-
Notifications
You must be signed in to change notification settings - Fork 5
/
k_prototypes.py
409 lines (386 loc) · 17 KB
/
k_prototypes.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
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
# -*- coding: utf-8 -*-
"""
Created on Mon Feb 3 15:18:14 2020
k-prototypes 聚类算法的实现
@author: zhenyu wu
"""
import pandas as pd
import numpy as np
from sklearn import preprocessing
from sklearn import metrics
from kmodes.kprototypes import KPrototypes
from copy import deepcopy
DEMO = True
N = 5
def Load_Data(demo=DEMO):
"""
加载测试数据集
Parameters
----------
demo : bool值
是否仅加载测试数据
Returns
-------
data : DataFrame
特征分离后的数据集
data_id : list
事件编号
len(numerical_features) : int
数值特征个数
len(category_features) : int
类别特征个数
"""
data = pd.read_csv('../data/clean_data.csv', low_memory=False)
data = data.drop(columns=['Unnamed: 0'])
data = data.sample(frac=1, random_state=2020).reset_index(drop=True)
data_id = list(data['eventid'])
numerical_features = ['latitude', 'longitude', 'nperps', 'nperpcap', 'nkill', 'nkillus',
'nkillter', 'nwound', 'nwoundus', 'nwoundte', 'nhostkid', 'nhostkidus',
'nkill_nwound', 'nkill_nwound_us', 'nkill_nwound_dte', 'kill_perct',
'wound_perct', 'nkill_perct', 'scite_count', ]
category_features = ['extended', 'country', 'region', 'specificity', 'vicinity', 'crit1',
'crit2', 'crit3', 'doubtterr', 'alternative', 'multiple', 'success',
'suicide', 'attacktype1', 'attacktype2', 'targtype1', 'targsubtype1',
'natlty1', 'guncertain1', 'individual', 'claimed', 'claimmode',
'weaptype1', 'weapsubtype1', 'property', 'propextent', 'ishostkid',
'INT_LOG', 'INT_IDEO', 'INT_MISC', 'INT_ANY', 'iyear', 'imonth', 'iday', ]
data = data[numerical_features+category_features]
if demo:
num_data = 400
data = data[:num_data]
data_id = data_id[:num_data]
numerical_data = data[numerical_features]
# 对连续特征进行归一化处理
min_max_scaler = preprocessing.MinMaxScaler()
numerical_data = pd.DataFrame(min_max_scaler.fit_transform(numerical_data))
numerical_data.columns = numerical_features
category_data = data[category_features]
data = pd.concat([numerical_data, category_data], axis=1)
return data, data_id, len(numerical_features), len(category_features)
def Calculate_Single_Distance(num_data, cat_data, num_center, cat_center):
"""
计算两个样本之间的距离
Parameters
----------
num_data : list
数据样本中的连续数值部分
cat_data : list
数据样本中的类别型特征部分
num_center : list
聚类中心点的连续数值部分
cat_center : list
聚类中心点的类别特征部分
num_numerical : float
欧式距离权重
num_category : float
汉明距离权重
Returns
-------
euclidean : float
两个样本点之间的欧几里得距离
hamming : float
两个样本点之间的汉明距离
"""
euclidean = np.linalg.norm(np.array(num_data)-np.array(num_center))**2
hamming = np.shape(np.nonzero(np.array(cat_data)-np.array(cat_center))[0])[0]
return euclidean, hamming
def Calculate_Center(data, n, num_numerical, num_category):
"""
更新聚类中心点
Parameters
----------
data : DataFrame
数据样本
n : int
聚类中心的个数，默认值为5
num_numerical : int
数值特征个数
num_category : int
类别特征个数
Returns
-------
numerical_centers : DataFrame
更新后的数值型特征的聚类中心点
category_centers : DataFrame
更新后的类别型特征的聚类中心点
"""
numerical_centers = []
category_centers = []
for i in range(n):
sub_data = data.loc[data.label==i]
sub_data_numerical = sub_data.iloc[:, :num_numerical]
sub_data_category = sub_data.iloc[:, num_numerical:-1]
numerical_center = []
for col in sub_data_numerical.columns:
numerical_center.append(sub_data_numerical[col].mean())
numerical_centers.append(numerical_center)
category_center = []
for col in sub_data_category.columns:
category_center.append(list(sub_data_category[col].mode())[0])
category_centers.append(category_center)
numerical_centers = pd.DataFrame(numerical_centers)
numerical_centers.columns = sub_data_numerical.columns
category_centers = pd.DataFrame(category_centers)
category_centers.columns = sub_data_category.columns
return numerical_centers, category_centers
def K_Prototypes(random_seed, n, data, num_numerical, num_category, max_iters, mode):
"""
K_Prototypes混合聚类算法
Parameters
----------
n : int
聚类中心的个数
data : DataFrame
用于聚类的样本
random_seed : int
随机数种子
num_numerical : int
数值特征个数
num_category : int
类别特征个数
max_iters : int
最大迭代次数
mode : int
计算模式
Returns
-------
newlabel : list
