k-mean算法实现
k-means 算法的工作过程说明如下：
初始化：聚类数k，初始聚类中心x,迭代次数或者收敛条件。
首先,从n个数据对象任意选择 k 个对象作为初始聚类中心；而对于所剩下其它对象，则根据它们与这些聚类中心的相似度（距离），分别将它们分配给与其最相似的（聚类中心所代表的）聚类；
然后,再计算每个所获新聚类的聚类中心（该聚类中所有对象的均值）；
再次，不断重复上面的过程直到满足收敛条件或者迭代次数为止.
目标：各聚类本身尽可能的紧凑，而各聚类之间尽可能的分开.
#################################################
import random
import time
t1 = time.time()
dataset = []
with open('iries.txt', 'r') as f:
for line in f:
ds = tuple([ float(x) for x in line.split() ])
dataset.append(ds)
#for x in dataset:
# print x[0], x[1], x[2], x[3]
#
def distance(x, y):
z = 0
for i in range(4):
z = z + pow((x[i] - y[i]), 2)
return pow(z, 0.5)
#print distance(dataset[1], dataset[2])
def get_means(ds):
a = b = c = d = 0.0
l = len(ds)
for x in ds:
a += x[0]
b += x[1]
c += x[2]
d += x[3]
return (a/l, b/l, c/l , d/l)
def gettotaldistance(clusters):
'''
[[(1,2,3,4), [(2,3,4,5),(3,4,5,6)...]], [(6,7,8,9),[...]],...]
'''
sum = 0
for cluster in clusters:
for point in cluster[1]:
sum += distance(cluster[0], point)
return sum
def push_into_cluster(clusters, point):
c = 0
d = distance(clusters[0][0], point)
for x in range(1, len(clusters)):
temp = distance(clusters[x][0], point)
if temp < d:
d = temp
c = x
clusters[c][1].append(point)
def kmeans(k = 3):
rn_center = random.sample(dataset, k)
clusters = []
for x in rn_center:
clusters.append([x, []])
print clusters
oldtotal = 999999999
while True:
for point in dataset:
push_into_cluster(clusters, point)
for cluster in clusters:
cluster[0] = get_means(cluster[1])
newtotal = gettotaldistance(clusters)
if oldtotal - newtotal > 1:
oldtotal = newtotal
for cluster in clusters:
cluster[1] = []
else:
print '============================'
for x in clusters:
print '-------------------------'
print x[0]
break
kmeans(k = 5)
print time.time() - t1
其中 iries.txt 来自 http://www.codeforge.cn/read/186226/irises.txt__html
下面是一个使用了pprocess模块来并行计算得k-means算法实现， 但不是太好，只有一点点用到了并行，对并行计算还不是太熟
import random
import time
import pprocess
t1 = time.time()
limit = 2 #core num
dataset = []
with open('iries.txt', 'r') as f:
for line in f:
ds = tuple([ float(x) for x in line.split() ])
dataset.append(ds)
#for x in dataset:
# print x[0], x[1], x[2], x[3]
#
def distance(x, y):
z = 0
for i in range(4):
z = z + pow((x[i] - y[i]), 2)
return pow(z, 0.5)
#print distance(dataset[1], dataset[2])
def get_means(ds):
a = b = c = d = 0.0
l = len(ds)
for x in ds:
a += x[0]
b += x[1]
c += x[2]
d += x[3]
return (a/l, b/l, c/l , d/l)
def gettotaldistance(clusters):
'''
[[(1,2,3,4), [(2,3,4,5),(3,4,5,6)...]], [(6,7,8,9),[...]],...]
'''
td = 0
results = pprocess.Map(limit = limit)
calc = results.manage(pprocess.MakeParallel(distance))
for cluster in clusters:
for point in cluster[1]:
calc(cluster[0], point)
td = sum(results)
print '------------------', results, td
return td
def push_into_cluster(clusters, point):
#results = pprocess.pmap(lambda x: distance(x, point), [ y[0] for y in clusters], limit = limit)
#print '#########', results
#minvalue = min(results)
#print minvalue
c = 0
d = distance(clusters[0][0], point)
for x in range(1, len(clusters)):
temp = distance(clusters[x][0], point)
if temp < d:
d = temp
c = x
clusters[c][1].append(point)
def kmeans(k = 3, eps = 1):
rn_center = random.sample(dataset, k)
clusters = []
for x in rn_center:
clusters.append([x, []])
print clusters
oldtotal = 999999999
while True:
for point in dataset:
push_into_cluster(clusters, point)
for cluster in clusters:
cluster[0] = get_means(cluster[1])
newtotal = gettotaldistance(clusters)
if oldtotal - newtotal > eps:
oldtotal = newtotal
for cluster in clusters:
cluster[1] = []
else:
print '============================'
for x in clusters:
print '-------------------------'
print x[0]
break
kmeans(k = 5, eps = 0.5)
print time.time() - t1