新老客户数据透视表答案

【问题标题】：pivot table for new and old customer新老客户数据透视表
【发布时间】：2021-05-28 00:43:22
【问题描述】：

  customer  purchase_id    payment_status   price currency  payment_date    
1   Andy        6         REPAID             100    GBP   2020-04-16 
2   Randy      10       IN_PROGRESS          10000  SEK   2020-04-17

我想做一个支点，在这里我可以看到有多少新老客户来购买。预期输出：

       new_customers    old_customers
Jan          1                3
Feb          5                2

我被卡住了：

df['year']=df['payment_date'].dt.year
df['month']=df['payment_date'].dt.month
df2=pd.DataFrame(df.groupby("customer", sort=False)["purchase_id"].count())
df2=number_of_purchase.reset_index()
df2.columns = ['merchant_code','number_of_purchase']
df2['repeat_customer']=np.where(df['number_of_purchase']>1,'old_customers','new_customers')

我不知道如何将df2 与df 集成在一起，在df 中，客户可以出现多次，具有不同的purchase_id、价格和付款日期。代码的最后部分大概是这样的：

df.groupby(["year","month", "repeat_customers"])["repeat_customers"].count()

但请随意更改我的代码，输出更重要。

【问题讨论】：

标签： python pandas numpy pandas-groupby

【解决方案1】：

您的样本数据没有足够的特征，所以我生成了一个与结构匹配的随机数据集
使用月份开始要简单得多，因此只需转到这些
新客户需要一个定义，我从您的代码中暗示了它的定义
用这个定义直接计算
最终将结果重塑为您想要的结构
您可能希望在输出 DF 中格式化月份

import numpy as np
d = pd.date_range("01-Jan-2020", periods=10, freq="W")
c = ['tenetur', 'quae', 'rem', 'maxime', 'sunt']
df = pd.DataFrame({"customer":np.random.choice(c, len(d)),
             "purchase_id":np.random.randint(1,10, len(d)),
             "payment_status":np.random.choice(["REPAID","IN_PROGRESS"],len(d)),
             "price":np.random.randint(100,10000, len(d)),
             "currency":np.random.choice(["GBP","SEK"],len(d)),
             "payment_date":d})

# only interested with month start
df2 = (df.assign(ms=df.payment_date - pd.to_timedelta(df.payment_date.dt.day-1, "d"),
           # find first time a customer made a purchase
          fms=lambda dfa: dfa.groupby("customer")["ms"].transform("first"),
           # if month of purchase and first month customer made a purchase are same, new ...
          new_customer=lambda dfa: np.where(dfa.ms==dfa.fms, "new_customers", "old_customers")
         )
 # with the prep it's a simple count
 .groupby(["ms","new_customer"])["customer"].count()
 # format the results
 .to_frame().unstack(1).fillna(0).droplevel(0, axis=1).rename_axis("", axis=1)
)

df

	customer	purchase_id	payment_status	price	currency	payment_date
0	sunt	5	REPAID	9228	SEK	2020-01-05 00:00:00
1	tenetur	6	IN_PROGRESS	1458	SEK	2020-01-12 00:00:00
2	maxime	7	IN_PROGRESS	9798	GBP	2020-01-19 00:00:00
3	sunt	1	IN_PROGRESS	2418	SEK	2020-01-26 00:00:00
4	maxime	8	IN_PROGRESS	6608	GBP	2020-02-02 00:00:00
5	sunt	4	IN_PROGRESS	2341	GBP	2020-02-09 00:00:00
6	rem	7	REPAID	8961	GBP	2020-02-16 00:00:00
7	quae	2	REPAID	7068	GBP	2020-02-23 00:00:00
8	maxime	1	IN_PROGRESS	4872	SEK	2020-03-01 00:00:00
9	tenetur	1	REPAID	2860	GBP	2020-03-08 00:00:00

df2

ms	new_customers	old_customers
2020-01-01 00:00:00	4	0
2020-02-01 00:00:00	2	2
2020-03-01 00:00:00	0	2

【讨论】：