apriori算法的学习

Apriori算法是最经典的挖掘频繁项集的算法，其核心思想是通过连接产生候选项与其支持度，然后通过剪枝生成频繁项集。

几个重要的概念：

(1) 关联规则的支持度（A,B同时发生的概率）

(2) 关联规则的置信度（A发生的前提下，B发生的概率）

(3) 最小支持度：是定义衡量支持度的一个阈值，表示项集在统计意义上的最低重要性

(4) 最小置信度：是定义衡量置信度的一个阈值，表示关联规则的最低可靠性

同时满足最小支持度阈值和最小置信度阈值的规则称为强规则。

(5) 项集

       项集是项的称为集合，包含K个项的集合K项集，如集合｛牛奶，麦片，糖｝是一个3项集。

       项集的出现频率是所有包含项集的事物计数，又称为绝对支持度或支持度计数。如果项集I的相对支持度满足预定义的最小支持度阈值，则I是频繁项集。频繁k项集通常记作Ik。

matlab代码如下：

Cal_apriori.m：

%% 使用Apriori算法挖掘菜品订单关联规则
clear;
% 参数初始化
inputfile = '../data/menu_orders.txt'; % 销量及其他属性数据
outputfile='../tmp/as.txt';% 输出转换后0,1矩阵文件
minSup = 0.2; % 最小支持度
minConf = 0.5;% 最小置信度
nRules = 1000;% 输出最大规则数
sortFlag = 1;% 按照支持度排序
rulefile = '../tmp/rules.txt'; % 规则输出文件
 
%% 调用转换程序 ，把数据转换为0,1矩阵，自定义函数
[transactions,code] = trans2matrix(inputfile,outputfile,','); 
 
%% 调用Apriori关联规则算法，自定义函数
[Rules,FreqItemsets] = findRules(transactions, minSup, minConf, nRules, sortFlag, code, rulefile);
 
disp('Apriori算法挖掘菜品订单关联规则完成！');

trans2matrix.m

function [ output,code] = trans2matrix( inputfile,outputfile,splitter )
%% 把输入事务转换为0、1矩阵；每行代表一个事务
 
% 输入参数：
% inputfile：输入文件，空格分隔每个项目；
% outputfile：输出文件，转换后的0,1矩阵文件；
% splitter: 输入文件项目的间隔符,默认为空格
 
% 输出参数：
% output : 转换后的0,1 矩阵
% code：编码规则；
 
if nargin<3
   splitter=' '; 
end
 
%% 读入文件, 获得编码规则
code={};
fid= fopen(inputfile);
tline = fgetl(fid);
lines=0;
while ischar(tline)
    lines=lines+1; % 记录行数
    tline = deblank(tline);
    tline = regexp(tline,splitter,'split');
    code=[code tline]; % 合并 
    code = unique(code); % 去除重复记录
%     disp(code)
    tline = fgetl(fid);
end
disp('编码规则为：')
disp(num2str(1:size(code,2)))
disp( code);
fclose(fid); % 关闭文档
 
%% 读取文件，根据编码规则对原始数据进行转换
itemsnum= size(code,2);
output=zeros(lines,itemsnum);
fid= fopen(inputfile);
tline = fgetl(fid);
lines=0;
while ischar(tline)
    lines=lines+1; % 记录行数
    tline = deblank(tline);
    tline = regexp(tline,splitter,'split');
    [~,icode,~] = intersect(code,tline);% 寻找下标
    output(lines,icode')=1;
    %disp(output(lines,:))
    tline = fgetl(fid);
end
fclose(fid);
 
