python实现RFM模型分析

python实现RFM模型分析

import numpy as np
import pandas as pd
import pyreadstat
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D  # 空间三维画图
import time
import datetime as dt


#================================================
path = 'E:/data/'#数据文件路径
file_name = 'eledata.sav'#数据集文件名
output_file_name = 'rfm_result' #输出文件名
#================================================
#注意：使用本模型尽量将不同用户群分开各自打分，例如，按照居民，商业用电，工业用电分开使用


'''设置rfm取值参数观察期(3个月，6个月，12个月)'''
time_window = 12
#观察期的长短，用户近3/6/12个月内的缴费次数和缴费总额确定F与M分值

'''设置R参数(最近一次缴费时间距离今天的天数)'''
param_r = [15,30,60,90]
#exp：param_r = [15,30,60,90]
#小于15天得分5,15-30天得分4,30-60得分3,60-90得分2,90以上得分1

'''自定义函数：'''
#1.计算Recency分值
def R_score(param_r,x):
    #param_r:    R分值划分区间
    #x： 需要被计算的具体数值
    if x<=param_r[0]:
        return 5
    elif x>param_r[0] and x<=param_r[1]:
        return 4
    elif x>param_r[1] and x<=param_r[2]:
        return 3
    elif x>param_r[2] and x<=param_r[3]:
        return 2
    elif x>param_r[3]:
        return 1

#2.计算f,m得分
def RFM_score(series,q,score_labels):
    a = series.quantile(q=q)
    if len(set(a))>len(score_labels):
        return pd.qcut(series,q=len(score_labels),labels=score_labels)
    else:
        return pd.cut(series,bins=len(score_labels),labels=score_labels)
    
#2.计算rfm用户分群
def RFM_group(df,idx,x,r,f,m):
    #df:  DataFrame of data set
    #idx: column name of user Index
    #x:   user index
    #r:   column name of Recency score
    #f:   column name of Frequency score
    #m:   column name of Monetary score
    df1 = df[df[idx]==x]
    if df1[r].values>=df[r].mean() and df1[f].values>=df[f].mean() and df1[m].values>=df[m].mean():
        return '重要价值客户'
    elif df1[r].values>=df[r].mean() and df1[f].values<df[f].mean() and df1[m].values>=df[m].mean():
        return '重要发展客户'
    elif df1[r].values<df[r].mean() and df1[f].values>=df[f].mean() and df1[m].values>=df[m].mean():
        return '重要保持客户'
    elif df1[r].values<df[r].mean() and df1[f].values<df[f].mean() and df1[m].values>=df[m].mean():
        return '重要挽留客户'
    elif df1[r].values>=df[r].mean() and df1[f].values>=df[f].mean() and df1[m].values<df[m].mean():
        return '一般价值客户'
    elif df1[r].values>=df[r].mean() and df1[f].values<df[f].mean() and df1[m].values<df[m].mean():
        return '一般发展客户'
    elif df1[r].values<df[r].mean() and df1[f].values>=df[f].mean() and df1[m].values<df[m].mean():
        return '一般保持客户'
    elif df1[r].values<df[r].mean() and df1[f].values<df[f].mean() and df1[m].values<df[m].mean():
        return '一般挽留客户'

df,meta = pyreadstat.read_sav(path+file_name)
print(meta.file_encoding)

'''确定now是今天还是19年底'''
# now = pd.to_datetime('today')
now = dt.datetime(2019,12,31)

'''重新给列命名'''
if '合计' in df.columns:
    df.rename(columns={'单位':'unit','交费ID':'payfee_id','交费方式':'pay_chnl','用户号':'acct_id','区编号':'tq_id','区名称':'tq_name',
                   '交费金额':'pay_fee','收费日期':'pay_date','等级':'vol_lvl','用户分类':'user_cat','合计':'total_use'},
         inplace=True)
else:
    df.rename(columns={'单位':'unit','交费ID':'payfee_id','交费方式':'pay_chnl','用户号':'acct_id','区编号':'tq_id','区名称':'tq_name',
                   '交费金额':'pay_fee','收费日期':'pay_date','等级':'vol_lvl','用户分类':'user_cat'},
         inplace=True)
    
    
'''取出有用字段'''
df = df[['acct_id', 'tq_id',  'voltage_lvl', 'user_cat' ,'pay_chnl', 
       'pay_fee', 'pay_date']]
'''对id进行取整'''
df['acct_id'] = df['acct_id'].astype('int64')
df['tq_id'] = df['tq_id'].astype('str')

UTF-8

'''一、加工自变量'''
starttime = dt.datetime.now()
df_fin = pd.DataFrame()
# df_mid = pd.DataFrame()
for i in df['acct_id'].unique():
    df1 = df[df['acct_id']==i].sort_values(by='pay_date',ascending=True)

    df1['time_span'] = df1['pay_date'].map(lambda x:(now-x).days)