最终的聚类结果
center_numerical : DataFrame
数值型特征的聚类中心点
center_category : DataFrame
类别型特征的聚类中心点
"""
all_features = num_numerical+num_category
# 当belta=0时，本算法为kmeans聚类
# 当alpha=0时，本算法为kmodes聚类
if mode==1:
alpha = 0
belta = 1
print('K_Modes聚类')
elif mode==2:
alpha = 1
belta = 0
print('K_Means聚类')
else:
alpha = num_numerical/all_features
belta = num_category/all_features
print('K_Prototypes聚类')
# 随机选定n个初始聚类中心点
init_center_points = data.sample(n=n, replace=False, random_state=random_seed, axis=0)
# 对数据特征按照类别进行划分
numerical_data = data.iloc[:, :num_numerical]
category_data = data.iloc[:, num_numerical:]
init_center_numerical = init_center_points.iloc[:, :num_numerical]
init_center_category = init_center_points.iloc[:, num_numerical:]
# 计算每个样本到各个聚类中心簇的距离
label = []
for i in range(len(data)):
all_distance = []
euclidean = []
hamming = []
for j in range(n):
sig_euclidean, sig_hamming = Calculate_Single_Distance(
numerical_data.iloc[[i]].values[0],
category_data.iloc[[i]].values[0],
init_center_numerical.iloc[[j]].values[0],
init_center_category.iloc[[j]].values[0],)
euclidean.append(sig_euclidean)
hamming.append(sig_hamming)
for j in range(n):
if sum(euclidean)==0:
distance = belta*hamming[j]/sum(hamming)
elif sum(hamming)==0:
distance = alpha*euclidean[j]/sum(euclidean)
else:
distance = alpha*euclidean[j]/sum(euclidean)+belta*hamming[j]/sum(hamming)
all_distance.append(distance)
label.append(np.argmin(np.array(all_distance)))
data['label'] = label
# 迭代更新聚类中心部分
err_distance = 1
iter_count = 0
for iter_counts in range(max_iters):
print('INFO--当前为第{}次迭代'.format(iter_count+1), end="\t")
if err_distance!=0:
iter_count += 1
center_numerical, center_category = Calculate_Center(data, n, num_numerical, num_category)
newlabel = []
for i in range(len(data)):
all_distance = []
euclidean = []
hamming = []
for j in range(n):
sig_euclidean, sig_hamming = Calculate_Single_Distance(
numerical_data.iloc[[i]].values[0],
category_data.iloc[[i]].values[0],
center_numerical.iloc[[j]].values[0],
center_category.iloc[[j]].values[0],)
euclidean.append(sig_euclidean)
hamming.append(sig_hamming)
for j in range(n):
if sum(euclidean)==0:
distance = belta*hamming[j]/sum(hamming)
elif sum(hamming)==0:
distance = alpha*euclidean[j]/sum(euclidean)
else:
distance = alpha*euclidean[j]/sum(euclidean)+belta*hamming[j]/sum(hamming)
all_distance.append(distance)
newlabel.append(np.argmin(np.array(all_distance)))
err_distance = np.shape(np.nonzero(np.array(list(data['label']))-np.array(newlabel))[0])[0]
print('loss: {}'.format(err_distance))
data['label'] = newlabel
else:
break
print('各类别的样本个数统计结果: {}'.format(data['label'].value_counts().values))
print('最终的迭代次数为: {}'.format(iter_count))
data.drop('label', axis=1, inplace=True)
return newlabel, center_numerical, center_category
def CUM_index(data, num_category, num_numerical, n, label, mode):
"""
计算CUM指标，用于对混合聚类模型的聚类效果评价，该指标越小越优
参考链接：https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/metrics/cluster/_unsupervised.py
Parameters
----------
data : DataFrame
用于聚类的样本
num_numerical : int
数值特征个数
num_category : int
类别特征个数
n : int
聚类中心的个数
mode : int
计算模式
Returns
-------
CUM : float
混合类别数据聚类结果性能度量指标
"""
eval_data = deepcopy(data)
eval_data_numerical = eval_data.iloc[:, :num_numerical]
eval_data_numerical['label'] = label
eval_data_category = eval_data.iloc[:, num_numerical:]
eval_data_category['label'] = label
num_features = num_category+num_numerical
if mode==1: # k-modes
alpha = 0
belta = 1
elif mode==2: # k-means
alpha = 1
belta = 0
else: # k-prototypes
alpha = num_numerical/num_features
belta = num_category/num_features
# 计算连续型特征部分的聚类分散度CUN，该指标越小越好
if num_numerical!=0:
sub_eval_data_numerical = []
center_numerical = []
for i in range(n):
sub_cluster_data = eval_data_numerical.loc[eval_data_numerical.label==i]
sub_cluster_data = sub_cluster_data[sub_cluster_data.columns[:-1]]
sub_eval_data_numerical.append(sub_cluster_data)
center_numerical.append(list(sub_cluster_data.mean(axis=0)))
center_numerical = pd.DataFrame(center_numerical)
center_numerical.columns = sub_cluster_data.columns
cluster_inner_distance = []
for i in range(n):
sub_data = sub_eval_data_numerical[i]
