python - 更改 LabelMaker 的一周开始日期(composeml 库)

Question

我正在尝试为客户和一周创建标签。
我需要labeling_function计算周一至周五的销售额。
但现在它从星期日计算到星期日。
如何更改 LabelMaker 的一周开始日期

def total_spent(df):
    total = df['amount'].sum()
    return total

label_maker = cp.LabelMaker(
    target_entity="customer_id",
    time_index="transaction_time",
    labeling_function=total_spent,
    window_size="W",
)

Answer 1

谢谢你的提问。您可以通过将窗口大小设置为 W-MON 来获取每周一的频率。 .我将通过这个数据快速举例。

import pandas as pd

records = []
for time in  pd.date_range(start='2020-11-16', periods=15, freq='d'):
    record = {'transaction_time': time, 'day_name': time.day_name()}
    records.append(record)

df = pd.DataFrame(records).assign(customer_id=0)

transaction_time   day_name  customer_id
      2020-11-16     Monday            0
      2020-11-17    Tuesday            0
      2020-11-18  Wednesday            0
      2020-11-19   Thursday            0
      2020-11-20     Friday            0
      2020-11-21   Saturday            0
      2020-11-22     Sunday            0
      2020-11-23     Monday            0
      2020-11-24    Tuesday            0
      2020-11-25  Wednesday            0
      2020-11-26   Thursday            0
      2020-11-27     Friday            0
      2020-11-28   Saturday            0
      2020-11-29     Sunday            0
      2020-11-30     Monday            0

在标签制作器中，我将窗口大小设置为 W-MON .这是星期一每周频率的偏移别名。窗口大小还支持许多其他 offset aliases来自 Pandas 。

import composeml as cp

lm = cp.LabelMaker(
    target_entity='customer_id',
    time_index='transaction_time',
    window_size='W-MON',
)

让我们检查标签生成器生成的数据切片。您应该在星期一获得每周频率。

slices = lm.slice(df, -1)
next(slices)

                   day_name  customer_id
transaction_time                        
2020-11-16           Monday            0
2020-11-17          Tuesday            0
2020-11-18        Wednesday            0
2020-11-19         Thursday            0
2020-11-20           Friday            0
2020-11-21         Saturday            0
2020-11-22           Sunday            0

next(slices)

                   day_name  customer_id
transaction_time                        
2020-11-23           Monday            0
2020-11-24          Tuesday            0
2020-11-25        Wednesday            0
2020-11-26         Thursday            0
2020-11-27           Friday            0
2020-11-28         Saturday            0
2020-11-29           Sunday            0