您可以设置一个布尔掩码来过滤30天内第二次购买的客户组，如下所示：

# Pre-processing to sort the data and convert date to the required date format
df = df.sort_values(['Customer_id', 'purchase_date'])
df['purchase_date'] = pd.to_datetime(df['purchase_date'])

# Set boolean mask
mask = (((df['purchase_date'] - df['purchase_date'].groupby(df['Customer_id']).shift()).dt.days <= 30)
            .groupby(df['Customer_id'])
            .transform('any')
       )

那么，我们已经可以通过以下代码过滤30天内第二次购买的客户的交易记录：

df[mask]

要进一步显示customer_id及其各自的首次购买日期，您可以使用：用途：

df[mask].groupby('Customer_id', as_index=False).first()

你可以用groupby来做

s = df.groupby('Customer_id')['purchase_date'].apply(lambda x : (x.iloc[1]-x.iloc[0]).days<30)
out = df.loc[df.Customer_id.isin(s.index[s])].drop_duplicates('Customer_id')

这里有一个方法：

df2 = (df.loc[df['purchase_date']
              .lt(df['Customer_id']
                  .map((df.sort_values('purchase_date').groupby('Customer_id').first() + pd.to_timedelta(30,'d'))
                       .squeeze()))])

df2 = (df2.loc[df2.duplicated('Customer_id',keep=False)]
       .groupby('Customer_id').first())

或

df = df.sort_values('purchase_date')
m = df.assign(c1 = 1).groupby('Customer_id').rolling('30D',on = 'purchase_date')['c1'].sum().gt(1).groupby(level=0).any()
df.loc[~df.duplicated('Customer_id',keep='first') & df['Customer_id'].map(m)].sort_index()

或

p = 'purchase_date'
c = 'Customer_id'
(df.loc[df[p].le(df.groupby(c)[p].transform('min') + pd.to_timedelta('30D'))
    .groupby(df[c])
    .transform('sum')
    .gt(1)]
    .groupby(c)[p]
    .min())

输出量：

Customer_id
1   2018-07-13
2   2018-06-16
3   2018-07-14