一些聚类算法的实现

1. 感知器算法

import numpy as np

class Data():
    POINT = [[0,2],[1,1],[2,0],[0,-2],[-1,-1],[-2,0]]
    C = 1
    W = []
    times = 0
    # Polength代表一共有几组数据
    Polength = 2
    Wlength = len(POINT[0])

def change_point():
    for i in range(0,Data.Polength):
            Data.POINT[i].append(Data.C)
def w_ready():
    # 这里3后期可以进行调整
    for i in range(0,Data.Polength):
        Data.W.append([0,0,0])

def sub_list(a,b):
    for i in range(0,len(a)):
        a[i] = a[i] - b[i]
    return a
def add_list(a,b):
    for i in range(0,len(a)):
        a[i] = a[i] + b[i]
    return a

def iteration():
    for i in range(0, Data.Polength):
        d_result = []
        w_ready_update = []
        Data.times += 1
        for w in Data.W:
            d_result.append(np.dot(Data.POINT[i],w))
        if np.argsort(d_result)[len(d_result) - 1] == i:
            print(Data.W)
        else:
            for m in range(0,Data.Polength):
                if i == m:
                    pass
                else:
                    if d_result[i] <= d_result[m]:
                        w_ready_update.append(m)
        if len(w_ready_update) != 0:
            for n in w_ready_update:
                Data.W[n] = sub_list(Data.W[n],Data.C * Data.POINT[i])
            Data.W[i] = add_list(Data.W[i],Data.C * Data.POINT[i])
            print('目前迭代' + str(Data.times) + '次')
            print('此次有变化')
            print(Data.W)
        # print('目前迭代' + str(Data.times) + '次')
        # print('此次没有变化')
        # print(Data.W)


    if len(w_ready_update) == 0:
        print('一共迭代' + str(Data.times) + '次')
        print(Data.W)
    else:
        print('目前迭代' + str(Data.times) + '次')
        print(Data.W)
        iteration()
if __name__=="__main__":
    change_point()
    w_ready()
    iteration()

2.最近邻规则的简单实验法

import numpy as np
def distance(x,y):
    sum = 0
    for i in range(0,len(x)):
        dtemp = x[i] - y[i]
        sum += pow(dtemp,2)
    return np.sqrt(sum)
'''
按最近邻规则的简单实验法
利用一个大列表list里面嵌套多个小列表llist
每个llist中的第一个数据是聚类中心
'''
# X代表数据集，T代表设定的阈值
def julei1(x,T,list):
    for z in list:
        if distance(x,z[0]) < T:
            z.append(x)
        else:
            llist = []
            llist.append(x)
            list.append(llist)
            break


number = [(0,0),(3,8),(2,2),(1,1),(5,3),(4,18),(6,3),(5,4),(6,4),(7,5)]
list = []
llist = [number[0]]
list.append(llist)
for x in number[1:]:
    julei1(x,2,list)
    print(list)

3.最大最小距离算法

import numpy as np
def distance(x,y):
    sum = 0
    for i in range(0,len(x)):
        dtemp = x[i] - y[i]
        sum += pow(dtemp,2)
    return np.sqrt(sum)
'''
按最近邻规则的简单实验法
利用一个大列表list里面嵌套多个小列表llist
每个llist中的第一个数据是聚类中心
'''
# X代表数据集，T代表设定的阈值
def julei1(x,T,list):
    for z in list:
        if distance(x,z[0]) < T:
            z.append(x)
        else:
            llist = []
            llist.append(x)
            list.append(llist)
            break


number = [(0,0),(3,8),(2,2),(1,1),(5,3),(4,18),(6,3),(5,4),(6,4),(7,5)]
list = []
llist = [number[0]]
list.append(llist)
for x in number[1:]:
    julei1(x,2,list)
    print(list)

4.Kmean算法

#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
@File    :   k-means clustering algorithm.py    
@Contact :   17751019280@163.com

