-
Notifications
You must be signed in to change notification settings - Fork 0
/
BSKM_MODIFIED.py
279 lines (243 loc) · 9.94 KB
/
BSKM_MODIFIED.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
import numpy as np
import random
from matplotlib import pyplot
from scipy.spatial.distance import cosine
def normalize(x):
#标准化向量
#print(np.linalg.norm(a[0]))
norm = np.linalg.norm(x)
y = x/norm
return y
def normalize_dataSet(dataset):
for i in range(len(dataset)):
dataset[i] = normalize(dataset[i])
return dataset
def cos_similarity(x,y):
#计算余弦相似度
'''
print('x: ')
print(x.shape)
print('y: ')
print(y.shape)
num = float(np.matmul(x, y))
s = np.linalg.norm(x) * np.linalg.norm(y) #np.linalg.norm 默认是求整体矩阵元素平方和再开根号 即是模
if s == 0:
result = 0.0
else:
result = num/s
return (result)
'''
return float(np.matmul(x,y))
def cos_dot_procuct(x,y):
#print(np.matmul(x,y))
#return (np.matmul(x,y))
return cosine(x,y)
def labeling(x):
for i in range(len(x)):
x[i].append(int(i))
return np.array(x)
def cos_distance(x,y):
return 1-cos_dot_procuct(x,y)
class SKM:
def __init__(self,dataSet,k,m,epislon=0.0001):
super().__init__()
self.dataSet = normalize_dataSet(dataSet) #数据集
self.dataSet_labeled = labeling(self.dataSet.tolist())
#self.dataSet = dataSet
self.k = k #簇的个数
self.epislon = epislon #精度
self.m = m #迭代次数
self.size = np.shape(dataSet)[0] #粒子个数
#self.dim = np.shape(dataSet)[1] #维度
def setDataSet(self,dataSet):
self.dataSet = dataSet
def initCenters(self):
#print(np.max(np.max(a,axis=0)))
#print(np.min(np.min(a,axis=0)))
center = []
n = np.shape(self.dataSet)[1]
min_value = np.min(np.min(self.dataSet,axis=0))
max_value = np.max(np.max(self.dataSet,axis=0))
range_value = float(max_value-min_value)
for i in range(n):
center.append(float(random.uniform(min_value,max_value)))
center = np.array(center)
return center
def fit(self):
self.centers = {} #存放质心 key:index value:point of centroid
for i in range(self.k):#从data中选质心 初始化
#print(self.dataSet)
self.centers[i] = self.initCenters()
for i in range(self.m):
n = np.shape(self.dataSet_labeled)[1] - 1
self.clf = {} #每个样本归属的簇 即分组情况
self.clf_label = {}
for i in range(self.k):
self.clf[i] = []
self.clf_label[i] = []
for feature in self.dataSet_labeled:
distances = []
for center in self.centers:
#print(type(feature))
#print(type(self.centers[center]))
#print(np.shape(feature))
#print(feature)
#print(feature[:n])
distances.append(cos_dot_procuct(feature[:n],self.centers[center])) #存放feature与各个质心的cos 此时即计算点乘
#print(distances)
#print(type(distances))
#np.array(distances)
#print(distances)
classification = distances.index(np.nanmin(distances)) #取余弦最小
self.clf[classification].append(feature[:n])
self.clf_label[classification].append(feature[-1])
prev_centers = dict(self.centers) #存放原本的质心
#print(self.centers)
#print('------------------')
#print(prev_centers)
for c in self.clf:
self.centers[c] = np.average(self.clf[c],axis=0) #求每个簇内更新之后的质心
self.centers[c] = normalize(self.centers[c])
#print(np.linalg.norm(self.centers[c]))
optimized = True #判断是否达到精度
for center in self.centers:
origin_centers = prev_centers[center] #上一次的质心
current_centers = self.centers[center] #当前质心
if np.sum((current_centers - origin_centers)/(origin_centers+1) *100.0) > self.epislon:
optimized = False
if optimized:
print('次数： ',i)
break
def clusting(self,_data):
#预测输入的样本属于哪个类
cos_ = [cos_similarity(_data,self.centers[center]) for center in self.centers]
index = distances.index[np.nanmax(cos_)]
return index
class BSKM:
def __init__(self,dataSet,k,m,epislon=0.0001):
super().__init__()
self.dataSet = normalize_dataSet(dataSet) #数据集
