|
1
|
|
|
"""!
|
|
2
|
|
|
|
|
3
|
|
|
@brief Cluster analysis algorithm: DBSCAN.
|
|
4
|
|
|
@details Implementation based on article:
|
|
5
|
|
|
- M.Ester, H.Kriegel, J.Sander, X.Xiaowei. A density-based algorithm for discovering clusters in large spatial databases with noise. 1996.
|
|
6
|
|
|
|
|
7
|
|
|
@authors Andrei Novikov ([email protected])
|
|
8
|
|
|
@date 2014-2018
|
|
9
|
|
|
@copyright GNU Public License
|
|
10
|
|
|
|
|
11
|
|
|
@cond GNU_PUBLIC_LICENSE
|
|
12
|
|
|
PyClustering is free software: you can redistribute it and/or modify
|
|
13
|
|
|
it under the terms of the GNU General Public License as published by
|
|
14
|
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
15
|
|
|
(at your option) any later version.
|
|
16
|
|
|
|
|
17
|
|
|
PyClustering is distributed in the hope that it will be useful,
|
|
18
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
19
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
20
|
|
|
GNU General Public License for more details.
|
|
21
|
|
|
|
|
22
|
|
|
You should have received a copy of the GNU General Public License
|
|
23
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
24
|
|
|
@endcond
|
|
25
|
|
|
|
|
26
|
|
|
"""
|
|
27
|
|
|
|
|
28
|
|
|
|
|
29
|
|
|
from enum import IntEnum;
|
|
30
|
|
|
|
|
31
|
|
|
from pyclustering.container.kdtree import kdtree;
|
|
32
|
|
|
|
|
33
|
|
|
from pyclustering.cluster.encoder import type_encoding;
|
|
34
|
|
|
|
|
35
|
|
|
from pyclustering.core.wrapper import ccore_library;
|
|
36
|
|
|
|
|
37
|
|
|
from pyclustering.utils import get_argument;
|
|
38
|
|
|
|
|
39
|
|
|
import pyclustering.core.dbscan_wrapper as wrapper;
|
|
40
|
|
|
|
|
41
|
|
|
|
|
42
|
|
|
|
|
43
|
|
|
class dbscan_data_type(IntEnum):
|
|
44
|
|
|
"""!
|
|
45
|
|
|
@brief Enumeration of DBSCAN input data types that is used for processing: points, adjacency matrix.
|
|
46
|
|
|
|
|
47
|
|
|
"""
|
|
48
|
|
|
|
|
49
|
|
|
## Input data is represented by points that are contained by array like container, for example, by list.
|
|
50
|
|
|
POINTS = 0;
|
|
51
|
|
|
|
|
52
|
|
|
## Input data is represented by distance matrix between points.
|
|
53
|
|
|
DISTANCE_MATRIX = 1;
|
|
54
|
|
|
|
|
55
|
|
|
|
|
56
|
|
|
class dbscan:
|
|
57
|
|
|
"""!
|
|
58
|
|
|
@brief Class represents clustering algorithm DBSCAN.
|
|
59
|
|
|
@details This DBSCAN algorithm is KD-tree optimized.
|
|
60
|
|
|
|
|
61
|
|
|
CCORE option can be used to use the pyclustering core - C/C++ shared library for processing that significantly increases performance.
|
|
62
|
|
|
|
|
63
|
|
|
Example:
|
|
64
|
|
|
@code
|
|
65
|
|
|
# sample for cluster analysis (represented by list)
|
|
66
|
|
|
sample = read_sample(path_to_sample);
|
|
67
|
|
|
|
|
68
|
|
|
# create object that uses CCORE for processing
|
|
69
|
|
|
dbscan_instance = dbscan(sample, 0.5, 3, True);
|
|
70
|
|
|
|
|
71
|
|
|
# cluster analysis
|
|
72
|
|
|
dbscan_instance.process();
|
|
73
|
|
|
|
|
74
|
|
|
# obtain results of clustering
|
|
75
|
|
|
clusters = dbscan_instance.get_clusters();
|
|
76
|
|
|
noise = dbscan_instance.get_noise();
|
|
77
|
|
|
@endcode
|
|
78
|
|
|
|
|
79
|
|
|
"""
|
|
80
|
|
|
|
|
81
|
|
|
def __init__(self, data, eps, neighbors, ccore = True, **kwargs):
|
|
82
|
|
|
"""!
|
|
83
|
|
|
@brief Constructor of clustering algorithm DBSCAN.
|
|
84
|
|
|
|
|
85
|
|
|
@param[in] data (list): Input data that is presented as list of points (objects), each point should be represented by list or tuple.
