Coverage for src/sensai/geoanalytics/geo_clustering.py: 0%
163 statements
« prev ^ index » next coverage.py v7.6.1, created at 2024-08-13 22:17 +0000
1import collections
2import itertools
3import math
4from abc import abstractmethod, ABC
5from typing import List, Tuple, Iterator, Optional
7import numpy as np
8import sklearn.cluster
10from .geo_coords import GeoCoord
11from .local_coords import LocalCoordinateSystem
12from ..clustering import GreedyAgglomerativeClustering
15class GeoCoordClusterer(ABC):
16 @abstractmethod
17 def fit_geo_coords(self, geo_coords: List[GeoCoord]):
18 """
19 :param geo_coords: the coordinates to be clustered
20 """
21 pass
23 @abstractmethod
24 def clusters_indices(self) -> Tuple[List[List[int]], List[int]]:
25 """
26 :return: a tuple (clusters, outliers), where clusters is a list of point indices, one list for each
27 cluster containing the indices of points within the cluster, and outliers is the list of indices of points not within
28 clusters
29 """
30 pass
33class GreedyAgglomerativeGeoCoordClusterer(GeoCoordClusterer):
34 def __init__(self,
35 max_min_distance_for_merge_m: float,
36 max_distance_m: float,
37 min_cluster_size: int,
38 lcs: LocalCoordinateSystem = None):
39 """
40 :param max_min_distance_for_merge_m: the maximum distance, in metres, for the minimum distance between two existing clusters for a merge
41 to be admissible
42 :param max_distance_m: the maximum distance, in metres, between any two points for the points to be allowed to be in the same cluster
43 :param min_cluster_size: the minimum number of points any valid cluster must ultimately contain; the points in any smaller clusters
44 shall be considered as outliers
45 :param lcs: the local coordinate system to use for clustering; if None, compute based on mean coordinates passed when fitting
46 """
47 self.lcs = lcs
48 self.min_cluster_size = min_cluster_size
49 self.max_min_distance_for_merge = max_min_distance_for_merge_m
50 self.max_distance_m = max_distance_m
51 self.squared_max_min_distance_for_merge = max_min_distance_for_merge_m * max_min_distance_for_merge_m
52 self.squared_max_distance = max_distance_m * max_distance_m
53 self.local_points = None
54 self.m_min_distance: Optional["GreedyAgglomerativeGeoCoordClusterer.Matrix"] = None
55 self.m_max_squared_distance: Optional["GreedyAgglomerativeGeoCoordClusterer.Matrix"] = None
56 self.clusters = None
58 class Matrix:
59 UNSET_VALUE = np.inf
61 def __init__(self, dim: int):
62 self.m = np.empty((dim, dim))
63 self.m.fill(np.inf)
65 def set(self, c1: int, c2: int, value: float):
66 self.m[c1][c2] = value
67 self.m[c2][c1] = value
69 def get(self, c1: int, c2: int) -> float:
70 return self.m[c1][c2]
72 class LocalPoint:
73 def __init__(self, xy: np.ndarray, idx: int):
74 self.idx = idx
75 self.xy = xy
77 class Cluster(GreedyAgglomerativeClustering.Cluster):
78 def __init__(self, point: "GreedyAgglomerativeGeoCoordClusterer.LocalPoint", idx: int,
79 clusterer: 'GreedyAgglomerativeGeoCoordClusterer'):
80 self.idx = idx
81 self.clusterer = clusterer
82 self.points = [point]
84 def merge_cost(self, other: "GreedyAgglomerativeGeoCoordClusterer.Cluster"):
85 cartesian_product = itertools.product(self.points, other.points)
86 min_squared_distance = math.inf
87 max_squared_distance = 0
88 for p1, p2 in cartesian_product:
89 diff = p1.xy - p2.xy
90 squared_distance = np.dot(diff, diff)
91 if squared_distance > self.clusterer.squared_max_distance:
92 max_squared_distance = math.inf
93 break
94 else:
95 min_squared_distance = min(squared_distance, min_squared_distance)
