-
Notifications
You must be signed in to change notification settings - Fork 2
/
KMeanAntColonySolver.py
255 lines (205 loc) · 11.1 KB
/
KMeanAntColonySolver.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
# Source: https://www.kaggle.com/jamesmcguigan/kmeans-ant-colony-optimization
import math
import random
import time
from itertools import chain
from itertools import combinations
from typing import Any
from typing import Callable
from typing import Dict
from typing import List
from typing import Tuple
import matplotlib.pyplot as plt
import numpy as np
from sklearn.cluster import KMeans
from sklearn.neighbors import NearestNeighbors
from search.ant_colony.AntColonySolver import AntColonySolver
from search.ant_colony.AntColonySolver import path_distance
class KmeansAntColonySolver(AntColonySolver):
def __init__(self,
animate: Callable=None,
cluster_factor=1.05, # Multiple for subdividing the problem
random_factor=1, # Create random subgroups - this doesn't work
distance_power_multiple=1.5, # Increase distance_power before final solve
intercity_merge=0, # For each pair of clusters, find a loop between the N nearest neighbours of each
intercity_loop=0, # Construct loops between clusters using N nearest neighbours for each cluster
intercity_random=0, # Construct loops between random members of each cluster
start_smell=2,
start_smell_normalization=0.5,
min_round_trips=2,
max_round_trips=0, # performance shortcut
best_path_smell=1.25, # 2*1 + 1.25*0.5 work best
best_path_smell_multiple=0.5, # Increase best_path_smell before final solve
min_clusters=2,
min_cluster_size=3,
**kwargs
):
self.min_clusters = min_clusters
self.min_cluster_size = min_cluster_size
self.animate = animate
self.intercity_merge = intercity_merge
self.intercity_loop = intercity_loop
self.intercity_random = intercity_random
self.cluster_factor = cluster_factor
self.random_factor = random_factor
self.distance_power_multiple = distance_power_multiple
self.best_path_smell_multiple = best_path_smell_multiple
self.start_smell_normalization = start_smell_normalization
self.kwargs = {
"start_smell": start_smell,
"min_round_trips": min_round_trips,
"max_round_trips": max_round_trips,
"best_path_smell": best_path_smell,
**kwargs
}
super().__init__(**self.kwargs)
### Heuristic Exports
self.ants_used = 0
self.epochs_used = 0
self.round_trips = 0
@staticmethod
def get_numeric_path( problem_path: List[Any] ) -> List[Tuple[int, int]]:
# KMeans requires: List[Tuple[int,int]]
numeric_path = list(problem_path)
try:
if isinstance(numeric_path, dict): numeric_path = list(numeric_path.values()) # if path == {"Name": (x,y)}
if isinstance(numeric_path[0][0], str): numeric_path = [ item[1] for item in numeric_path ] # if path == ("Name", (x,y))
except: pass
return numeric_path
@staticmethod
def group_by_random( problem_path: List[Any], n_clusters ) -> List[Any]:
clusters = [
random.sample(problem_path, math.ceil(len(problem_path) / n_clusters) )
for _ in range(int(n_clusters))
]
return clusters
def group_by_kmeans(self, problem_path: List[Any], n_clusters) -> List[List[Any]]:
if n_clusters == 1: return [ problem_path ]
# Group the cities into KMeans cluster groups of increasing size
numeric_path = self.get_numeric_path(problem_path)
cluster_ids = KMeans(n_clusters=n_clusters).fit_predict(numeric_path)
clusters = [
list({ problem_path[n] for n in range(len(problem_path)) if cluster_ids[n] == cluster_id })
for cluster_id in np.unique(cluster_ids)
]
return clusters
def centroid(self, problem_path: List[Tuple[Any]]) -> Tuple[int,int]:
numeric_path = self.get_numeric_path(problem_path)
return tuple(np.median(numeric_path, axis=0))
# Returns the two nearest neighbours to the centroid for each cluster
def nearest_neighbors(self, clusters: List[List[Any]], n_neighbors=2) -> List[List[Any]]:
center_point = self.centroid(list(chain(*clusters)))
clusters_of_nearest = []
for cluster in clusters:
numeric_path = self.get_numeric_path(cluster)
nn = NearestNeighbors(n_neighbors).fit(numeric_path)
dist, indicies = nn.kneighbors([center_point]) # inputs and outputs are both arrays
clusters_of_nearest.append([ cluster[i] for i in indicies[0] ])
return clusters_of_nearest
def normalize_pheromones(self, norm: float=None):
norm = norm or self.start_smell
# mean = np.mean(list(chain(*[ d.values() for d in self.pheromones.values() ])))
for source in self.pheromones.keys():
for dest in self.pheromones.keys():
self.pheromones[source][dest] *= norm
self.pheromones[source][dest] += norm * self.start_smell_normalization
def solve(self,
problem_path: List[Any],
restart=True,
) -> List[Tuple[int,int]]:
