Apriori算法python实现（含mlxend实现+手动实现）

开发环境

• 基础环境：anaconda3

• 开发环境

  jupyter lab
  

  使用教程：https://blog.csdn.net/tangyi2008/article/details/123761210

• 安装mlxtend包

  pip install mlxtend -i https://pypi.tuna.tsinghua.edu.cn/simple/
  

数据集

购物篮数据集：data/basket.txt
网盘链接：https://pan.baidu.com/s/1SbDJXwGCoCUM4JdU_NTjAQ?pwd=jiau
提取码：jiau

一、使用mlxend进行关联分析

1. 导入数据并进行预处理

import pandas as pd
import numpy as np
from mlxtend.preprocessing import TransactionEncoder

with open('data/basket.txt') as fp:
    products = [line.strip().split('\t') for line in fp]
onehot_enc = TransactionEncoder()
tmp_array = onehot_enc.fit_transform(products)
products_ = pd.DataFrame(tmp_array,columns=onehot_enc.columns_)
products_.head(5)

2. 寻找频繁项集

from mlxtend.frequent_patterns import apriori

freq_items = apriori(products_, min_support=0.1, 
                            use_colnames=True, max_len=None)

#计算项集的长度
freq_items['length'] = freq_items['itemsets'].apply(lambda x: len(x))

freq_items.head(5)

3. 寻找关联规则

from mlxtend.frequent_patterns import association_rules

rules = association_rules(freq_items, metric='lift', min_threshold=1)
# 选取提升度大于1且置信度大于0.5的关联规则
rules[(rules['lift'] > 1) & (rules['confidence'] > 0.5)]

#可以尝试其他的度量指标

度量指标:

• support(A->C) = support(A+C) [aka ‘support’], range: [0, 1]

• confidence(A->C) = support(A+C) / support(A), range: [0, 1]

• lift(A->C) = confidence(A->C) / support©, range: [0, inf]

• leverage(A->C) = support(A->C) - support(A)*support©,
  range: [-1, 1]

• conviction = [1 - support©] / [1 - confidence(A->C)],
  range: [0, inf]

• zhangs_metric(A->C) =
  leverage(A->C) / max(support(A->C)(1-support(A)), support(A)(support©-support(A->C)))
  range: [-1,1]

二、手动代码实现数据预处理

import pandas as pd
def my_transaction_encoder(file):
    tmp_items = []
    with open(file) as fp:
        tmp_items = [{item:True for item in line.strip().split('\t')} for line in fp]
    return pd.DataFrame(tmp_items).fillna(False)

调用测试

my_products = my_transaction_encoder('data/basket.txt')
my_products

三、手动代码实现apriori算法

1. 实现候选集生成函数

自定义函数，实现k项集 -> k+1项集

def gen_candidate_sets(x):
    x = list(map(sorted, x))
    r = []
    for i in range(len(x)):
        for j in range(i + 1, len(x)):
            if x[i][:-1] == x[j][:-1] and x[i][-1] != x[j][-1]:
                r.append(tuple(x[i][:-1] + sorted([x[j][-1], x[i][-1]])))
    return r

2. 实现频繁项集查找

import numpy as np
#寻找频繁项集
def my_find_items(df, min_support):
    # 单项集支持度筛选
    support_dict = {(k,):v for k,v in (df.sum()/len(df)).to_dict().items() if v > min_support}
    items = list(support_dict.keys())   
    #第几次遍历
    cn = 0

    while len(items) > 1:  
        cn += 1
        print(f"\n正在进行第{cn}次搜索..." )
        items = gen_candidate_sets(items)  #生产k+1项候选集
        print(f"数目{len(items)}...")

        # 新的支持度函数
        #计算k+1项集(即对应k+1列)的连乘
        support_dict_tmp = {cols : df[list(cols)].prod(axis=1, numeric_only=True).sum()/len(df)
                                     for cols in items}
        #筛选支持度，获取k+1项频繁项集
        items = [k for k,v in support_dict_tmp.items() if v > min_support]

        #更新支持度字典
        support_dict.update(support_dict_tmp)
        
    return support_dict

调用测试

items = my_find_items(my_products, 0.1)
items

3. 实现关联规则查找

def my_find_rule(support_dict, confidence):
    result = {}  
            
    for k,v in support_dict.items():
        if len(k) < 2:
            continue
        #遍历每一种可能的规则
        for i in range(len(k)):
            cond = k[:i] + k[i+1:]
            r = k[i]
            tmp_cofidence = support_dict[k]/support_dict[cond]
            if tmp_cofidence > confidence:
                result[f'{cond} => {r}'] = dict(confidence = tmp_cofidence,
                        support = support_dict[k])
  
    sorted(result.items(), key=lambda x:( x[1]["confidence"], x[1]["support"]), reverse=True)
    return result

调用测试

my_find_rule(items, 0.5)