【机器学习小案例篇】关于RFM模型的小案例

import pandas as pd

1. 导入数据

In [75]:

trad_flow = pd.read_csv('D:\python\Script\RFM_TRAD_FLOW.csv',encoding='gbk')  #编码格式需要进行修改
trad_flow.head() #默认前五行

Out[75]:

	transID	cumid	time	amount	type_label	type
0	9407	10001	14JUN09:17:58:34	199.0	正常	Normal
1	9625	10001	16JUN09:15:09:13	369.0	正常	Normal
2	11837	10001	01JUL09:14:50:36	369.0	正常	Normal
3	26629	10001	14DEC09:18:05:32	359.0	正常	Normal
4	30850	10001	12APR10:13:02:20	399.0	正常	Normal

2.计算 RFM

In [21]:

M = trad_flow.groupby(['cumid','type'])[['amount']].sum()

In [48]:

trains_M = pd.pivot_table(M,index='cumid',columns='type',values='amount')
trains_M.head()

Out[48]:

type	Normal	Presented	Special_offer	returned_goods
cumid
10001	3608.0	0.0	420.0	-694.0
10002	1894.0	0.0	NaN	-242.0
10003	3503.0	0.0	156.0	-224.0
10004	2979.0	0.0	373.0	-40.0
10005	2368.0	0.0	NaN	-249.0

In [47]:

F = trad_flow.groupby(['cumid','type'])[['transID']].count()
F.head()

Out[47]:

		transID
cumid	type
10001	Normal	15
	Presented	8
	Special_offer	2
	returned_goods	2
10002	Normal	12

In [46]:

R = trad_flow.groupby(['cumid','type'])[['time']].max()
R.head()

Out[46]:

		time
cumid	type
10001	Normal	21JUL09:09:31:26
	Presented	31MAR10:20:29:48
	Special_offer	12OCT09:10:59:13
	returned_goods	10JUL10:20:41:54
10002	Normal	29JUL09:19:21:41

3.衡量客户对打折商品的偏好

In [53]:

trains_M['Special_offer'] = trains_M['Special_offer'].fillna(0)
trains_M['Special_offer'].head()

Out[53]:

cumid
10001    420.0
10002      0.0
10003    156.0
10004    373.0
10005      0.0
Name: Special_offer, dtype: float64

In [67]:

trains_M['spe_ratio'] = trains_M['Special_offer']/(trains_M['Special_offer']+trains_M['Normal'])
trains_M['spe_ratio'].head()

Out[67]:

cumid
10001    0.104270
10002    0.000000
10003    0.042635
10004    0.111277
10005    0.000000
Name: spe_ratio, dtype: float64

In [68]:

M_rank = trains_M.sort_values('spe_ratio',ascending=False).head()
M_rank.head()

Out[68]:

type	Normal	Presented	Special_offer	returned_goods	spe_ratio
cumid
10151	765.0	0.0	870.0	NaN	0.532110
40033	1206.0	0.0	761.0	-848.0	0.386884
40236	1155.0	0.0	691.0	-793.0	0.374323
30225	1475.0	0.0	738.0	-301.0	0.333484
20068	1631.0	0.0	731.0	-239.0	0.309483

In [74]:

pd.qcut(M_rank['spe_ratio'],4)

Out[74]:

cumid
10151    (0.387, 0.532]
40033    (0.374, 0.387]
40236    (0.333, 0.374]
30225    (0.308, 0.333]
20068    (0.308, 0.333]
Name: spe_ratio, dtype: category
Categories (4, interval[float64]): [(0.308, 0.333] < (0.333, 0.374] < (0.374, 0.387] < (0.387, 0.532]]

 

相关：

【机器学习业务篇】基于RFM模型的用户分群方法

【机器学习杂烩篇】详解 Pandas 透视表（pivot_table）

【机器学习杂烩篇】pandas fillna()函数详解

【机器学习杂烩篇】Pandas 排序sort_values

【机器学习杂烩篇】Pandas中的qcut和cut

 

数据链接：https://pan.baidu.com/s/1Qet7LWD89jVGBQK9QUQnDg