-
Notifications
You must be signed in to change notification settings - Fork 0
/
generate_experiments_moves.py
256 lines (220 loc) · 13 KB
/
generate_experiments_moves.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
import os
import sys
import time
import numpy as np
import sklearn.metrics.pairwise as skl
import moves
import clustering_utils as cu
import data_generation_tools as dg
import data_visualization_tools as dv
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
np.set_printoptions(linewidth=1000, precision=4, threshold=np.nan, suppress=True)
# noinspection PyStringFormat
def save_data(props, num_trials, dirpath, filename="test"):
num_trials = float(num_trials)
header = ['property',
'num_exp',
'tim', 'tim_std',
'pur', 'pur_std', 'pur_max', 'pur_max_n',
'pur_min', 'pur_min_std', 'pur_min_max', 'pur_min_max_n',
'ene', 'ene_std', 'ene_max', 'ene_max_n',
'per', 'per_std', 'per_max', 'per_max_n',
'CH', 'CH_std', 'CH_max',
'SI', 'SI_std', 'SI_max',
'DB', 'DB_std', 'DB_min',
'DU', 'DU_std', 'DU_max']
props = ['value',
'%d' % num_trials,
'%.3f' % np.mean(props['times']), '%.2f' % np.std(props['times']),
'%.3f' % np.mean(props['purities']), '%.2f' % np.std(props['purities']), '%.3f' % np.max(props['purities']), '%.2f' % (np.count_nonzero(props['purities'] == np.max(props['purities']))/num_trials),
'%.3f' % np.mean(props['min_purities']), '%.2f' % np.std(props['min_purities']), '%.3f' % np.max(props['min_purities']), '%.2f' % (np.count_nonzero(props['min_purities'] == np.max(props['min_purities']))/num_trials),
'%.3f' % np.mean(props['energies']), '%.2f' % np.std(props['energies']), '%.3f' % np.max(props['energies']), '%.2f' % (np.count_nonzero(props['energies'] == np.max(props['energies']))/num_trials),
'%.3f' % np.mean(props['percentages_energy']), '%.2f' % np.std(props['percentages_energy']), '%.3f' % np.max(props['percentages_energy']), '%.2f' % (np.count_nonzero(props['percentages_energy'] == np.max(props['percentages_energy']))/num_trials),
'%.3f' % np.mean(props['CH']), '%.2f' % np.std(props['CH']), '%.3f' % np.max(props['CH']),
'%.3f' % np.mean(props['SI']), '%.2f' % np.std(props['SI']), '%.3f' % np.max(props['SI']),
'%.3f' % np.mean(props['DB']), '%.2f' % np.std(props['DB']), '%.3f' % np.max(props['DB']),
'%.3f' % np.mean(props['DU']), '%.2f' % np.std(props['DU']), '%.3f' % np.max(props['DU'])]
f = open(str(dirpath) + '/' + filename + '.csv', 'a')
for i in range(len(header)):
f.write(header[i] + ',')
f.write('\n')
for i in range(len(props)):
f.write(props[i] + ',')
f.close()
def save_params(params, dirpath):
f = open(str(dirpath) + '/params_data.dat', 'a')
f.write('param|value\n')
f.write('n|%d\n' % params['n'])
f.write('k|%d\n' % params['K'])
f.write('dim_space|%d\n' % params['dim_space'])
f.write('l|%d\n' % params['l'])
f.write('sigma_1|%.4f\n' % params['sigma_1'])
f.write('sigma_2|%.4f\n' % params['sigma_2'])
f.write('min_dist|%.4f\n' % params['min_dist'])
f.close()
# MAIN CODE ============================================================================================================
# noinspection PyStringFormat
def run_test(n, k, sigma, min_dist, num_trials, random_init, use_previous_dataset=-1, use_D31=False, dir_name='test'):
time_start = time.time()
# Data generation and visualization --------------------------------------------------------------------------------
if use_D31:
P, ground_truth = dg.get_D31_data()
