364 lines
11 KiB
Python
364 lines
11 KiB
Python
# -*- coding: utf-8 -*-
|
||
|
||
|
||
"""
|
||
基于RFM模型生成数据分析报告
|
||
"""
|
||
|
||
# 导入模块
|
||
|
||
import statistics
|
||
from datetime import datetime
|
||
from decimal import Decimal, ROUND_HALF_UP
|
||
|
||
import pandas
|
||
from jinja2 import Environment, FileSystemLoader
|
||
|
||
from utils.client import MySQLClient
|
||
from utils.pandas_extension import DrawAsHTML
|
||
|
||
|
||
# 函数说明：根据RFM编码映射为客户分类
|
||
def map_classification(r_encoded, f_encoded, m_encoded):
|
||
# 就R、F、M指标构建独热编码并匹配客户分类
|
||
match f"{r_encoded}{f_encoded}{m_encoded}":
|
||
case "000":
|
||
classification = "流失客户"
|
||
case "010":
|
||
classification = "一般维持客户"
|
||
case "100":
|
||
classification = "新客户"
|
||
case "110":
|
||
classification = "潜力客户"
|
||
case "001":
|
||
classification = "重要挽留客户"
|
||
case "101":
|
||
classification = "重要深耕客户"
|
||
case "011":
|
||
classification = "重要唤回客户"
|
||
case "111":
|
||
classification = "重要价值客户"
|
||
|
||
# noinspection PyUnboundLocalVariable
|
||
return classification
|
||
|
||
|
||
print("1 加载数据集...", end="")
|
||
|
||
client = MySQLClient(database="data_analysis")
|
||
|
||
dataframe = client.execute_query(
|
||
sql="select 客户ID, 交易金额, 交易日期 from rfm_dataset"
|
||
) # customer_id 客户ID STRING，trade_date 交易日期 DATETIME.DATE，trade_amount 交易金额 DECIMAL
|
||
|
||
print("已完成")
|
||
|
||
print("2 预处理，删除包含缺失值的样本和重复样本...", end="")
|
||
|
||
# 删除包含缺失值的样本
|
||
dataframe.dropna(inplace=True)
|
||
|
||
# 删除重复样本（保留第一例重复样本、重置索引）
|
||
dataframe = dataframe.drop_duplicates(ignore_index=True, inplace=False)
|
||
|
||
# 仅保留交易日期为2012和2013年的样本
|
||
dataframe = dataframe[
|
||
dataframe["交易日期"].apply(lambda x: x.year in [2012, 2013])
|
||
].reset_index(
|
||
drop=True
|
||
) # 因交易日期数据类型为DATETIME.DATE非DATETIME64，故无法使用SERIES.DT.YEAR方法
|
||
|
||
sample_size = Decimal(dataframe.shape[0]).quantize(Decimal("0"))
|
||
|
||
print("已完成")
|
||
|
||
print("3 构建RFM...", end="")
|
||
|
||
# 最远交易日期
|
||
min_trade_date = dataframe["交易日期"].min()
|
||
|
||
# R为最近一次交易日期距离样本中最远一次交易日期的天数（单位：日），DECIMAL；F为交易频率（单位；次），DECIMAL；M为交易金额（单位：元），DECIMAL。均正向化
|
||
rfm = (
|
||
dataframe.groupby(by="客户ID")
|
||
.agg(
|
||
R=(
|
||
"交易日期",
|
||
lambda x: Decimal((x.max() - min_trade_date).days).quantize(
|
||
Decimal("0"), rounding=ROUND_HALF_UP
|
||
),
|
||
),
|
||
F=(
|
||
"客户ID",
|
||
lambda x: Decimal(len(x)).quantize(Decimal("0"), rounding=ROUND_HALF_UP),
|
||
),
|
||
M=(
|
||
"交易金额",
|
||
lambda x: sum(x, Decimal("0")).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
), # 求和时指定初始值为DECIMAL("0")
|
||
),
|
||
)
|
||
.reset_index()
|
||
)
|
||
|
||
# 客户数
|
||
customer_counts = Decimal(rfm.shape[0]).quantize(Decimal("0"))
|
||
|
||
# 总交易金额
|
||
trade_amounts = sum(rfm["M"], Decimal("0.00")).quantize(Decimal("0.00"))
|
||
|
||
print("已完成")
|
||
|
||
print("4 基于平均数将R、F和M分为低、高两个等级并组合为八种客户分类...", end="")
|
||
|
||
# R、F和M的平均数，使用STATISTICS.MEAN统计平均值，保证精度
|
||
# noinspection PyUnresolvedReferences
|
||
means = {
|
||
"R": statistics.mean(rfm["R"]).quantize(Decimal("0.00"), rounding=ROUND_HALF_UP),
