python实现层次聚类

1.层次聚类的简单介绍

层次聚类方法对给定的数据集进行层次的分解，直到某种条件满足或者达到最大迭代次数。具体又可分为：
凝聚的层次聚类（AGNES算法）：一种自底向上的策略，首先将每个对象作为一个簇，然后合并这些原子簇为越来越大的簇（一般是计算所有簇的中心之间的距离，选取距离最小的两个簇合并），直到某个终结条件被满足或者达到最大迭代次数。
分裂的层次聚类（DIANA算法）：采用自顶向下的策略，它首先将所有对象置于一个簇中，然后逐渐细分为越来越小的簇（一般是每次迭代分裂一个簇为两个），直到达到了某个终结条件或者达到最大迭代次数。



2.凝聚型层次聚类的代码如下：

本例中合并之后会把两个簇的中心求平均值，然后当成新的簇的中心。而且计算量比较大。可以说本例是最最最朴素的层次聚类的一种python实现。谁有好的方法，欢迎一起讨论！

代码如下：

import sys, os

import math

class Hierarchical:

def __init__(self, center, left = None, right = None, flag = None, distance = 0.0):

self.center = center

self.left = left

self.right = right

self.flag = flag

self.distance = distance

def traverse(node):

if node.left == None and node.right == None:

return [node.center]

else:

return traverse(node.left) + traverse(node.right)

def distance(v1, v2):

if len(v1) != len(v2):

print sys.stderr, "invalid v1 and v2 !"

sys.exit(1)

distance = 0

for i in range(len(v1)):

distance += (v1[i] - v2[i]) ** 2

distance = math.sqrt(distance)

return distance

def hcluster(data, n):

if len(data) <= 0:

print sys.stderr, "invalid data"

sys.exit(1)

clusters = [Hierarchical(data[i], flag = i) for i in range(len(data))]

distances = {}

min_id1 = None

min_id2 = None

currentCluster = -100

while(len(clusters) > n):

minDist = 1000000000000

for i in range(len(clusters) - 1):

for j in range(i + 1, len(clusters)):

# save distance, pick up speed

if distances.get((clusters[i].flag, clusters[j].flag)) == None:

distances[(clusters[i].flag, clusters[j].flag)] = distance(clusters[i].center, clusters[j].center)

if distances[(clusters[i].flag, clusters[j].flag)] <= minDist:

min_id1 = i

min_id2 = j

minDist = distances[(clusters[i].flag, clusters[j].flag)]

if min_id1 != None and min_id2 != None and minDist != 1000000000000:

newCenter = [(clusters[min_id1].center[i] + clusters[min_id2].center[i])/2 for i in range(len(clusters[min_id2].center))]

newFlag = currentCluster

currentCluster -= 1

newCluster = Hierarchical(newCenter, clusters[min_id1], clusters[min_id2], newFlag, minDist)

del clusters[min_id2]

del clusters[min_id1]

clusters.append(newCluster)

finalCluster = [traverse(clusters[i]) for i in range(len(clusters))]

return finalCluster

def loadData(filename):

infile = open(filename, 'r')

line = infile.readline()

dataList = []

tempList = []

while line:

lineArr = line.strip().split()

if len(lineArr) < 2:

line = infile.readline()

continue

for i in range(len(lineArr)):

tempList.append(float(lineArr[i]))

dataList.append(tempList)

tempList = []

line = infile.readline()

return dataList

if __name__ == '__main__':

data = [[123,321,434,4325,345345],[23124,141241,434234,9837489,34743],\

[128937,127,12381,424,8945],[322,4348,5040,8189,2348],\

[51249,42190,2713,2319,4328],[13957,1871829,8712847,34589,30945],\

[1234,45094,23409,13495,348052],[49853,3847,4728,4059,5389]]

finalCluster = hcluster(data, 3)

print finalCluster

data = loadData("testSet2.txt")

finalCluster = hcluster(data, 3)

print finalCluster

参考博客：http://www.cnblogs.com/coser/archive/2013/04/10/3013044.html。，然后根据自己的理解实现了一遍。