数据挖掘 —— 无监督学习（关联）

Apriori算法

代码

from itertools import combinations
def comb(lst):
    ret=[]
    for i in range(1,len(lst)+1):
        ret+=list(combinations(lst,i))
    return ret
class AprLayer(object):
    d=dict()
    def __init__(self):
        self.d=dict()
class AprNode(object):
    def __init__(self,node):
        self.s=set(node)
        self.size=len(self.s)
        self.lnk_nodes=dict()
        self.num=0
    def __hash__(self):
        return hash("__".join(sorted([str(itm) for itm in list(self.s)])))
    def __eq__(self, other):
        if "__".join(sorted([str(itm) for itm in list(self.s)]))=="__".join(sorted([str(itm) for itm in list(other.s)])):
            return True
        return False
    def isSubnode(self,node):
        return self.s.issubset(node.s)
    def incNum(self,num=1):
        self.num+=num
    def addLnk(self,node):
        self.lnk_nodes[node]=node.s

class AprBlk():
    def __init__(self,data):
        cnt=0
        self.apr_layers = dict()
        self.data_num=len(data)
        for datum in data:
            cnt+=1
            datum=comb(datum)
            nodes=[AprNode(da) for da in datum]
            for node in nodes:
                if not node.size in self.apr_layers:
                    self.apr_layers[node.size]=AprLayer()
                if not node in self.apr_layers[node.size].d:
                    self.apr_layers[node.size].d[node]=node
                self.apr_layers[node.size].d[node].incNum()
            for node in nodes:
                if node.size==1:
                    continue
                for sn in node.s:
                    sub_n=AprNode(node.s-set([sn]))
                    self.apr_layers[node.size-1].d[sub_n].addLnk(node)

    def getFreqItems(self,thd=1,hd=1):
        freq_items=[]
        for layer in self.apr_layers:
            for node in self.apr_layers[layer].d:
                if self.apr_layers[layer].d[node].num<thd:
                    continue
                freq_items.append((self.apr_layers[layer].d[node].s,self.apr_layers[layer].d[node].num))
        freq_items.sort(key=lambda x:x[1],reverse = True)
        return freq_items[:hd]

    def getConf(self,low=True, h_thd=10, l_thd=1, hd=1):
        confidence = []
        for layer in self.apr_layers:
            for node in self.apr_layers[layer].d:
                if self.apr_layers[layer].d[node].num < h_thd:
                    continue
                for lnk_node in node.lnk_nodes:
                    if lnk_node.num < l_thd:
                        continue
                    conf = float(lnk_node.num) / float(node.num)
                    confidence.append([node.s, node.num, lnk_node.s, lnk_node.num, conf])

        confidence.sort(key=lambda x: x[4])
        if low:
            return confidence[:hd]
        else:
            return confidence[-hd::-1]

class AssctAnaClass():
    def fit(self,data):
        self.apr_blk=AprBlk(data)
        return self
    def get_freq(self,thd=1,hd=1):
        return self.apr_blk.getFreqItems(thd=thd,hd=hd)
    def get_conf_high(self,thd,h_thd=10):
        return self.apr_blk.getConf(low=False, h_thd=h_thd, l_thd=thd)
    def get_conf_low(self,thd,hd,l_thd=1):
        return self.apr_blk.getConf(h_thd=thd,l_thd=l_thd,hd=hd)


def main():
    data=[
        ["牛奶","啤酒","尿布"],
        ["牛奶","啤酒","咖啡","尿布"],
        ["香肠","牛奶","饼干"],
        ["尿布","果汁","啤酒"],
        ["钉子","啤酒"],
        ["尿布","毛巾","香肠"],
        ["啤酒","毛巾","尿布","饼干"]
    ]
    print("Freq",AssctAnaClass().fit(data).get_freq(thd=3,hd=10))
    print("Conf",AssctAnaClass().fit(data).get_conf_high(thd=3,h_thd=3))
if __name__=="__main__":
    main()

结果

Freq [({'啤酒'}, 5), ({'尿布'}, 5), ({'啤酒', '尿布'}, 4), ({'牛奶'}, 3)]
Conf [[{'尿布'}, 5, {'啤酒', '尿布'}, 4, 0.8], [{'啤酒'}, 5, {'啤酒', '尿布'}, 4, 0.8]]

by CyrusMay 2022 04 05