|
| 1 | +# coding:UTF-8 |
| 2 | +''' |
| 3 | +Date:20160923 |
| 4 | +@author: zhaozhiyong |
| 5 | +''' |
| 6 | +import numpy as np |
| 7 | + |
| 8 | +def load_data(file_path): |
| 9 | + '''导入数据 |
| 10 | + input: file_path(string):文件的存储位置 |
| 11 | + output: data(mat):数据 |
| 12 | + ''' |
| 13 | + f = open(file_path) |
| 14 | + data = [] |
| 15 | + for line in f.readlines(): |
| 16 | + row = [] # 记录每一行 |
| 17 | + lines = line.strip().split("\t") |
| 18 | + for x in lines: |
| 19 | + row.append(float(x)) # 将文本中的特征转换成浮点数 |
| 20 | + data.append(row) |
| 21 | + f.close() |
| 22 | + return np.mat(data) |
| 23 | + |
| 24 | +def distance(vecA, vecB): |
| 25 | + '''计算vecA与vecB之间的欧式距离的平方 |
| 26 | + input: vecA(mat)A点坐标 |
| 27 | + vecB(mat)B点坐标 |
| 28 | + output: dist[0, 0](float)A点与B点距离的平方 |
| 29 | + ''' |
| 30 | + dist = (vecA - vecB) * (vecA - vecB).T |
| 31 | + return dist[0, 0] |
| 32 | + |
| 33 | +def randCent(data, k): |
| 34 | + '''随机初始化聚类中心 |
| 35 | + input: data(mat):训练数据 |
| 36 | + k(int):类别个数 |
| 37 | + output: centroids(mat):聚类中心 |
| 38 | + ''' |
| 39 | + n = np.shape(data)[1] # 属性的个数 |
| 40 | + centroids = np.mat(np.zeros((k, n))) # 初始化k个聚类中心 |
| 41 | + for j in xrange(n): # 初始化聚类中心每一维的坐标 |
| 42 | + minJ = np.min(data[:, j]) |
| 43 | + rangeJ = np.max(data[:, j]) - minJ |
| 44 | + # 在最大值和最小值之间随机初始化 |
| 45 | + centroids[:, j] = minJ * np.mat(np.ones((k , 1))) \ |
| 46 | + + np.random.rand(k, 1) * rangeJ |
| 47 | + return centroids |
| 48 | + |
| 49 | +def kmeans(data, k, centroids): |
| 50 | + '''根据KMeans算法求解聚类中心 |
| 51 | + input: data(mat):训练数据 |
| 52 | + k(int):类别个数 |
| 53 | + centroids(mat):随机初始化的聚类中心 |
| 54 | + output: centroids(mat):训练完成的聚类中心 |
| 55 | + subCenter(mat):每一个样本所属的类别 |
| 56 | + ''' |
| 57 | + m, n = np.shape(data) # m:样本的个数,n:特征的维度 |
| 58 | + subCenter = np.mat(np.zeros((m, 2))) # 初始化每一个样本所属的类别 |
| 59 | + change = True # 判断是否需要重新计算聚类中心 |
| 60 | + while change == True: |
| 61 | + change = False # 重置 |
| 62 | + for i in xrange(m): |
| 63 | + minDist = np.inf # 设置样本与聚类中心之间的最小的距离,初始值为争取穷 |
| 64 | + minIndex = 0 # 所属的类别 |
| 65 | + for j in xrange(k): |
| 66 | + # 计算i和每个聚类中心之间的距离 |
| 67 | + dist = distance(data[i, ], centroids[j, ]) |
| 68 | + if dist < minDist: |
| 69 | + minDist = dist |
| 70 | + minIndex = j |
| 71 | + # 判断是否需要改变 |
| 72 | + if subCenter[i, 0] <> minIndex: # 需要改变 |
| 73 | + change = True |
| 74 | + subCenter[i, ] = np.mat([minIndex, minDist]) |
| 75 | + # 重新计算聚类中心 |
| 76 | + for j in xrange(k): |
| 77 | + sum_all = np.mat(np.zeros((1, n))) |
| 78 | + r = 0 # 每个类别中的样本的个数 |
| 79 | + for i in xrange(m): |
| 80 | + if subCenter[i, 0] == j: # 计算第j个类别 |
| 81 | + sum_all += data[i, ] |
| 82 | + r += 1 |
| 83 | + for z in xrange(n): |
| 84 | + try: |
| 85 | + centroids[j, z] = sum_all[0, z] / r |
| 86 | + except: |
| 87 | + print " r is zero" |
| 88 | + return subCenter |
| 89 | + |
| 90 | +def save_result(file_name, source): |
| 91 | + '''保存source中的结果到file_name文件中 |
| 92 | + input: file_name(string):文件名 |
| 93 | + source(mat):需要保存的数据 |
| 94 | + output: |
| 95 | + ''' |
| 96 | + m, n = np.shape(source) |
| 97 | + f = open(file_name, "w") |
| 98 | + for i in xrange(m): |
| 99 | + tmp = [] |
| 100 | + for j in xrange(n): |
| 101 | + tmp.append(str(source[i, j])) |
| 102 | + f.write("\t".join(tmp) + "\n") |
| 103 | + f.close() |
| 104 | + |
| 105 | +if __name__ == "__main__": |
| 106 | + k = 4 # 聚类中心的个数 |
| 107 | + file_path = "data.txt" |
| 108 | + # 1、导入数据 |
| 109 | + print "---------- 1.load data ------------" |
| 110 | + data = load_data(file_path) |
| 111 | + # 2、随机初始化k个聚类中心 |
| 112 | + print "---------- 2.random center ------------" |
| 113 | + centroids = randCent(data, k) |
| 114 | + # 3、聚类计算 |
| 115 | + print "---------- 3.kmeans ------------" |
| 116 | + subCenter = kmeans(data, k, centroids) |
| 117 | + # 4、保存所属的类别文件 |
| 118 | + print "---------- 4.save subCenter ------------" |
| 119 | + save_result("sub", subCenter) |
| 120 | + # 5、保存聚类中心 |
| 121 | + print "---------- 5.save centroids ------------" |
| 122 | + save_result("center", centroids) |
0 commit comments