|
|
86
|
|
|
@param[in] eps (double): Connectivity radius between points, points may be connected if distance between them less then the radius.
|
|
87
|
|
|
@param[in] neighbors (uint): minimum number of shared neighbors that is required for establish links between points.
|
|
88
|
|
|
@param[in] ccore (bool): if True than DLL CCORE (C++ solution) will be used for solving the problem.
|
|
89
|
|
|
@param[in] **kwargs: Arbitrary keyword arguments (available arguments: 'data_type').
|
|
90
|
|
|
|
|
91
|
|
|
Keyword Args:
|
|
92
|
|
|
data_type (dbscan_data_type): Data type of input sample 'data' that is processed by the algorithm (simple sequence of points or distance matrix).
|
|
93
|
|
|
|
|
94
|
|
|
"""
|
|
95
|
|
|
|
|
96
|
|
|
self.__pointer_data = data;
|
|
97
|
|
|
self.__kdtree = None;
|
|
98
|
|
|
self.__eps = eps;
|
|
99
|
|
|
self.__sqrt_eps = eps * eps;
|
|
100
|
|
|
self.__neighbors = neighbors;
|
|
101
|
|
|
|
|
102
|
|
|
self.__visited = [False] * len(self.__pointer_data);
|
|
103
|
|
|
self.__belong = [False] * len(self.__pointer_data);
|
|
104
|
|
|
|
|
105
|
|
|
self.__data_type = get_argument('data_type', dbscan_data_type.POINTS, **kwargs);
|
|
106
|
|
|
|
|
107
|
|
|
self.__clusters = [];
|
|
108
|
|
|
self.__noise = [];
|
|
109
|
|
|
|
|
110
|
|
|
self.__ccore = ccore;
|
|
111
|
|
|
if (self.__ccore):
|
|
112
|
|
|
self.__ccore = ccore_library.workable();
|
|
113
|
|
|
|
|
114
|
|
|
|
|
115
|
|
|
def process(self):
|
|
116
|
|
|
"""!
|
|
117
|
|
|
@brief Performs cluster analysis in line with rules of DBSCAN algorithm.
|
|
118
|
|
|
|
|
119
|
|
|
@see get_clusters()
|
|
120
|
|
|
@see get_noise()
|
|
121
|
|
|
|
|
122
|
|
|
"""
|
|
123
|
|
|
|
|
124
|
|
|
if self.__ccore is True:
|
|
125
|
|
|
(self.__clusters, self.__noise) = wrapper.dbscan(self.__pointer_data, self.__eps, self.__neighbors, self.__data_type);
|
|
126
|
|
|
|
|
127
|
|
|
else:
|
|
128
|
|
|
if self.__data_type == dbscan_data_type.POINTS:
|
|
129
|
|
|
self.__kdtree = kdtree(self.__pointer_data, range(len(self.__pointer_data)));
|
|
130
|
|
|
|
|
131
|
|
|
for i in range(0, len(self.__pointer_data)):
|
|
132
|
|
|
if self.__visited[i] is False:
|
|
133
|
|
|
|
|
134
|
|
|
cluster = self.__expand_cluster(i);
|
|
135
|
|
|
if cluster is not None:
|
|
136
|
|
|
self.__clusters.append(cluster);
|
|
137
|
|
|
else:
|
|
138
|
|
|
self.__noise.append(i);
|
|
139
|
|
|
self.__belong[i] = True;
|
|
140
|
|
|
|
|
141
|
|
|
|
|
142
|
|
|
def get_clusters(self):
|
|
143
|
|
|
"""!
|
|
144
|
|
|
@brief Returns allocated clusters.
|
|
145
|
|
|
|
|
146
|
|
|
@remark Allocated clusters can be returned only after data processing (use method process()). Otherwise empty list is returned.
|
|
147
|
|
|
|
|
148
|
|
|
@return (list) List of allocated clusters, each cluster contains indexes of objects in list of data.
|
|
149
|
|
|
|
|
150
|
|
|
@see process()
|
|
151
|
|
|
@see get_noise()
|
|
152
|
|
|
|
|
153
|
|
|
"""
|
|
154
|
|
|
|
|
155
|
|
|
return self.__clusters;
|
|
156
|
|
|
|
|
157
|
|
|
|
|
158
|
|
|
def get_noise(self):
|
|
159
|
|
|
"""!
|
|
160
|
|
|
@brief Returns allocated noise.
|
|
161
|
|
|
|
|
162
|
|
|
@remark Allocated noise can be returned only after data processing (use method process() before). Otherwise empty list is returned.