# Initialize the Solver - preserve the pheromone trail between runs
self.solve_initialize(problem_path)
# Break the Travelling Salesman problem down into local clusters of nodes, as detected by KMeans
# Iteratively decrease the number of clusters, until we are back at the starting problem
n_clusters = int( len(problem_path) / ( self.cluster_factor * self.random_factor ) )
random_clusters = self.group_by_random(problem_path, self.random_factor)
results_plot = {}
while n_clusters > self.min_clusters:
results_plot[n_clusters] = []
results_plot[f"{n_clusters}_loop"] = []
results_plot[f"{n_clusters}_merge"] = []
for random_cluster in random_clusters:
kmeans_clusters = self.group_by_kmeans(random_cluster, int(n_clusters))
kmeans_clusters = [ cluster for cluster in kmeans_clusters if len(cluster) >= self.min_cluster_size ]
# Break the map down into kmeans subclusters and create a heuristic pheromone trail
for kmeans_cluster in kmeans_clusters:
if len(kmeans_cluster) < self.min_cluster_size: continue
results = self.solve_subproblem(kmeans_cluster, restart=False)
results_plot[n_clusters] += [ results ]
if len(kmeans_clusters) <= 1: continue # Can't do intercity with a single cluster
# Construct a loop between clusters, using the N closest members to the centroid from each cluster
if self.intercity_loop:
intercity = self.nearest_neighbors(kmeans_clusters, self.intercity_loop)
intercity = list(chain(*intercity))
results = self.solve_subproblem(intercity, restart=False)
results_plot[f"{n_clusters}_loop"] += [ results ]
if self.intercity_random:
intercity = [ random.sample(cluster, max(self.intercity_random, len(cluster)-1)) for cluster in kmeans_clusters ]
intercity = list(chain(*intercity))
results = self.solve_subproblem(intercity, restart=False)
results_plot[f"{n_clusters}_loop"] += [ results ]
# For each pair of clusters, find the optimal path to join them using their N nearest neighbours
if self.intercity_merge:
for clusters in combinations(kmeans_clusters, 2):
intercity = self.nearest_neighbors(clusters, self.intercity_merge)
intercity = list(chain(*intercity))
results = self.solve_subproblem(intercity, restart=False)
results_plot[f"{n_clusters}_merge"] += [ results ]
# self.normalize_pheromones()
n_clusters = int( (n_clusters) // ( self.cluster_factor * self.random_factor ) )
# Display the growth of clusters
if callable(self.animate):
self.animate_results(results_plot, problem_path)
# Now solve the original problem
for key, value in self.kwargs.items():
if hasattr(self, key): setattr(self, key, value)
self.normalize_pheromones()
self.distance_power *= self.distance_power_multiple
self.best_path_smell *= self.best_path_smell_multiple
self.round_trips = 0
self.ant_count = 4 * len(problem_path)
#self.min_ants = self.ants_used + len(problem_path) ** 2 / 2
self.max_ants = self.ants_used + len(problem_path) ** 2 * 2
result = super().solve(problem_path)
if callable(self.animate):
plt.figure()
self.animate(result)
return result
def solve_subproblem(self, problem_path: List[Any], restart=True) -> List[Tuple[int,int]]:
verbose = self.verbose
self.round_trips = 0
self.ant_count = 4 * len(problem_path)
#self.min_ants = 0 # len(problem_path) ** 2 / 2
#self.max_ants = 0 # self.ants_used + len(problem_path) ** 2
time_start = time.perf_counter()
self.verbose = False
result = super().solve(problem_path, restart=False)
# self.normalize_pheromones_path(problem_path, 10000)
self.verbose = verbose
if self.verbose:
print(
f'solve({len(problem_path)})', path_distance(problem_path), '->', path_distance(result),
{ "ant_count": self.ant_count, "ants_used": self.ants_used, "round_trips": self.round_trips, "time": round(time.perf_counter() - time_start, 1) }
)
return result
def animate_results(self, results_plot: Dict[int, List[Any]], problem_path: List=None) -> None:
results_plot = { k:v for k,v in results_plot.items() if len(v) } # remove empty results
if not len(results_plot): return
if not callable(self.animate): return
if problem_path is None: problem_path = []
grid_cols = max(4, math.ceil(np.sqrt(len(results_plot))))
grid_cols = min(grid_cols,len(results_plot))
grid_rows = math.ceil(len(results_plot)/grid_cols)
grid_size = ( grid_rows, grid_cols )
figure, axes = plt.subplots(*grid_size, figsize=(grid_size[0]*10, grid_size[1]*10))
plt.tight_layout(pad=5)
try:
for ax in chain(*axes): ax.axis('off')
except: pass
for index, N in enumerate(results_plot.keys()):
plt.subplot(*grid_size, index+1)
# unique_lengths = list(np.unique(list(map(len,results_plot[N]))))
plt.title(f'{len(problem_path)}/{N} = {len(results_plot[N])} clusters')
for results in results_plot[N]:
self.animate(results)
#plt.close(figure)