97 # fill cache: the max value takes precedence; if it is inf (no merge admissible), then the min value is also set to inf;
98 # the min value valid only if the max value is finite
99 self.clusterer.m_max_squared_distance.set(self.idx, other.idx, max_squared_distance)
100 if np.isinf(max_squared_distance):
101 self.clusterer.m_min_distance.set(self.idx, other.idx, np.inf)
102 else:
103 self.clusterer.m_min_distance.set(self.idx, other.idx, min_squared_distance)
105 if np.isinf(max_squared_distance):
106 return math.inf
107 if min_squared_distance <= self.clusterer.squared_max_min_distance_for_merge:
108 return min_squared_distance
109 return math.inf
111 def merge(self, other):
112 self.points += other.points
114 def fit_geo_coords(self, geo_coords: List[GeoCoord]) -> None:
115 self.m_min_distance = self.Matrix(len(geo_coords))
116 self.m_max_squared_distance = self.Matrix(len(geo_coords))
117 if self.lcs is None:
118 mean_coord = GeoCoord.mean_coord(geo_coords)
119 self.lcs = LocalCoordinateSystem(mean_coord.lat, mean_coord.lon)
120 self.local_points = [self.LocalPoint(np.array(self.lcs.get_local_coords(p.lat, p.lon)), idx) for idx, p in enumerate(geo_coords)]
121 clusters = [self.Cluster(lp, i, self) for i, lp in enumerate(self.local_points)]
122 gac = GreedyAgglomerativeClustering(clusters,
123 merge_candidate_determination_strategy=self.MergeCandidateDeterminationStrategy(self.max_distance_m, self))
124 clusters = gac.apply_clustering()
125 self.clusters = clusters
127 def clusters_indices(self) -> Tuple[List[List[int]], List[int]]:
128 outliers = []
129 clusters = []
130 for c in self.clusters:
131 indices = [p.idx for p in c.points]
132 if len(c.points) < self.min_cluster_size:
133 outliers.extend(indices)
134 else:
135 clusters.append(indices)
136 return clusters, outliers
138 class MergeCandidateDeterminationStrategy(GreedyAgglomerativeClustering.MergeCandidateDeterminationStrategy):
139 def __init__(self, search_radius_m: float, parent: "GreedyAgglomerativeGeoCoordClusterer"):
140 super().__init__()
141 self.parent = parent
142 self.searchRadiusM = search_radius_m
144 def set_clusterer(self, clusterer: GreedyAgglomerativeClustering):
145 super().set_clusterer(clusterer)
146 points = []
147 for wc in self.clusterer.wrapped_clusters:
148 c: GreedyAgglomerativeGeoCoordClusterer.Cluster = wc.cluster
149 for p in c.points:
150 points.append(p.xy)
151 assert len(points) == len(self.clusterer.wrapped_clusters)
152 points = np.stack(points)
153 self.kdtree = sklearn.neighbors.KDTree(points)
155 def iter_candidate_indices(self, wc: "GreedyAgglomerativeClustering.WrappedCluster", initial: bool,
156 merged_cluster_indices: Tuple[int, int] = None) -> Iterator[int]:
157 c: GreedyAgglomerativeGeoCoordClusterer.Cluster = wc.cluster
158 if initial:
159 local_point = c.points[0] # pick any point from wc, since we use maximum cluster extension as search radius
160 indices = self.kdtree.query_radius(np.reshape(local_point.xy, (1, 2)), self.searchRadiusM)[0]
161 candidate_set = set()
162 for idx in indices:
163 wc = self.clusterer.wrapped_clusters[idx]
164 candidate_set.add(wc.get_cluster_association().idx)
165 yield from sorted(candidate_set)
166 else:
167 # The new distance values (max/min) between wc and any cluster index otherIdx can be computed from the cached distance
168 # values of the two clusters from which wc was created through a merge:
169 # The max distance is the maximum of the squared distances of the original clusters (and if either is inf, then
170 # a merge is definitely inadmissible, because one of the original clusters was already too far away).