|
|
163
|
|
|
|
|
164
|
|
|
@return (list) List of indexes that are marked as a noise.
|
|
165
|
|
|
|
|
166
|
|
|
@see process()
|
|
167
|
|
|
@see get_clusters()
|
|
168
|
|
|
|
|
169
|
|
|
"""
|
|
170
|
|
|
|
|
171
|
|
|
return self.__noise;
|
|
172
|
|
|
|
|
173
|
|
|
|
|
174
|
|
|
def get_cluster_encoding(self):
|
|
175
|
|
|
"""!
|
|
176
|
|
|
@brief Returns clustering result representation type that indicate how clusters are encoded.
|
|
177
|
|
|
|
|
178
|
|
|
@return (type_encoding) Clustering result representation.
|
|
179
|
|
|
|
|
180
|
|
|
@see get_clusters()
|
|
181
|
|
|
|
|
182
|
|
|
"""
|
|
183
|
|
|
|
|
184
|
|
|
return type_encoding.CLUSTER_INDEX_LIST_SEPARATION;
|
|
185
|
|
|
|
|
186
|
|
|
|
|
187
|
|
|
def __expand_cluster(self, index_point):
|
|
188
|
|
|
"""!
|
|
189
|
|
|
@brief Expands cluster from specified point in the input data space.
|
|
190
|
|
|
|
|
191
|
|
|
@param[in] index_point (list): Index of a point from the data.
|
|
192
|
|
|
|
|
193
|
|
|
@return (list) Return tuple of list of indexes that belong to the same cluster and list of points that are marked as noise: (cluster, noise), or None if nothing has been expanded.
|
|
194
|
|
|
|
|
195
|
|
|
"""
|
|
196
|
|
|
|
|
197
|
|
|
cluster = None;
|
|
198
|
|
|
self.__visited[index_point] = True;
|
|
199
|
|
|
neighbors = self.__neighbor_indexes(index_point);
|
|
200
|
|
|
|
|
201
|
|
|
if len(neighbors) >= self.__neighbors:
|
|
202
|
|
|
cluster = [ index_point ];
|
|
203
|
|
|
|
|
204
|
|
|
self.__belong[index_point] = True;
|
|
205
|
|
|
|
|
206
|
|
|
for i in neighbors:
|
|
207
|
|
|
if self.__visited[i] is False:
|
|
208
|
|
|
self.__visited[i] = True;
|
|
209
|
|
|
next_neighbors = self.__neighbor_indexes(i);
|
|
210
|
|
|
|
|
211
|
|
|
if len(next_neighbors) >= self.__neighbors:
|
|
212
|
|
|
# if some node has less then minimal number of neighbors than we shouldn't look at them
|
|
213
|
|
|
# because maybe it's a noise.
|
|
214
|
|
|
neighbors += [k for k in next_neighbors if ( (k in neighbors) == False)];
|
|
215
|
|
|
|
|
216
|
|
|
if self.__belong[i] is False:
|
|
217
|
|
|
cluster.append(i);
|
|
218
|
|
|
self.__belong[i] = True;
|
|
219
|
|
|
|
|
220
|
|
|
return cluster;
|
|
221
|
|
|
|
|
222
|
|
|
def __neighbor_indexes(self, index_point):
|
|
223
|
|
|
"""!
|
|
224
|
|
|
@brief Return list of indexes of neighbors of specified point for the data.
|
|
225
|
|
|
|
|
226
|
|
|
@param[in] index_point (list): An index of a point for which potential neighbors should be returned in line with connectivity radius.
|
|
227
|
|
|
|
|
228
|
|
|
@return (list) Return list of indexes of neighbors in line the connectivity radius.
|
|
229
|
|
|
|
|
230
|
|
|
"""
|
|
231
|
|
|
|
|
232
|
|
|
if self.__data_type == dbscan_data_type.POINTS:
|
|
233
|
|
|
kdnodes = self.__kdtree.find_nearest_dist_nodes(self.__pointer_data[index_point], self.__eps);
|
|
234
|
|
|
return [node_tuple[1].payload for node_tuple in kdnodes if node_tuple[1].payload != index_point];
|
|
235
|
|
|
|
|
236
|
|
|
else:
|
|
237
|
|
|
distances = self.__pointer_data[index_point];
|
|
238
|
|
|
return [ index_neighbor for index_neighbor in range(len(distances))
|
|
239
|
|
|
if ( (distances[index_neighbor] <= self.__eps) and (index_neighbor != index_point) ) ];
|
|
240
|
|
|
|
annoviko /
pyclustering