inner_distance = []
for j in range(len(sub_data)):
sub_inner_distance = np.linalg.norm(np.array(center_numerical.iloc[[i]].values[0])-
np.array(sub_data.iloc[[j]].values[0]))
inner_distance.append(sub_inner_distance)
cluster_inner_distance.append(sum(inner_distance)/len(sub_data))
temp = 0
sub_dbi = []
for i in range(n):
inner_sub_dbi = []
for j in range(n):
if i!=j:
temp = (cluster_inner_distance[i]+cluster_inner_distance[j])/np.linalg.norm(np.array(center_numerical.iloc[[i]].values[0])-
np.array(center_numerical.iloc[[j]].values[0]))
inner_sub_dbi.append(temp)
sub_dbi.append(max(inner_sub_dbi))
CUN = sum(sub_dbi)/n
else:
CUN = 0
# 计算类别型特征部分的聚类分散度CUC，该指标越小越好
if num_category!=0:
sub_eval_data_category = []
center_category = []
for i in range(n):
sub_cluster_data = eval_data_category.loc[eval_data_category.label==i]
sub_cluster_data = sub_cluster_data[sub_cluster_data.columns[:-1]]
sub_eval_data_category.append(sub_cluster_data)
data_category_mod = []
for col in sub_cluster_data.columns:
data_category_mod.append(list(sub_cluster_data[col].mode())[0])
center_category.append(data_category_mod)
center_category = pd.DataFrame(center_category)
center_category.columns = sub_cluster_data.columns
cluster_inner_distance = []
for i in range(n):
sub_data = sub_eval_data_category[i]
inner_distance = []
for j in range(len(sub_data)):
sub_inner_distance = np.shape(np.nonzero(np.array(center_category.iloc[[i]].values[0])-
np.array(sub_data.iloc[[j]].values[0]))[0])[0]
inner_distance.append(sub_inner_distance)
cluster_inner_distance.append(sum(inner_distance)/len(sub_data))
temp = 0
sub_dbi = []
for i in range(n):
inner_sub_dbi = []
for j in range(n):
if i!=j:
temp = (cluster_inner_distance[i]+cluster_inner_distance[j])/np.shape(np.nonzero(np.array(center_category.iloc[[i]].values[0])-
np.array(center_category.iloc[[j]].values[0]))[0])[0]
inner_sub_dbi.append(temp)
sub_dbi.append(max(inner_sub_dbi))
CUC = sum(sub_dbi)/n
else:
CUC = 0
CUM = alpha*CUN+belta*CUC
return CUM
if __name__ == '__main__':
data, data_id, num_numerical_features, num_category_features = Load_Data()
for n in range(2, 10):
print('聚为{}类'.format(n))
# 本文算法
label, center_numerical, center_category = K_Prototypes(random_seed=2020, n=n, data=data,
num_numerical=num_numerical_features,
num_category=num_category_features,
max_iters = 20, mode=3)
CUM = CUM_index(data=data, num_category=num_category_features,
num_numerical=num_numerical_features, n=n,
label=label, mode=3)
print("K_Prototypes算法的Calinski-Harabaz Index值为：{}".format(metrics.calinski_harabasz_score(data, label)))
print("K_Prototypes算法的CUM值为：{}".format(CUM))
# 开源包
kp = KPrototypes(n_clusters=n, init='Huang', n_init=1, verbose=True,
n_jobs=4, random_state=2020, gamma=num_category_features/num_numerical_features)
KPrototypes_results = kp.fit_predict(data, categorical=list(range(num_numerical_features, num_numerical_features+num_category_features-1)))
print("K_Prototypes算法包的Calinski-Harabaz Index值为：{}".format(metrics.calinski_harabasz_score(data, KPrototypes_results)))
CUM = CUM_index(data=data, num_category=num_category_features,
num_numerical=num_numerical_features, n=n,
label=KPrototypes_results, mode=3)
print("K_Prototypes算法包的CUM值为：{}".format(CUM))
label_2, center_numerical_2, center_category_2 = K_Prototypes(random_seed=2020, n=N, data=data,
num_numerical=num_numerical_features+num_category_features,
num_category=0,
max_iters = 20, mode=2)
CUM = CUM_index(data=data, num_category=0,
num_numerical=num_numerical_features+num_category_features, n=N,
label=label_2, mode=2)
print("K_Means算法的Calinski-Harabaz Index值为：{}".format(metrics.calinski_harabasz_score(data, label_2)))
print("K_Means算法的CUM值为：{}".format(CUM))
label_3, center_numerical_3, center_category_3 = K_Prototypes(random_seed=2020, n=N, data=data,
num_numerical=0,
num_category=num_numerical_features+num_category_features,
max_iters = 20, mode=1)
CUM = CUM_index(data=data, num_category=num_numerical_features+num_category_features,
num_numerical=0, n=N,
label=label_3, mode=1)
print("K_Modes算法的Calinski-Harabaz Index值为：{}".format(metrics.calinski_harabasz_score(data, label_3)))
print("K_Modes算法的CUM值为：{}".format(CUM))