%% 把转换后的矩阵写入文件
fid = fopen(outputfile, 'w');
for i=1:lines
   fprintf(fid,'%s\n',num2str(output(i,:))); 
end
fclose(fid);
end

findRules.m

<span style="font-size:18px;">function [Rules,FreqItemsets] = findRules(transactions, minSup, minConf, nRules, sortFlag, code, rulesfile)
%
% This function performs Association Analysis (Apriori Algorithm):  Given a set of transactions,
% find rules that will predict  the occurrence of an item based on the occurrences of other
% items in the transaction
% 
% Rules are of the form  A-> B (e.g., {milk, diaper} - > {Coke}), where
% support = minSup (minimum support threshold)
% confidence = minConf (minimum confidence threshold)
% 
% Support is the fraction of transactions that contain both A and B:
% Support(A,B) = P(A,B)
% 
% Confidence is the fraction of transactions where items in B appear in transactions  that contain A:
% Confidence(A,B) = P(B|A)
%
%
% INPUT:
%          transactions:  M x N matrix of binary transactions, where each row
%                                  represents one transaction and each column represents
%                                  one attribute/item
%          minSup:          scalar value that represents the minimum
%                                  threshold for support for each rule
%          minConf:        scalar value that represents the minimum
%                                  threshold for confidence of each rule
%          nRules:           scalar value indicating the number of rules
%                                  the user wants to find
%          sortFlag:         binary value indicating if the rules should be
%                                  sorted by support level or confidence level
%                                  1: sort by rule support level
%                                  2: sort by rule confidence level
%          code (labels): 编码规则            optional parameter that provides labels for
%                                  each attribute (columns of transactions),
%                                  by default attributes are represented
%                                  with increasing numerical values 1:N
%           
%          fname:            optional file name where rules are saved
%
% OUTPUT:
%          Rules:             2 x 1 cell array, where the first cell (Rules{1}{:})
%                                 contains the itemsets in the left side of the rule and second
%                                 cell (Rules{2}{:}) contains the itemsets
%                                 in the right side of the rule (e.g., if
%                                 the first rule is {1, 2} -> 3,
%                                 Rules{1}{1} = [1,2], Rules{2}{1} = [3])
%         FreqItemsets: A cell array of frequent itemsets of size 1, 2,
%                                 etc., with itemset support >= minSup,
%                                 where FreqItemSets{1} represents itemsets
%                                 of size 1, FreqItemSets{2} itemsets of
%                                 size 2, etc.
%         fname.txt:      The code creates a text file and stores all the
%                                 rules in the form left_side -> right_side.
%
% author: Narine Manukyan 07/08/2013
 
% Number of transactions in the dataset
M = size(transactions,1);
% Number of attributes in the dataset
N = size(transactions,2);
 
if nargin < 7
    fname = 'default';
end
 
if nargin < 6
    labels = cellfun(@(x){num2str(x)}, num2cell(1:N));
end
 
if nargin < 5
    sortFlag = 1;
end
 
if nargin < 4
    nRules = 100;
end
 
if nargin < 3
    minConf = 0.5;
end
 
if nargin < 2
    minSup = 0.5;
end
 
if nargin == 0
    error('No input arguments were supplied.  At least one is expected.');
end
 
% Preallocate memory for Rules and FreqItemsets
maxSize = 10^2;
Rules = cell(2,1);
Rules{1} = cell(nRules,1);
Rules{2} = cell(nRules,1);
FreqItemsets = cell(maxSize);
RuleConf = zeros(nRules,1);
RuleSup = zeros(nRules,1);
ct = 1;
 
% Find frequent item sets of size one (list of all items with minSup)
T = [];
for i = 1:N
    S = sum(transactions(:,i))/M;
    if S >= minSup
        T = [T; i];
    end
end
FreqItemsets{1} = T;
 
%Find frequent item sets of size >=2 and from those identify rules with minConf
 
for steps = 2:N
    
    % If there aren't at least two items  with minSup terminate
    U = unique(T);
    if isempty(U) || size(U,1) == 1
        Rules{1}(ct:end) = [];
        Rules{2}(ct:end) = [];
        FreqItemsets(steps-1:end) = [];
        break
    end
    
    % Generate all combinations of items that are in frequent itemset
    Combinations = nchoosek(U',steps);
    TOld = T;
    T = [];
    
    for j = 1:size(Combinations,1)
        if ct > nRules
            break;
        else
            % Apriori rule: if any subset of items are not in frequent itemset do not
            % consider the superset (e.g., if {A, B} does not have minSup do not consider {A,B,*})
            if sum(ismember(nchoosek(Combinations(j,:),steps-1),TOld,'rows')) - steps+1>0
                
