FPgrowth用python3实现挖掘频繁项集

参考的是以下的博客：
https://blog.csdn.net/Gamer_gyt/article/details/51113753

输入：

    simpDat = [['r', 'z', 'h', 'j', 'p'],
               ['z', 'y', 'x', 'w', 'v', 'u', 't', 's'],
               ['z'],
               ['r', 'x', 'n', 'o', 's'],
               ['y', 'r', 'x', 'z', 'q', 't', 'p'],
               ['y', 'z', 'x', 'e', 'q', 's', 't', 'm']]

过程：

dataSet=simpDat
freqItems=fpGrowth(dataSet,4)  #4代表最小频繁度，即要找出出现4次或4次以上的频繁项集
freqItems

dataSet是输入，可按实际情况整理

输出：

[{'x'}, {'z'}]  #simDat中出现4次货以上的频繁项集

以下是代码，当成一个FPgrowth的黑盒子：

#FP树中节点的类定义
class treeNode:
    def __init__(self, nameValue, numOccur, parentNode):
        self.name = nameValue
        self.count = numOccur
        self.nodeLink = None #nodeLink 变量用于链接相似的元素项
        self.parent = parentNode #指向当前节点的父节点
        self.children = {} #空字典，存放节点的子节点

    def inc(self, numOccur):
        self.count += numOccur

#将树以文本形式显示
    def disp(self, ind=1):
        print ('  ' * ind, self.name, ' ', self.count)
        for child in self.children.values():
            child.disp(ind + 1)


#构建FP-tree
def createTree(dataSet, minSup=1):
    headerTable = {}
    for trans in dataSet:  #第一次遍历：统计各个数据的频繁度
        for item in trans:
            headerTable[item] = headerTable.get(item, 0) + dataSet[trans]
            #用头指针表统计各个类别的出现的次数，计算频繁量：头指针表[类别]=出现次数
    for  k in list(headerTable):  #删除未达到最小频繁度的数据
        if headerTable[k] < minSup:
            del (headerTable[k])
    freqItemSet = set(headerTable.keys())#保存达到要求的数据
    # print ('freqItemSet: ',freqItemSet)
    if len(freqItemSet) == 0:
        return None, None  #若达到要求的数目为0
    for k in headerTable: #遍历头指针表
        headerTable[k] = [headerTable[k], None]  #保存计数值及指向每种类型第一个元素项的指针
    # print ('headerTable: ',headerTable)
    retTree = treeNode('Null Set', 1, None)  #初始化tree
    for tranSet, count in dataSet.items():  # 第二次遍历：
        localD = {}
        for item in tranSet:  # put transaction items in order
            if item in freqItemSet:#只对频繁项集进行排序
                localD[item] = headerTable[item][0]

        #使用排序后的频率项集对树进行填充
        if len(localD) > 0:
            orderedItems = [v[0] for v in sorted(localD.items(), key=lambda p: p[1], reverse=True)]
            updateTree(orderedItems, retTree, headerTable, count)  # populate tree with ordered freq itemset
    return retTree, headerTable  #返回树和头指针表


def updateTree(items, inTree, headerTable, count):
    if items[0] in inTree.children:  # 首先检查是否存在该节点
        inTree.children[items[0]].inc(count)  # 存在则计数增加
    else:  # 不存在则将新建该节点
        inTree.children[items[0]] = treeNode(items[0], count, inTree)#创建一个新节点
        if headerTable[items[0]][1] == None:  # 若原来不存在该类别，更新头指针列表
            headerTable[items[0]][1] = inTree.children[items[0]]#更新指向
        else:#更新指向
            updateHeader(headerTable[items[0]][1], inTree.children[items[0]])
    if len(items) > 1:  #仍有未分配完的树，迭代
        updateTree(items[1::], inTree.children[items[0]], headerTable, count)

#节点链接指向树中该元素项的每一个实例。
# 从头指针表的 nodeLink 开始,一直沿着nodeLink直到到达链表末尾
def updateHeader(nodeToTest, targetNode):
    while (nodeToTest.nodeLink != None):
        nodeToTest = nodeToTest.nodeLink
    nodeToTest.nodeLink = targetNode


def loadSimpDat():
        simpDat = [['r', 'z', 'h', 'j', 'p'],
                   ['z', 'y', 'x', 'w', 'v', 'u', 't', 's'],
                   ['z'],
                   ['r', 'x', 'n', 'o', 's'],
                   ['y', 'r', 'x', 'z', 'q', 't', 'p'],
                   ['y', 'z', 'x', 'e', 'q', 's', 't', 'm']]
    return simpDat
#createInitSet() 用于实现上述从列表到字典的类型转换过程
def createInitSet(dataSet):
    retDict = {}
    for trans in dataSet:
        retDict[frozenset(trans)] = 1
    return retDict

#从FP树中发现频繁项集
def ascendTree(leafNode, prefixPath):  #递归上溯整棵树
    if leafNode.parent != None:
        prefixPath.append(leafNode.name)
        ascendTree(leafNode.parent, prefixPath)


def findPrefixPath(basePat, treeNode):  #参数：指针，节点；
    condPats = {}
    while treeNode != None:
        prefixPath = []
        ascendTree(treeNode, prefixPath)#寻找当前非空节点的前缀
        if len(prefixPath) > 1:
            condPats[frozenset(prefixPath[1:])] = treeNode.count #将条件模式基添加到字典中
        treeNode = treeNode.nodeLink
    return condPats

#递归查找频繁项集
def mineTree(inTree, headerTable, minSup, preFix, freqItemList):
    # 头指针表中的元素项按照频繁度排序,从小到大
    bigL = [v[0] for v in sorted(headerTable.items(), key=lambda p: str(p[1]))]#python3修改
    for basePat in bigL:  #从底层开始
        #加入频繁项列表
        newFreqSet = preFix.copy()
        newFreqSet.add(basePat)
        print ('finalFrequent Item: ',newFreqSet)
        freqItemList.append(newFreqSet)
        #递归调用函数来创建基
        condPattBases = findPrefixPath(basePat, headerTable[basePat][1])
        print ('condPattBases :',basePat, condPattBases)

        #2. 构建条件模式Tree
        myCondTree, myHead = createTree(condPattBases, minSup)
        #将创建的条件基作为新的数据集添加到fp-tree
        print ('head from conditional tree: ', myHead)
        if myHead != None: #3. 递归
            print ('conditional tree for: ',newFreqSet)
            myCondTree.disp(1)
            mineTree(myCondTree, myHead, minSup, newFreqSet, freqItemList)

def fpGrowth(dataSet, minSup=3):
    initSet = createInitSet(dataSet)
    myFPtree, myHeaderTab = createTree(initSet, minSup)
    freqItems = []
    mineTree(myFPtree, myHeaderTab, minSup, set([]), freqItems)
    return freqItems