Python pandas RFM模型应用实例详解

#读取数据
rfm<-read.csv('~/desktop/rfm1_7.csv',header=TRUE)
summary(rfm)
#数据分布
par(mfrow=c(1,3))
boxplot(rfm$rankR1) 
boxplot(rfm$rankF1) 
boxplot(rfm$rankM1)
#rfm分级
breaks1<-quantile(rfm$Recency, probs = seq(0, 1, 0.2),names = FALSE)
breaks1<-c(1,14,30,57，111,181) #以流失用户的定义来设置分级 30天以上为流失用户
breaks2<-quantile(rfm$Frequency, probs = seq(0, 1, 0.2),names = FALSE)
breaks2<-c(1,2,3,6,14,164) 
breaks3<-quantile(rfm$Monetary, probs = seq(0, 1, 0.2),names = FALSE)
rfm$rankR1<- cut(rfm$Recency,breaks1, 5,labels=F)
rfm$rankR1<- 6-rfm$rankR1
rfm$rankF1<- cut(rfm$Frequency,breaks2, 5,labels=F)
rfm$rankM1<- cut(rfm$Monetary,breaks3, 5,labels=F)

#聚类
df<-rfm[,c(6,7,8)]
p1<-fviz_nbclust(df, kmeans, method = "wss")
p2<-p1 + geom_vline(xintercept = 5, linetype = 2)
km_result <- kmeans(df, 5)
dd <- cbind(rfm,df, cluster = km_result$cluster)
##查看每一类的数目
table(dd$cluster)
picture<-fviz_cluster(km_result, df, geom = "point")
####聚类结果解释####
rfm_final <- within(dd,{Custom = NA
Custom[cluster == 1] = '高价值客户' 
Custom[cluster == 2 ] = '无价值客户' 
Custom[ cluster == 3] = '无价值客户' 
Custom[cluster == 4] = '重点发展客户'  
Custom[cluster == 5] = '重点挽留客户' 
})

import pandas as pd
import numpy as np
import time
#todo 读取数据
data = pd.read_csv('RFM_TRAD_FLOW.csv',encoding='gbk')
# print(ret)
# todo RFM------>R(最近一次消费)
#todo 时间与字符串相互转换
data['time'] = data['time'].map(lambda x:time.mktime(time.strptime(x,'%d%b%y:%H:%M:%S')))
# print(data)
# todo 分组
groupby_obj = data.groupby(['cumid','type'])
# for name,data in groupby_obj:
#   print(name)
#   print(data)
# todo 取值
R = groupby_obj[['time']].max()
# print(
# todo 转为透视表
r_trans = pd.pivot_table(R,index='cumid',columns='type',values='time')
# print(data_trans)
# todo 替换缺失值 有缺失值，替换成最远的值
r_trans[['Special_offer','returned_goods']] = r_trans[['Special_offer','returned_goods']].apply(lambda x:x.replace(np.nan,min(x)),axis = 0)
# print(data_trans)
r_trans['r_max'] = r_trans.apply(lambda x:sum(x),axis=1)
# print(r_trans)
# todo RFM------>F(消费频率)
# 取值
F =groupby_obj[['transID']].count()
# print(F)
#转为透视表
f_trans = pd.pivot_table(F,index='cumid',columns='type',values='transID')
# print(f_trans)
#替换缺失值
f_trans[['Special_offer','returned_goods']]= f_trans[['Special_offer','returned_goods']].fillna(0)
# print(f_trans)
#
f_trans['returned_goods'] = f_trans['returned_goods'].map(lambda x:-x)
# print(f_trans)
f_trans['f_total'] = f_trans.apply(lambda x:sum(x),axis=1)
# print(f_trans)
# todo RFM------>M(消费金额)
# 取值
M =groupby_obj[['amount']].sum()
# print(M)
#转为透视表
m_trans = pd.pivot_table(M,index='cumid',columns='type',values='amount')
# print(f_trans)
#替换缺失值
m_trans[['Special_offer','returned_goods']]= m_trans[['Special_offer','returned_goods']].fillna(0)
# print(f_trans)
#
m_trans['m_total'] = m_trans.apply(lambda x:sum(x),axis=1)
# print(m_trans)
# 合并
RFM=pd.concat([r_trans["r_max"],f_trans['f_total'],m_trans['m_total']],axis=1)
print(RFM)
r_score = pd.cut(RFM.r_max,3,labels=[0,1,2])
f_score = pd.cut(RFM.r_max,3,labels=[0,1,2])
m_score = pd.cut(RFM.r_max,3,labels=[0,1,2])