    '''1.基础信息'''
    df_mid = df1[['acct_id','tq_id','voltage_lvl','user_cat']][:1]

    '''2.缴费渠道个数'''
    df_mid['pay_chnl_cnt'] = df1['pay_chnl'].unique().shape[0]

    '''3.近一次缴费时间，间隔'''
    df_mid['last_pay_date'] = df1[-1:]['pay_date'].values
    df_mid['last_pay_span'] = df1[-1:]['time_span'].values

    '''4.按照日为单位汇总金额'''
    df1 = df1[['acct_id','time_span','pay_fee']].groupby(['acct_id','time_span']).agg({'pay_fee':lambda x:x.sum()}).sort_values(by='time_span',ascending=False).reset_index()

    '''5.加工时间切片内缴费金额的总额，次数'''
    #1.近3个月
    df_mid['last_3m_fee_sum'] = df1[df1['time_span']<=90]['pay_fee'].sum()
    df_mid['last_3m_fee_cnt'] = df1[df1['time_span']<=90]['pay_fee'].count()
    #2.近6个月
    df_mid['last_6m_fee_sum'] = df1[df1['time_span']<=180]['pay_fee'].sum()
    df_mid['last_6m_fee_cnt'] = df1[df1['time_span']<=180]['pay_fee'].count()
    #3.近12个月
    df_mid['last_12m_fee_sum'] = df1[df1['time_span']<=365]['pay_fee'].sum()
    df_mid['last_12m_fee_cnt'] = df1[df1['time_span']<=365]['pay_fee'].count()
    
    '''6.保存加工好的数据'''
    df_fin = df_fin.append(df_mid)
df_fin.reset_index(drop=True,inplace=True)
endtime = dt.datetime.now()
print('用时',(endtime - starttime).seconds/60,'分钟')

用时 0.0 分钟

'''三、建立rfm模型进行用户群体划分'''
#1.计算R,F,M得分
df_fin['Recency'] = df_fin['last_pay_span'].map(lambda x:R_score(param_r,x))
df_fin['Frequency'] = RFM_score(df_fin['last_'+str(time_window)+'m_fee_cnt'],[0,0.2,0.4,0.6,0.8,1],[1,2,3,4,5])
df_fin['Monetary'] = RFM_score(df_fin['last_'+str(time_window)+'m_fee_sum'],[0,0.2,0.4,0.6,0.8,1],[1,2,3,4,5])   
df_fin['Frequency'] = df_fin['Frequency'].astype('int')
df_fin['Monetary'] = df_fin['Monetary'].astype('int')
#2.计算rfm得分：
df_fin['RFM_score'] = df_fin['Recency']*100+df_fin['Frequency']*10+df_fin['Monetary']*1
#3.用户分群
# df_fin['user_group'] = df_fin['acct_id'].map(lambda x:RFM_group(df_fin,'acct_id',x,'Recency','Frequency','Monetary'))
rfm_q = [0,1/8,2/8,3/8,4/8,5/8,6/8,7/8,1]
rfm_cat = ['一般挽留客户','一般保持客户','一般发展客户','一般价值客户','重要挽留客户','重要保持客户','重要发展客户','重要价值客户']
df_fin['user_group'] = RFM_score(df_fin['RFM_score'],rfm_q,rfm_cat)
#输出结果到文件
df_fin.to_excel(path+output_file_name+'.xlsx')

#做三维散点图
fig = plt.figure()
ax = Axes3D(fig)
color_dict = {'一般挽留客户':'k','一般保持客户':'darkgreen','一般发展客户':'tan','一般价值客户':'midnightblue',
              '重要挽留客户':'pink','重要保持客户':'yellow','重要发展客户':'cyan','重要价值客户':'r'}
ax.scatter(df_fin['Recency'], df_fin['Frequency'], df_fin['Monetary'],zdir='y',c=df_fin['user_group'].map(lambda x:color_dict[x]))


# 添加坐标轴(顺序是x,y,z)
ax.set_xlabel('Recency', fontdict={'size': 15, 'color': 'k'})
ax.set_ylabel('Frequency', fontdict={'size': 15, 'color': 'k'})
ax.set_zlabel('Monetary', fontdict={'size': 15, 'color': 'k'})
plt.show()
fig.savefig(path+'rfm.png')

df_fin['user_group']

0     一般挽留客户
1     重要挽留客户
2     重要价值客户
3     重要价值客户
4     一般价值客户
5     重要价值客户
6     重要保持客户
7     重要价值客户
8     重要保持客户
9     重要价值客户
10    一般价值客户
Name: user_group, dtype: category
Categories (8, object): [一般挽留客户 < 一般保持客户 < 一般发展客户 < 一般价值客户 < 重要挽留客户 < 重要保持客户 < 重要发展客户 < 重要价值客户]