-
Notifications
You must be signed in to change notification settings - Fork 0
/
genshapelet.py
280 lines (237 loc) · 11.8 KB
/
genshapelet.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
import numpy as np
import pandas as pd
import time
import os
import sys
from copy import copy
from fastdtw import fastdtw
from scipy import interpolate
from scipy.stats import levy, zscore, mode
from sklearn.metrics import silhouette_score
from sklearn.metrics.pairwise import *
from scipy.spatial.distance import *
from sklearn.linear_model import LogisticRegression
from sklearn.cluster import DBSCAN
def pairwise_fastdtw(X, **kwargs):
X = [list(enumerate(pattern)) for pattern in X]
triu = [fastdtw(X[i], X[j], **kwargs)[0] if i != j else 0 for i in range(len(X)) for j in range(i, len(X))]
matrix = np.zeros([len(X)] * 2)
matrix[np.triu_indices(len(X))] = triu
matrix += np.tril(matrix.T, -1)
return matrix
class individual:
def __init__(self, start: list = None, slen: list = None):
if start is None:
start = []
if slen is None:
slen = []
self.start = start
self.slen = slen
self.cluster = None
class genshapelet:
def __init__(self, ts_path: 'path to file', nsegments, min_support, smin, smax, output_folder=''):
self.ts = pd.read_csv(ts_path, header=None)
self.ts_path = ts_path
self.nsegments = nsegments
if self.nsegments is None:
self.nsegments = int(len(self.ts) / (2 * smax) + 1)
if self.nsegments < 2:
sys.exit('nsegments must be at least 2 for computing clustering quality')
self.min_support = min_support
self.smin = smin
self.smax = smax
if os.path.exists(output_folder):
pass
elif os.access(output_folder, os.W_OK):
pass
else:
sys.exit('output_folder not createable.')
self.output_folder = output_folder
self.probability = 2 / self.nsegments
self.random_walk = False
def run(self, popsize: dict(type=int, help='> 3, should be odd'), sigma: dict(type=float, help='mutation factor'),
t_max, pairwise_distmeasures=[
(pairwise_distances, {'metric': 'cosine'}),
(pairwise_distances, {'metric': 'chebyshev'}),
(pairwise_distances, {'metric': 'euclidean'}),
(pairwise_fastdtw, {'dist': euclidean})],
fusion=True, notes=''):
# For pairwise_distances
# From scikit-learn: ['cityblock', 'cosine', 'euclidean', pairwise_distancesl1', 'l2', 'manhattan'].
# From scipy.spatial.distance: ['braycurtis', 'canberra', 'chebyshev', 'correlation', 'dice', 'hamming',
# 'jaccard', 'kulsinski', 'mahalanobis', 'matching', 'minkowski', 'rogerstanimoto', 'russellrao',
# 'seuclidean', 'sokalmichener', 'sokalsneath', 'sqeuclidean', 'yule']
print('-->')
# print('working with ' + str(self.nsegments) + ' windows')
t_max *= 60
t_start = time.time()
population, fitness = [], []
for i in range(0, popsize):
population.append(self.make_individual())
fitness.append(self.evaluate(population[i], pairwise_distmeasures, fusion))
fitness_curve = []
best_fit = -np.inf
iterations = 0
t_elapsed = time.time() - t_start
while(t_elapsed < t_max):
order = np.argsort(fitness)[::-1]
ix_maxfitness = order[0]
if(fitness[ix_maxfitness] > best_fit):
best_fit = fitness[ix_maxfitness]
fitness_curve.append((t_elapsed, iterations, best_fit))
# print((t_elapsed, iterations, best_fit))
# if(iterations % 500) == 0:
# print((t_elapsed, iterations, best_fit))
new_population = []
new_population.append(population[ix_maxfitness]) # elite
fitness[0] = fitness[ix_maxfitness]
if self.random_walk:
for i in range(1, popsize):
new_population.append(self.make_individual())
fitness[i] = self.evaluate(population[i], pairwise_distmeasures, fusion)
else:
for i in range(1, int(popsize / 2), 2):
new_population.append(population[order[i]])
new_population.append(population[order[i + 1]])
self.crossover(new_population[i], new_population[i + 1])
self.mutate(new_population[i], sigma)
self.mutate(new_population[i + 1], sigma)
fitness[i] = self.evaluate(new_population[i], pairwise_distmeasures, fusion)
fitness[i + 1] = self.evaluate(new_population[i + 1], pairwise_distmeasures, fusion)
for i in range(int(popsize / 2), popsize):
new_population.append(self.make_individual())
fitness[i] = self.evaluate(new_population[i], pairwise_distmeasures, fusion)
population = new_population
iterations += 1
t_elapsed = time.time() - t_start
ix_maxfitness = np.argmax(fitness)
fitness_curve.append((t_max, iterations, fitness[ix_maxfitness]))
# print('t_elapsed: ' + str(t_elapsed))
# print('iterations: ' + str(iterations))
# print('fitness: ' + str(fitness[ix_maxfitness]))
name = self.make_filename(popsize, sigma, t_max, notes)
self.write_shapelets(population[ix_maxfitness], name)
self.write_fitness(fitness_curve, name)
# print(population[ix_maxfitness].start)
# print(population[ix_maxfitness].slen)
# print(population[ix_maxfitness].cluster)
# print(self.evaluate(population[ix_maxfitness], pairwise_distmeasures, fusion))
print('--<')
return 0
def evaluate(self, x: individual, pairwise_distmeasures, fusion):
