1+
2+ import random
3+
4+ from collections import defaultdict
5+
6+ from sklearn .cluster import KMeans
7+ import numpy as np
8+
9+ from utils .misc_utils import distance , check_random_state
10+ from utils .data_generater import makeRandomPoint
11+
12+
13+
14+ class KMeansBase (object ):
15+
16+ def __init__ (self , n_clusters = 8 , init = "random" , max_iter = 300 , random_state = None ):
17+ self .k = n_clusters
18+ self .init = init
19+ self .max_iter = max_iter
20+ self .random_state = random_state
21+
22+
23+ def fit (self , dataset ):
24+ self .dataset = np .array (dataset )
25+ k_points = self ._init_centroids (dataset )
26+ assignments = self .assign_points (dataset , k_points )
27+ old_assignments = None
28+ for i in range (self .max_iter ):
29+ if assignments == old_assignments :
30+ break
31+ error = self .update_error (dataset , assignments )
32+ print ("error" , error )
33+ new_centers = self .update_centers (dataset , assignments )
34+ old_assignments = assignments
35+ assignments = self .assign_points (dataset , new_centers )
36+ return zip (assignments , dataset )
37+
38+
39+ # k个数据点，随机生成
40+ def _init_centroids (self , dataset ):
41+ random_state = check_random_state (self .random_state )
42+ n_samples = dataset .shape [0 ]
43+ if self .init == "random" :
44+ seeds = random_state .permutation (n_samples )[:self .k ]
45+ centers = dataset [seeds ]
46+ elif self .init == "k-means++" :
47+ centers = []
48+
49+ return centers
50+
51+
52+ # 输入：points是一个聚类的点，维度相同
53+ # 输出：这些点的中心点
54+ def point_avg (self , points ):
55+ return np .mean (points , axis = 0 )
56+
57+
58+ #输入：data_set是数据集的点，assignments是每个点在当前归为的类别
59+ #输出：新的中心点list
60+ def update_centers (self , dataset , assignments ):
61+ new_means = defaultdict (list )
62+ centers = []
63+ for assignment , point in zip (assignments , dataset ):
64+ new_means [assignment ].append (point )
65+
66+ for points in new_means .values ():
67+ newCenter = self .point_avg (points )
68+ centers .append (newCenter )
69+
70+ return centers
71+
72+ #输入：data_set是数据集的点，assignments是每个点在当前归为的类别
73+ #输出：新的误差值
74+ def update_error (self , dataset , assignments ):
75+ new_means = defaultdict (list )
76+ error = 0
77+ for assignment , point in zip (assignments , dataset ):
78+ new_means [assignment ].append (point )
79+
80+ for points in new_means .values ():
81+ newCenter = self .point_avg (points )
82+ error += np .sqrt (np .sum (np .square (points - newCenter )))
83+
84+ return error
85+
86+ #输入：data_set原始数据集，centers所有的中心点
87+ #输出：每个点对应的聚类类别
88+ def assign_points (self , dataset , centers ):
89+ assignments = []
90+ for point in dataset :
91+ shortest = float ("inf" ) # 正无穷
92+ shortest_index = 0
93+ for i in range (len (centers )):
94+ val = distance (point , centers [i ])
95+ if val < shortest :
96+ shortest = val
97+ shortest_index = i
98+ assignments .append (shortest_index )
99+ return assignments
100+
101+
102+
103+
104+
105+
106+ if __name__ == "__main__" :
107+
108+ iris = datasets .load_iris ()
109+ km = KMeansBase (3 )
110+ for k in km .fit (iris .data ):
111+ print (k )
112+
113+ kmeans = KMeans (init = 'k-means++' , n_clusters = 10 , n_init = 10 )
114+
115+ # pointList = []
116+ # numPoints = 10000
117+ # dim = 1000
118+ # numClusters = 10
119+ # k = 0
120+ # for i in range(0,numClusters):
121+ # num = int(numPoints/numClusters)
122+ # p = makeRandomPoint(num,dim,k)
123+ # k += 5
124+ # pointList += p.tolist()
125+ #
126+ # start = time.time()
127+ # config= k_means(np.array(pointList), numClusters)
128+ # print("Time taken:",time.time() - start)
0 commit comments