@Modify Time          
------------          
2020/11/7 2:22 下午       
@Author    @Version    @Desciption
-------    --------    -----------
SunBoyan      3.0         None
'''

import numpy as np
import random
# 把常用数据放到一起
class data():
    # 基本数据集
    number = [(0, 0), (3, 8), (2, 2), (1, 1), (5, 3), (4, 8), (6, 3), (5, 4), (6, 4), (7, 5)]
    # clusteResult聚类合集
    clusterResult = []
    clusterPoint = []
    # 随机选取聚类中心,i为个数
    for i in range(0,3):
        clusterPoint.append(number[i])
        clusterResult.append([])
def distance(x,y):
    sum = 0
    for i in range(0,len(x)):
        dtemp = x[i] - y[i]
        sum += pow(dtemp,2)
    return np.sqrt(sum)
def getAverage():
    sign = 0
    i = 0
    for li in data.clusterResult:
        sumx = 0
        sumy = 0
        for num in li:
            sumx += num[0]
            sumy += num[1]
        if data.clusterPoint[i] == (sumx / len(li),sumy / len(li)):
            sign += 1
        else:
            data.clusterPoint[i] = (sumx / len(li),sumy / len(li))
        i += 1
    #         3是手动确定要分的集合有几个
    if sign == 3:
        return 0
    else:
        return 1
#  每次迭代前对列表进行更新
def clearlist():
    data.clusterResult.clear()
    for i in range(0,len(data.clusterPoint)):
        data.clusterResult.append([])
def clusterByDis():
    clearlist()
    for j in data.number:
        tempdis = 99999999
        for i in range(3):
            if tempdis > distance(data.clusterPoint[i],j):
                tempdis = distance(data.clusterPoint[i],j)
                tempweb = i
        data.clusterResult[tempweb].append(j)
    if getAverage() == 0:
        print(data.clusterResult)
    else:
        clusterByDis()


clusterByDis()

5.系统聚类法算法

import numpy as np
import numpy as np

def distance(x,y):
    sum = 0
    for i in range(0,len(x)):
        dtemp = x[i] - y[i]
        sum += pow(dtemp,2)
    return np.sqrt(sum)
class data():
    # number = [(0,3,1,2,0),(1,3,0,1,0),(3,3,0,0,1),(1,1,0,2,0),(3,2,1,2,1),(4,1,1,1,0)]
    number = [(0, 0), (3, 8), (2, 2), (1, 1), (5, 3), (4, 8), (6, 3), (5, 4), (6, 4), (7, 5)]
    # number = [(0, 0), (3, 8), (2, 2), (1, 1), (5, 3), (4, 8)]
    length = len(number)
    X = np.zeros((length,length))
    # 计算合并次数
    times = 0
    #存储聚类的点
    clusterResult = {}
    #存储已经聚类的点
    alredyCluster = []

# 获取最小值位置
def getmin():
    min = 99999
    minwebx = 0
    minweby = 0
    for i in range(0,len(data.X[0])):
        for j in range(0,len(data.X[0])):
            if min > data.X[i][j] and data.X[i][j] != 0:
                min = data.X[i][j]
                minwebx = i
                minweby = j
    # np.unravel_index获取索引值在多维数组的位置，第一个是目标值，第二个是目标矩阵
    # print(np.unravel_index(min,data.X.shape))不能用浮点数做索引
    # pos = np.unravel_index(min,data.X.shape)
    print('输出最小值位置：'+str(minwebx) + ' ' + str(minweby))
    pos = [minwebx,minweby]
    return pos
def makeNumMatrix(temppos):
    tempal1 = 0
    tempal2 = 0
    for i in data.alredyCluster:
        if i <= temppos[0]:
            tempal1 += 1
        if i <= temppos[1]:
            tempal2 += 1
    temppos[0] -= tempal1
    temppos[1] -= tempal2
    return temppos
# 第一次判断两列最小并覆盖
def firJudgeMinForTwo(pos):
    # 获取合并后的矩阵
    li = []
    data.times += 1
    length = data.length - data.times
    temp = np.array((length,2))
    temp = data.X[: , [pos[0],pos[1]]]
    li = temp.min(axis = 1)
    li = np.delete(li,max(pos[0],pos[1]),0)
    # 整合合并后的矩阵
    data.X = np.delete(data.X,[pos[0],pos[1]],1)
    data.X = np.delete(data.X, [pos[0], pos[1]], 0)
    temp = np.zeros(length - 1)
    data.X = np.insert(data.X,min(pos[0],pos[1]),values=temp,axis=0)
    temp = np.zeros(length)
    data.X = np.insert(data.X, min(pos[0], pos[1]), values=temp, axis=1)
    data.X[: , min(pos[0],pos[1])] = li
    data.X[min(pos[0],pos[1]) , :] = li

