1 预备项目

关联规则分析实践———购物车分析

import warnings
import numpy as np
import pandas as pd
from mlxtend.frequent_patterns import apriori
from mlxtend.frequent_patterns import association_rules
warnings.filterwarnings("ignore", category=DeprecationWarning)
# 读入数据
df_Retails = pd.read_excel('input/Online Retail.xlsx')
df_Retails.head()

# 数据理解
print(df_Retails.shape)

df_Retails.columns

df_Retails.describe()

#查看国家一列的取值
df_Retails.Country.unique()
#各国家的购物数量
df_Retails["Country"].value_counts()
#查看InvoiceNo一列中是否有重复的值
df_Retails.duplicated(subset=["InvoiceNo"]).any()
#是否有缺失值
df_Retails.isna().sum()
df_Retails['InvoiceNo'].isnull().sum(axis = 0)

#空格处理
df_Retails['Description'] = df_Retails['Description'].str.strip()
# Description: Product (item) name. Nominal.
#查看是否有缺失值
df_Retails['Description'].isna().sum()
#缺失值处理
df_Retails.dropna(axis=0
                  , subset=['Description']
                  , inplace=True)
print(df_Retails.shape)
#查看是否有缺失值
print(df_Retails['Description'].isna().sum())
#删除含有C字母的已取消订单
df_Retails['InvoiceNo'] = df_Retails['InvoiceNo'].astype('str')

df_Retails = df_Retails[~df_Retails['InvoiceNo'].str.contains('C')]
df_Retails.shape

#将数据改为每一行一条购物记录
#考虑到内存限制只计算Germany，全部计算则计算量太大
df_ShoppingCarts = (df_Retails[df_Retails['Country'] =="Germany"]
                    .groupby(['InvoiceNo', 'Description'])['Quantity']
                    .sum()
                    .unstack()
                    .reset_index()
                    .fillna(0)
                    .set_index('InvoiceNo'))

print(df_ShoppingCarts.shape)

df_ShoppingCarts.head()

#查看InvoiceNo一列中是否有重复的值
df_Retails.duplicated(subset=["InvoiceNo"]).any()

def encode_units(x):
    if x <= 0:
        return 0
    if x >= 1:
        return 1

df_ShoppingCarts_sets = df_ShoppingCarts.map(encode_units)

df_ShoppingCarts_sets.head()

1 产生频繁集

# 产生频繁集 最小支持度为0.07, 在输出中使用原始列名
df_Frequent_Itemsets = apriori(df_ShoppingCarts_sets
                               , min_support=0.07
                               , use_colnames=True)
df_Frequent_Itemsets

2 产生关联规则

# 生成关联规则，使用提升度（lift）作为度量 置提升度的最小阈值为 1，表示无正相关关系的规则也会被计算
df_AssociationRules = association_rules(df_Frequent_Itemsets
                                        , metric="lift"
                                        , min_threshold=1)
#输出结果的解读：https://rasbt.github.io/mlxtend/user_guide/frequent_patterns/association_rules/
df_AssociationRules.head()

# 筛选关联规则 筛选提升度不小于 2 置信度不小于 0.8 的关联规则
df_A= df_AssociationRules[(df_AssociationRules['lift'] >= 2) &
                          (df_AssociationRules['confidence'] >= 0.8) ]
df_A

# 可视化结果
import seaborn as sns
import matplotlib.pyplot as plt

sns.scatterplot(x = "support"
                , y = "confidence"
                , size = "lift"
                , data = df_AssociationRules)
plt.show()