|
||
"F": statistics.mean(rfm["F"]).quantize(Decimal("0.00"), rounding=ROUND_HALF_UP),
|
||
"M": statistics.mean(rfm["M"]).quantize(Decimal("0.00"), rounding=ROUND_HALF_UP),
|
||
}
|
||
|
||
rfm = rfm.assign(
|
||
客户分类=lambda dataframe: dataframe.apply(
|
||
lambda row: map_classification(
|
||
r_encoded=0 if row["R"] <= means["R"] else 1,
|
||
f_encoded=0 if row["F"] <= means["F"] else 1,
|
||
m_encoded=0 if row["M"] <= means["M"] else 1,
|
||
),
|
||
axis="columns",
|
||
)
|
||
)
|
||
|
||
dataframe = dataframe.merge(right=rfm[["客户ID", "客户分类"]], on="客户ID", how="left")
|
||
|
||
print("已完成")
|
||
|
||
print("5 生成分析报告...", end="")
|
||
|
||
draw = DrawAsHTML()
|
||
|
||
# 生成数据预览
|
||
draw.table(
|
||
dataframe=dataframe.sample(5),
|
||
file_name="数据预览.html",
|
||
)
|
||
|
||
# 客户分类维度
|
||
customer_types = (
|
||
rfm.groupby(by="客户分类") # 按照客户分类分组
|
||
.agg(
|
||
R=(
|
||
"R",
|
||
lambda x: statistics.mean(x).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
),
|
||
), # R平均值
|
||
F=(
|
||
"F",
|
||
lambda x: statistics.mean(x).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
),
|
||
), # F平均值
|
||
M=(
|
||
"M",
|
||
lambda x: statistics.mean(x).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
),
|
||
), # M平均值
|
||
客户占比=(
|
||
"客户分类",
|
||
lambda x: (Decimal(len(x)) / customer_counts * Decimal("100")).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
),
|
||
), # 统计各客户分类的客户占比
|
||
交易金额占比=(
|
||
"M",
|
||
lambda x: Decimal(
|
||
sum(x, Decimal("0.00")) / trade_amounts * Decimal("100")
|
||
).quantize(Decimal("0.00"), rounding=ROUND_HALF_UP),
|
||
), # 统计各客户分类的交易金额占比
|
||
)
|
||
.reset_index()
|
||
)
|
||
|
||
# 生成客户分类分布
|
||
draw.scatter(
|
||
dataframe=customer_types[["客户分类", "R", "F", "M"]],
|
||
xaxis_opts_min=475,
|
||
xaxis_opts_max=750,
|
||
file_name="客户分类分布.html",
|
||
)
|
||
|
||
# 生成客户占比
|
||
draw.pie(
|
||
dataframe=customer_types[["客户分类", "客户占比"]].sort_values(
|
||
by="客户占比", ascending=False
|
||
), # 按照客户占比降序
|
||
file_name="客户占比.html",
|
||
)
|
||
|
||
# 生成交易金额占比
|
||
draw.pie(
|
||
dataframe=customer_types[["客户分类", "交易金额占比"]].sort_values(
|
||
by="交易金额占比", ascending=False
|
||
), # 按照交易金额占比降序
|
||
file_name="交易金额占比.html",
|
||
)
|
||
|
||
report_backward = pandas.DataFrame(data=[], columns=["客户分类", "窗口期", "客户数"])
|
||
|
||
for customer_type in customer_types["客户分类"]:
|
||
|
||
for month in range(1, 13):
|
||
|
||
# 窗口期，从2013-01至2013-12
|
||
period = f"2013-{month:02d}"
|
||
|
||
# 窗口期起期（向前滑动十二个月，包括当月）
|
||
period_start = (
|
||
pandas.Period(value=f"2013-{month:02d}", freq="M") - 11
|
||
).start_time.date()
|
||
|
||
# 窗口期止期
|
||
period_end = pandas.Period(value=f"2013-{month:02d}", freq="M").end_time.date()
|
||
|
||
# 指定客户分类窗口期内客户数
|
||
customer_counts = dataframe.loc[
|
||
(dataframe["客户分类"] == customer_type)
|
||
& (dataframe["交易日期"] >= period_start)
|
||
& (dataframe["交易日期"] <= period_end),
|
||
"客户ID",
|
||
].nunique()
|
||
|
||
report_backward.loc[report_backward.shape[0]] = [
|
||
customer_type,
|
||
period,
|
||
customer_counts,
|
||
]
|
||
|
||
# 生成近十二个自然月客户数趋势
|