    # 对number列表调整做准备
    # 这里不定义一个新的会导致浅复制，即一个变化另一个随着变化
    '''
    本意是使A = B，B为一个列表，结果在后续对A的操作中，导致B中的值也改变了，才回忆起python中均为“引用”这话。
    为避免这种情况，对这篇博文总结有两种情况可避免。
    1. 利用列表切片 A = B[:]
    2 . 深拷贝 A = copy.deepcopy(B)
    '''
    print('输出即将删除的数据：')

    #将聚类的点放到一起
    if data.times == 1:
        print(data.number[pos[0]], data.number[pos[1]])
        llist = [data.number[pos[0]],data.number[pos[1]]]
        data.clusterResult[min(pos[0],pos[1])] = llist
        # data.alredyCluster.append(max(pos[0],pos[1]))
        little = max(pos[0],pos[1])
        data.number.pop(little)
    else:
        # 遍历时不能修改字典元素
        # list是函数名，慎用
        sign = 0
        for k1 in list(data.clusterResult.keys()):
            if data.clusterResult[k1][0] == data.number[pos[0]]:
                for k2 in list(data.clusterResult.keys()):
                    if data.clusterResult[k2][0] == data.number[pos[1]]:
                        for i in data.clusterResult[max(k1,k2)]:
                            data.clusterResult[min(k1,k2)].append(i)
                        data.number.remove(data.clusterResult[max(k1, k2)][0])
                        data.clusterResult.pop(max(k1,k2))
                        sign += 2
                        break
                if sign == 0:
                    print(data.number[pos[1]])
                    data.clusterResult[k1].append(data.number[pos[1]])
                    sign += 1
                    data.number.pop(pos[1])
                    break
        if sign == 0:
            for k1 in list(data.clusterResult.keys()):
                if data.clusterResult[k1][0] == data.number[pos[1]]:
                    print(data.number[pos[0]])
                    data.clusterResult[k1].append(data.number[pos[0]])
                    sign += 1
                    data.number.pop(pos[0])
                    break
        if sign == 0:
            print(data.number[pos[0]], data.number[pos[1]])
            llist = [data.number[pos[0]], data.number[pos[1]]]
            data.clusterResult[min(pos[0], pos[1])] = llist
            data.number.pop(max(pos[0], pos[1]))
#     开始调整
    print('输出已有的字典：')
    print(data.clusterResult)
    print('输出删除后的number数据')
    print(data.number)

# 生成初始矩阵
def getMatrix():
    list = []
    # 求值
    for x in data.number:
        llist = []
        for y in range(0, len(data.number)):
            llist.append(distance(x, data.number[y]))
        list.append(llist)
    #     将数据存入矩阵中
    for temp in range(0, len(data.number)):
        data.X[:, temp] = list[temp]

def hierarchical_cluster():
    # 理想值设定，递归跳出条件
    if data.times== 999 or len(data.number) == 3:
        return 0
    firJudgeMinForTwo(getmin())

    hierarchical_cluster()

# 主函数
def Main():
    getMatrix()
    hierarchical_cluster()
    print(data.clusterResult)


Main()