subgroup2  group2                          NaN
14 2017-02-01    200  subgroup1  group2                          NaN
15 2017-02-01    400  subgroup2  group2                          NaN
16 2018-01-01    400  subgroup1  group2                        600.0
17 2018-01-01    800  subgroup2  group2                        600.0
18 2018-02-01    400  subgroup1  group2                        600.0
19 2018-02-01    800  subgroup2  group2                        600.0
20 2019-01-01    600  subgroup1  group2                        900.0
21 2019-01-01   1200  subgroup2  group2                        900.0
22 2019-02-01    600  subgroup1  group2                        900.0
23 2019-02-01   1200  subgroup2  group2                        900.0

如果对第一次求和使用变换，则其工作方式不同：

df['value'] = df.groupby(['group', 'date'], as_index=False)['value'].transform('sum')
df['month'] = df['date'].dt.month


def rolling_mean(serie):
    return serie.shift(1).rolling(2, min_periods=1).mean()

df['month_rolling_mean_by_group'] = (df.groupby(['group', 'month'])['value']
                                        .transform(rolling_mean))

与列列表一起使用-此处缺少，因为通过Interscion合并两个数据帧的所有公共列：

df = df.merge(df_total_by_group[['group','date','month_rolling_mean_by_group']], how='left')

所以它的工作原理是一样的：

df = df.merge(df_total_by_group[['group','date','month_rolling_mean_by_group']], 
              how='left',
              on=['group','date'])

print (df)
         date  value   subgroup   group  month_rolling_mean_by_group
0  2017-01-01    100  subgroup1  group1                          NaN
1  2017-01-01    200  subgroup2  group1                          NaN
2  2017-02-01    100  subgroup1  group1                          NaN
3  2017-02-01    200  subgroup2  group1                          NaN
4  2018-01-01    200  subgroup1  group1                        300.0
5  2018-01-01    400  subgroup2  group1                        300.0
6  2018-02-01    200  subgroup1  group1                        300.0
7  2018-02-01    400  subgroup2  group1                        300.0
8  2019-01-01    300  subgroup1  group1                        450.0
9  2019-01-01    600  subgroup2  group1                        450.0
10 2019-02-01    300  subgroup1  group1                        450.0
11 2019-02-01    600  subgroup2  group1                        450.0
12 2017-01-01    200  subgroup1  group2                          NaN
13 2017-01-01    400  subgroup2  group2                          NaN
14 2017-02-01    200  subgroup1  group2                          NaN
15 2017-02-01    400  subgroup2  group2                          NaN
16 2018-01-01    400  subgroup1  group2                        600.0
17 2018-01-01    800  subgroup2  group2                        600.0
18 2018-02-01    400  subgroup1  group2                        600.0
19 2018-02-01    800  subgroup2  group2                        600.0
20 2019-01-01    600  subgroup1  group2                        900.0
21 2019-01-01   1200  subgroup2  group2                        900.0
22 2019-02-01    600  subgroup1  group2                        900.0
23 2019-02-01   1200  subgroup2  group2                        900.0

如果对第一次求和使用变换，则其工作方式不同：

df['value'] = df.groupby(['group', 'date'], as_index=False)['value'].transform('sum')
df['month'] = df['date'].dt.month


def rolling_mean(serie):
    return serie.shift(1).rolling(2, min_periods=1).mean()

df['month_rolling_mean_by_group'] = (df.groupby(['group', 'month'])['value']
                                        .transform(rolling_mean))

---

我一直在做的是为我的聚合创建一个新的数据帧，然后连接回原始数据帧

pd.merge(df1, df2, on=['group, 'date'], how='left')

---

我一直在做的是为我的聚合创建一个新的数据帧，然后连接回原始数据帧

pd.merge(df1, df2, on=['group, 'date'], how='left')

---

必须是['group'，date']我认为这是一种新的方式来做它lol-也被编辑以包括date必须是['group'，date']我认为这是一种新的方式来做它lol-也被编辑以包括date