171 # The min distance is the minimum of the squred distances of the original clusters.
172 c1, c2 = merged_cluster_indices
173 max1 = self.parent.m_max_squared_distance.m[c1]
174 max2 = self.parent.m_max_squared_distance.m[c2]
175 max_combined = np.maximum(max1, max2)
176 for otherIdx, maxSqDistance in enumerate(max_combined):
177 min_sq_distance = np.inf
178 if maxSqDistance <= self.parent.squared_max_distance:
179 wc_other = self.clusterer.wrapped_clusters[otherIdx]
180 if wc_other.is_merged():
181 continue
182 min1 = self.parent.m_min_distance.get(c1, otherIdx)
183 min2 = self.parent.m_min_distance.get(c2, otherIdx)
184 min_sq_distance = min(min1, min2)
185 if min_sq_distance <= self.parent.squared_max_min_distance_for_merge:
186 yield GreedyAgglomerativeClustering.ClusterMerge(wc, wc_other, min_sq_distance)
187 # update cache
188 self.parent.m_max_squared_distance.set(wc.idx, otherIdx, maxSqDistance)
189 self.parent.m_min_distance.set(wc.idx, otherIdx, min_sq_distance)
192class SkLearnGeoCoordClusterer(GeoCoordClusterer):
193 def __init__(self, clusterer, lcs: LocalCoordinateSystem = None):
194 """
195 :param clusterer: a clusterer from sklearn.cluster
196 :param lcs: the local coordinate system to use for Euclidian conversion; if None, determine from data (using mean coordinate as
197 centre)
198 """
199 self.lcs = lcs
200 self.clusterer = clusterer
201 self.local_points = None
203 def fit_geo_coords(self, geo_coords: List[GeoCoord]):
204 if self.lcs is None:
205 mean_coord = GeoCoord.mean_coord(geo_coords)
206 self.lcs = LocalCoordinateSystem(mean_coord.lat, mean_coord.lon)
207 self.local_points = [self.lcs.get_local_coords(p.lat, p.lon) for p in geo_coords]
208 self.clusterer.fit(self.local_points)
210 def _clusters(self, mode):
211 clusters = collections.defaultdict(list)
212 outliers = []
213 for idxPoint, idxCluster in enumerate(self.clusterer.labels_):
214 if mode == "localPoints":
215 item = self.local_points[idxPoint]
216 elif mode == "indices":
217 item = idxPoint
218 else:
219 raise ValueError()
220 if idxCluster >= 0:
221 clusters[idxCluster].append(item)
222 else:
223 outliers.append(item)
224 return list(clusters.values()), outliers
226 def clusters_local_points(self) -> Tuple[List[List[Tuple[float, float]]], List[Tuple[float, float]]]:
227 """
228 :return: a tuple (clusters, outliers), where clusters is a dictionary mapping from cluster index to
229 the list of local points within the cluster and outliers is a list of local points not within
230 clusters
231 """
232 return self._clusters("localPoints")
234 def clusters_indices(self) -> Tuple[List[List[int]], List[int]]:
235 return self._clusters("indices")
238class DBSCANGeoCoordClusterer(SkLearnGeoCoordClusterer):
239 def __init__(self, eps, min_samples, lcs: LocalCoordinateSystem = None, **kwargs):
240 """
241 :param eps: the maximum distance between two samples for one to be considered as in the neighbourhood of the other
242 :param min_samples: the minimum number of samples that must be within a neighbourhood for a cluster to be formed
243 :param lcs: the local coordinate system for conversion to a Euclidian space
244 :param kwargs: additional arguments to pass to DBSCAN (see https://scikit-learn.org/stable/modules/generated/sklearn.cluster.DBSCAN.html)
245 """
246 super().__init__(sklearn.cluster.DBSCAN(eps=eps, min_samples=min_samples, **kwargs), lcs)