||
draw.area(
|
||
dataframe=report_backward.groupby(by="窗口期", as_index=False).agg(
|
||
客户数=("客户数", "sum")
|
||
),
|
||
file_name="近十二个自然月客户数趋势.html",
|
||
yaxis_opts_min=1350,
|
||
)
|
||
|
||
report_backward = report_backward.loc[
|
||
report_backward["客户分类"].isin(
|
||
["新客户", "流失客户", "重要价值客户"]
|
||
) # 仅考虑新客户、流水客户、重要价值客户
|
||
].assign(
|
||
总客户数=lambda x: x.groupby(by="窗口期")["客户数"].transform(
|
||
"sum"
|
||
), # 统计窗口期总客户数并新增值各行
|
||
客户占比=lambda x: x.apply(
|
||
lambda y: (
|
||
Decimal(y["客户数"]) / Decimal(y["总客户数"]) * Decimal("100")
|
||
).quantize(
|
||
Decimal("0.00"), rounding=ROUND_HALF_UP
|
||
), # 运算各项使用DECIMAL以控制精度
|
||
axis="columns",
|
||
),
|
||
)
|
||
|
||
# 生成近十二个自然月客户占比趋势（仅考虑新客户、流失客户和重要价值客户）
|
||
draw.bar(
|
||
dataframe=report_backward[
|
||
["客户分类", "窗口期", "客户占比"]
|
||
], # 仅保留客户分类、窗口期和占比
|
||
file_name="近十二个自然月客户占比趋势.html",
|
||
stack=True,
|
||
)
|
||
|
||
report_forward = (
|
||
dataframe.assign(
|
||
最早交易日期=lambda dataframe: dataframe.groupby(by="客户ID")[
|
||
"交易日期"
|
||
].transform("min"),
|
||
) # 统计每位客户最早交易日期
|
||
.assign(
|
||
周期=lambda dataframe: (
|
||
dataframe["交易日期"].apply(lambda x: x.year)
|
||
- dataframe["最早交易日期"].apply(lambda x: x.year)
|
||
)
|
||
* 12
|
||
+ (
|
||
dataframe["交易日期"].apply(lambda x: x.month)
|
||
- dataframe["最早交易日期"].apply(lambda x: x.month)
|
||
)
|
||
) # 每笔交易的交易日期和对客户最早交易日期的间隔作为周期，单位为月
|
||
.assign(
|
||
周期=lambda dataframe: dataframe["周期"].apply(lambda x: f"M+{x:02d}")
|
||
) # 格式化周期
|
||
.assign(
|
||
群组=lambda dataframe: dataframe["最早交易日期"].apply(
|
||
lambda x: f"{x.year}-{x.month:02d}"
|
||
)
|
||
) # 截取最早交易日期的年月作为群组
|
||
.groupby(by=["客户分类", "群组", "周期"], as_index=False)
|
||
.agg(客户数=("客户ID", "nunique")) # COHORT-ANALYSIS，群组-周期矩阵
|
||
.groupby(by=["客户分类", "周期"], as_index=False)
|
||
.agg(客户数=("客户数", "sum")) # 统计各客户分类各周期用户数
|
||
.assign(
|
||
基准客户数=lambda dataframe: dataframe.groupby(by=["客户分类"])[
|
||
"客户数"
|
||
].transform("first")
|
||
)
|
||
.assign(
|
||
留存率=lambda dataframe: dataframe.apply(
|
||
lambda x: (
|
||
Decimal(x["客户数"]) / Decimal(x["基准客户数"]) * Decimal("100")
|
||
).quantize(Decimal("0.00"), rounding=ROUND_HALF_UP),
|
||
axis="columns",
|
||
)
|
||
)
|
||
.loc[
|
||
lambda dataframe: dataframe["客户分类"].isin(
|
||
["新客户", "流失客户", "重要价值客户"]
|
||
)
|
||
& dataframe["周期"].isin(
|
||
[
|
||
"M+01",
|
||
"M+02",
|
||
"M+03",
|
||
"M+04",
|
||
"M+05",
|
||
"M+06",
|
||
"M+07",
|
||
"M+08",
|
||
"M+09",
|
||
"M+10",
|
||
"M+11",
|
||
"M+12",
|
||
]
|
||
)
|
||
]
|
||
.reset_index(drop=True)
|
||
)
|
||
|
||
# 生成近十二个自然月留存率趋势
|
||
draw.line(
|
||
dataframe=report_forward[["客户分类", "周期", "留存率"]],
|
||
file_name="近十二个自然月留存率趋势.html",
|
||
)
|
||
|
||
# 获取报告模版
|
||
template = Environment(loader=FileSystemLoader(".")).get_template("template.html")
|
||
|
||
# 渲染模版
|
||
rfm_report = template.render(
|
||
{
|
||
# 报告日期
|
||
"report_date": datetime.now().strftime("%Y-%m-%d"),
|
||
"sample_size": sample_size,
|
||
}
|
||
)
|
||
|
||
with open("rfm_report.html", "w", encoding="utf8") as file:
|
||
file.write(rfm_report)
|
||
|
||
print("已完成")
|