self.dataSet_labeled = labeling(self.dataSet.tolist()) #[-1]是标签
#self.dataSet = dataSet
self.k = k #簇的个数
self.epislon = epislon #精度
self.m = m #迭代次数
self.size = np.shape(dataSet)[0] #粒子个数
#self.dim = np.shape(dataSet)[1] - 1#维度
#print(self.dataSet)
self.skm = SKM(self.dataSet,2,self.m) #SKM实例
def fit(self):
n = np.shape(self.dataSet_labeled)[1] - 1
self.centers_ = {}
self.centers_[0] = np.average(self.dataSet,axis=0)
self.centers_[0] = normalize(self.centers_[0]) #初始化簇心 为所有样本点的中心点
self.clf_ = {}
self.clf_[0] = []
self.clf_label_ = {}
self.clf_label_[0] = []
index = 0
self.centers = {}
self.clf = {}
self.clf_label = {}
for feature in self.dataSet_labeled:
self.clf_[0].append(feature[:n])
self.clf_label_[0].append(feature[-1])
#print(len(self.clf_[0]))
while(len(self.centers) < self.k):
#print(len(self.centers))
#print(self.k)
#print('len of centers:',len(self.centers))
#重复 直至出现4个簇
cos_distances_ = []
for center in self.centers_:
#print(len(self.centers_))
cos = 0
for clf in self.clf_:
for j in range(len(self.clf_[clf])):
cos += cos_distance(self.clf_[clf][j],self.centers_[center])
#print(cos)
#print(cos_distances_)
cos_distances_.append(cos)
#print(cos)
#print(cos_distances_)
choosed_class_ = cos_distances_.index(np.nanmax(cos_distances_)) #获得SCE最大的簇
#print(choosed_class_)
self.skm.setDataSet(self.clf_[choosed_class_])
#print(len(self.clf_[choosed_class_]))
#print((self.clf_[choosed_class_]))
self.skm.fit()
#note : 应该用一个临时变量存储skm分出的SCE较大簇的聚类结果
temp_centers_ = self.skm.centers
temp_clf_ = self.skm.clf
temp_clf_label_ = self.skm.clf_label
cos_distances = []
for clf in temp_clf_: #遍历每个clf
cos_ = 0
for j in range(len(temp_clf_[clf])): #遍历每个clf中的每个样本
for center in temp_centers_: #计算SCE #此处float转换出现nan 需要解决
#print('temp_clf_[clf][j]:',temp_clf_[clf][j])
#print('temp_centers_[center]:',temp_centers_[center])
cos_ += cos_distance(temp_clf_[clf][j],temp_centers_[center])
#print('cos_',cos_)
cos_distances.append((cos_))
#print('cos_distances:',cos_distances)
choosed_class = cos_distances.index(np.nanmin(cos_distances)) #选取SCE最小的簇
self.centers[index] = temp_centers_[choosed_class]
self.clf[index] = temp_clf_[choosed_class]
self.clf_ = self.clf
self.clf_label[index] = temp_clf_label_[choosed_class]
self.clf_label_ = self.clf_label
self.centers_ = self.centers
index += 1
def run_skm():
x = np.random.rand(100,2)
skm = SKM(x,2,300,0.0001)
skm.fit()
print(skm.clf_label)
#print(skm.clf)
flag = 1
for center in skm.centers:
if flag == 1:
pyplot.scatter(skm.centers[center][0],skm.centers[center][1],marker='*',s=200,c='r')
flag = 0
elif flag == 0:
pyplot.scatter(skm.centers[center][0],skm.centers[center][1],marker='*',s=200,c='b')
for catter in skm.clf:
for point in skm.clf[catter]:
if catter == 0:
pyplot.scatter(point[0],point[1],c='r')
else:
pyplot.scatter(point[0],point[1],c='b')
pyplot.show()
def run_bskm():
x = np.random.rand(100,2)
#print(type(x))
#print(type(y))
#print(x.shape)
#print(y.shape)
bskm = BSKM(x,2,300,0.0001)
bskm.fit()
print(bskm.clf_label)
#print((bskm.centers))
#print(skm.clf)
#print(bskm.centers)
#for i in bskm.clf:
#print(bskm.clf[i])
flag = 1
for center in bskm.centers:
if flag == 1:
pyplot.scatter(bskm.centers[center][0],bskm.centers[center][1],marker='*',s=200,c='r')
flag = 0
elif flag == 0:
pyplot.scatter(bskm.centers[center][0],bskm.centers[center][1],marker='*',s=200,c='b')
for catter in bskm.clf:
for point in bskm.clf[catter]:
if catter == 0:
pyplot.scatter(point[0],point[1],c='r')
else:
pyplot.scatter(point[0],point[1],c='b')
pyplot.show()
if __name__ == '__main__':
#run_skm()
#run_bskm()
run_skm()