                % Calculate the support for the new itemset
                S = mean((sum(transactions(:,Combinations(j,:)),2)-steps)>=0);
                if S >= minSup
                    T = [T; Combinations(j,:)];
                    
                    % Generate potential rules and check for minConf
                    for depth = 1:steps-1
                        R = nchoosek(Combinations(j,:),depth);
                        for r = 1:size(R,1)
                            if ct > nRules
                                break;
                            else
                                % Calculate the confidence of the rule
                                Ctemp = S/mean((sum(transactions(:,R(r,:)),2)-depth)==0);
                                if Ctemp > minConf
                                    
                                    % Store the rules that have minSup and minConf
                                    Rules{1}{ct} = R(r,:);
                                    Rules{2}{ct} = setdiff(Combinations(j,:),R(r,:));
                                    RuleConf(ct) = Ctemp;
                                    RuleSup(ct) = S;
                                    ct = ct+1;
                                end
                            end
                        end
                    end
                end
            end
        end
    end
    
    % Store the freqent itemsets
    FreqItemsets{steps} = T;
end
 
% Get rid of unnecessary rows due to preallocation (helps with speed)
FreqItemsets(steps-1:end) = [];
RuleConf = RuleConf(1:ct-1);
RuleSup = RuleSup(1:ct-1);
 
% Sort the rules in descending order based on the confidence or support level
switch sortFlag
    case 1 % Sort by Support level
        [V ind] = sort(RuleSup,'descend');
    case 2 % Sort by Confidence level
        [V ind] = sort(RuleConf,'descend');
end
 
RuleConf = RuleConf(ind);
RuleSup = RuleSup(ind);
 
for i = 1:2
    temp = Rules{i,1};
    temp = temp(ind);
    Rules{i,1} = temp;
end
 
disp(['关联规则算法完成,规则数为：' num2str(size(RuleSup,1))]);
 
% Save the rule in a text file and print them on display
fid = fopen(rulesfile, 'w');
fprintf(fid, '%s   (%s, %s) \n', 'Rule', 'Support', 'Confidence');
 
for i = 1:size(Rules{1},1)
    s1 = '';
    s2 = '';
    for j = 1:size(Rules{1}{i},2)
        if j == size(Rules{1}{i},2)
            s1 = [s1 code{Rules{1}{i}(j)}];
        else
            s1 = [s1 code{Rules{1}{i}(j)} ','];
        end
    end
    for k = 1:size(Rules{2}{i},2)
        if k == size(Rules{2}{i},2)
            s2 = [s2 code{Rules{2}{i}(k)}];
        else
            s2 = [s2 code{Rules{2}{i}(k)} ','];
        end
    end
    s3 = num2str(RuleSup(i)*100);
    s4 = num2str(RuleConf(i)*100);
    fprintf(fid, '%s -> %s  (%s%%, %s%%)\n', s1, s2, s3, s4);
end
fclose(fid);
disp(['存储规则到文件‘' rulesfile '’完成'])
end</span>

 

运行的数据集：

a,c,e
b,d
b,c
a,b,c,d
a,b
b,c
a,b
a,b,c,e
a,b,c
a,c,e

运行结果：

Rule   (Support, Confidence) 
a -> b  (50%, 71.4286%)
b -> a  (50%, 62.5%)
a -> c  (50%, 71.4286%)
c -> a  (50%, 71.4286%)
b -> c  (50%, 62.5%)
c -> b  (50%, 71.4286%)
e -> a  (30%, 100%)
e -> c  (30%, 100%)
a,b -> c  (30%, 60%)
a,c -> b  (30%, 60%)
b,c -> a  (30%, 60%)
e -> a,c  (30%, 100%)
a,c -> e  (30%, 60%)
a,e -> c  (30%, 100%)
c,e -> a  (30%, 100%)
d -> b  (20%, 100%)

参考资料：MATLAB数据分析与挖掘实战