# get patterns from individual
patterns, classlabels = [], []
for i in range(len(x.start)):
df = self.ts.loc[x.start[i]:x.start[i] + x.slen[i] - 1, :]
classlabels.append(mode(df.loc[:, [0]])[0][0][0]) # xD
df = df.loc[:, [1]].apply(zscore).fillna(0) # consider extending for multivariate ts
upsampled_ix = np.linspace(0, len(df) - 1, self.smax) # upsampling
new_values = interpolate.interp1d(np.arange(len(df)), np.array(df).flatten(), kind='cubic')(upsampled_ix)
patterns.append(new_values)
patterns = np.array(patterns)
classlabels = np.array(classlabels)
# print('patterns\n' + str(patterns)) # DEBUG
# print('classlabels ' + str(classlabels)) # DEBUG
distances = {}
cols = len(patterns)
for measure, params in pairwise_distmeasures:
distances[str(measure) + str(params)] = measure(patterns, **params)[np.triu_indices(cols)]
distances = pd.DataFrame(distances)
if fusion:
clf = LogisticRegression()
different_class = np.zeros([cols] * 2)
different_class[classlabels[:, None] != classlabels] = 1
different_class = different_class[np.triu_indices(cols)]
if 1 in different_class:
clf.fit(distances, different_class)
combined_distance = clf.predict_proba(distances)[:, 1]
else:
return -np.inf
dist_matrix = np.zeros([cols] * 2)
dist_matrix[np.triu_indices(cols)] = combined_distance
dist_matrix += np.tril(dist_matrix.T, -1)
else:
measure, params = pairwise_distmeasures[0]
dist_matrix = measure(patterns, **params)
# print('dist_matrix\n' + str(dist_matrix)) # DEBUG
# epsilon! consider: eps=dist_matrix.mean()/1.5
db = DBSCAN(eps=dist_matrix.mean(), min_samples=self.min_support, metric='precomputed', n_jobs=-1).fit(dist_matrix)
x.cluster = db.labels_
try:
fitness = silhouette_score(dist_matrix, x.cluster)
except Exception as e:
fitness = -np.inf
# print(fitness) # DEBUG
return fitness
def validate(self, x):
order = np.argsort(x.start)
for i in range(len(order)):
for j in range(1, len(order) - i):
if(x.start[order[i + j]] - x.start[order[i]] > self.smax):
break
if(x.start[order[i]] + x.slen[order[i]] > x.start[order[i + j]]):
return False
return True
def mutate(self, x, sigma):
for i in range(len(x.start)):
if(np.random.uniform() < self.probability):
tmp_start, tmp_slen = copy(x.start[i]), copy(x.slen[i])
x.slen[i] += int(sigma * (self.smax + 1 - self.smin) * levy.rvs())
x.slen[i] = (x.slen[i] - self.smin) % (self.smax + 1 - self.smin) + self.smin
x.start[i] = (x.start[i] + int(sigma * len(self.ts) * levy.rvs())) % (len(self.ts) - x.slen[i])
if not self.validate(x):
x.start[i], x.slen[i] = copy(tmp_start), copy(tmp_slen)
return 0
def crossover(self, x, y):
for i in range(min(len(x.start), len(y.start))):
if(np.random.uniform() < self.probability):
tmp_start_x, tmp_slen_x = copy(x.start[i]), copy(x.slen[i])
tmp_start_y, tmp_slen_y = copy(y.start[i]), copy(y.slen[i])
x.start[i], y.start[i] = y.start[i], x.start[i]
x.slen[i], y.slen[i] = y.slen[i], x.slen[i]
if not self.validate(x):
x.start[i], x.slen[i] = copy(tmp_start_x), copy(tmp_slen_x)
if not self.validate(y):
y.start[i], y.slen[i] = copy(tmp_start_y), copy(tmp_slen_y)
return 0
def write_fitness(self, x: 'fitness curve', filename):
df = pd.DataFrame(x)
df.to_csv(self.output_folder + '/' + filename + '.fitness.csv', index=False, header=False)
def write_shapelets(self, x: individual, filename):
out = {}
out['start'] = [start for start in x.start]
out['slen'] = [slen for slen in x.slen]
out['cluster'] = [cluster for cluster in x.cluster] if x.cluster is not None else [-2] * len(x.start)
df = pd.DataFrame(out, columns=['start', 'slen', 'cluster'])
df.sort_values('cluster', inplace=True) # unordered indizes .reset_index(inplace=True, drop=True)
df.to_csv(self.output_folder + '/' + filename + '.shapelets.csv', index=False)
return 0
def make_filename(self, popsize, sigma, t_max, notes):
filename = os.path.splitext(os.path.basename(self.ts_path))[0] # get name without path and extension
motifs = str(self.nsegments) + 'x' + str(self.min_support) + 'motifs'
window_length = str(self.smin) + '-' + str(self.smax) + 'window'
hyperparameter = str(popsize) + '_' + str(sigma) + '_' + str(t_max / 60) + '_' + str(notes)
return 'genshapelet_' + filename + '_' + motifs + '_' + window_length + '_' + hyperparameter
def make_individual(self):
x = individual()
for i in range(self.nsegments):
x.slen.append(np.random.randint(self.smin, self.smax + 1))
x.start.append(np.random.randint(0, len(self.ts) - x.slen[i]))
valid = False
attempts = 5 # this is random, right; but the whole should stay random so .. ¯\_(ツ)_/¯
while(not valid and attempts > 0):
valid = True
for j in range(i):
if((x.start[i] + x.slen[i] <= x.start[j]) or (x.start[j] + x.slen[j] <= x.start[i])):
continue
else:
valid = False
attempts -= 1
x.start[i] = np.random.randint(0, len(self.ts) - x.slen[i])
break
if (attempts == 0):
# print('The individual isn\'t complete. Check nsegments and smax parameter.')
x.slen.pop()
x.start.pop()
break
return x