计算滚动 3 天熊猫的不同计数？

Question

我想统计按城市分组的 3 天窗口内的唯一客户

输入：

    df = pd.DataFrame([['1A', 'Cairo', '2020-12-01'],
                ["2A", 'Cairo', '2020-12-01'],
                ['1A', 'Cairo', '2020-12-02'],
                ['1A', 'Cairo', '2020-12-03'],
                ['3A', 'Alex', '2020-12-01'],
                ['3A', 'Alex', '2020-12-02'],
                ['3A', 'Alex', '2020-12-03'],
                ['4A', 'Giza', '2020-12-02'],
                ['4A', 'Giza', '2020-12-02'],
                ['5A', 'Giza', '2020-12-03'],
                ['6A', 'Giza', '2020-12-01']], columns=
    ['customer_id', 'city', 'day'])

预期输出：

    output = pd.DataFrame([['Alex', '2020-12-01',1],
                ['Alex', '2020-12-02',1],
                ['Alex', '2020-12-03',1],
                ['Cairo', '2020-12-01',2],
                ['Cairo', '2020-12-02',2],
                ['Cairo', '2020-12-03',2],
                ['Giza', '2020-12-01',1],
                ['Giza', '2020-12-02',2],
                ['Giza', '2020-12-03',3]], columns=
    ['city', 'day', 'unique_customers_last3Days'])

我试过了：

df['day'] = pd.to_datetime(df['day'])
df.set_index('day',inplace=True)
df.sort_index(inplace=True)
df.groupby('city').rolling("3D").agg({'customer_id':'nun'})

但它给了我错误

AttributeError: 'nunique' is not a valid function for 'RollingGroupby' object

Answer 1

将数据框的索引设置为day，然后将sort设置为索引值，现在factorizecustomer_id列，以便为每个客户ID分配唯一的代码，然后group@上的数据框987654326@ 和 apply rolling nunique 操作，窗口大小为 3 days。 （可选）dropday 中每个 city 的重复值

df = df.set_index('day').sort_index()
df['codes'] = df['customer_id'].factorize()[0]

df.groupby('city')\
  .rolling('3D')['codes'].apply(pd.Series.nunique)\
  .reset_index(name='unique').drop_duplicates(['city', 'day'], keep='last')

---

     city        day  unique
0    Alex 2020-12-01     1.0
1    Alex 2020-12-02     1.0
2    Alex 2020-12-03     1.0
4   Cairo 2020-12-01     2.0
5   Cairo 2020-12-02     2.0
6   Cairo 2020-12-03     2.0
7    Giza 2020-12-01     1.0
9    Giza 2020-12-02     2.0
10   Giza 2020-12-03     3.0