k = len(np.unique(ground_truth))
N = P.shape[0]
else:
if use_previous_dataset == -1:
N = n * k
params = {'sigma_1': 1, 'sigma_2': sigma, 'min_dist': min_dist, 'simplex': False, 'K': k, 'dim_space': 2,
'l': 2, 'n': n, 'use_prev_p': False, 'shuffle': False}
P, ground_truth = dg.generate_data_random(params)
else:
P = np.loadtxt(dir_name + "/" + str(use_previous_dataset) + "/P.dat")
ground_truth = np.loadtxt(dir_name + "/" + str(use_previous_dataset) + "/gt.dat", dtype=np.int16)
k = len(np.unique(ground_truth))
N = P.shape[0]
C = skl.pairwise_distances(P, metric='sqeuclidean')
dv.plot_data(P, k, ground_truth, 2, show_data=True)
# Other parameters -------------------------------------------------------------------------------------------------
use_IPM = 0
num_max_it = 20
ab_pur, ab_min_pur, ab_ene, ab_per, ab_tim, ab_ch, ab_si, ab_db, ab_du, ab_it = \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials)
# ae_pur, ae_min_pur, ae_ene, ae_per, ae_tim, ae_ch, ae_si, ae_db, ae_du, ae_it = \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials)
# aebs_pur, aebs_min_pur, aebs_ene, aebs_per, aebs_tim, aebs_ch, aebs_si, aebs_db, aebs_du, aebs_it\
# = \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials), \
# np.zeros(num_trials), np.zeros(num_trials)
ls_pur, ls_min_pur, ls_ene, ls_per, ls_tim, ls_ch, ls_si, ls_db, ls_du, ls_it = \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials), \
np.zeros(num_trials), np.zeros(num_trials)
# Iterate ----------------------------------------------------------------------------------------------------------
for t in range(num_trials):
print ("Running experiment %d of %d" % (t + 1, num_trials))
if random_init:
lb_init = np.random.randint(0, k, N)
while len(np.unique(lb_init)) < k:
lb_init = np.random.randint(0, k, N)
ab_sequence = None
else:
lb_init = np.zeros(N, dtype=int)
ab_sequence = np.array([np.random.choice(k, k, replace=False), np.random.choice(k, k, replace=False)])
start_t = time.time()
ab_labeling, ab_it[t] = moves.large_move_maxcut(C, k, lb_init,
move_type="ab", ab_sequence=ab_sequence,
num_max_it=num_max_it, use_IPM=use_IPM)
ab_pur[t], ab_min_pur[t], ab_ene[t], ab_per[t], ab_ch[t], ab_si[t], ab_db[t], ab_du[t] \
= cu.stats_clustering_pairwise(C, ab_labeling, ground_truth, P, use_other_measures=True)
ab_tim[t] = time.time() - start_t
print time.time() - start_t
print ab_it[t]
# start_t = time.time()
# ae_labeling, ae_it[t] = moves.large_move_maxcut(C, k, lb_init,
# move_type="ae", ab_sequence=ab_sequence,
# num_max_it=num_max_it, use_IPM=use_IPM)
# ae_pur[t], ae_min_pur[t], ae_ene[t], ae_per[t], ae_ch[t], ae_si[t], ae_db[t], ae_du[t]\
# = cu.stats_clustering(P, C, ae_labeling, ground_truth)
# ae_tim[t] = time.time() - start_t
# print time.time() - start_t
#
# start_t = time.time()
# aebs_labeling, aebs_it[t] = moves.large_move_maxcut(C, k, lb_init,
# move_type="ae_bs", ab_sequence=ab_sequence,
# num_max_it=num_max_it, use_IPM=use_IPM)
# aebs_pur[t], aebs_min_pur[t], aebs_ene[t], aebs_per[t], aebs_ch[t], aebs_si[t], aebs_db[t], aebs_du[t] \
# = cu.stats_clustering(P, C, aebs_labeling, ground_truth)
# aebs_tim[t] = time.time() - start_t
# print time.time() - start_t
start_t = time.time()
ls_labeling, ls_it[t] = cu.local_search(C, k, lb_init)
ls_pur[t], ls_min_pur[t], ls_ene[t], ls_per[t], ls_ch[t], ls_si[t], ls_db[t], ls_du[t] \
= cu.stats_clustering_pairwise(C, ls_labeling, ground_truth, P, use_other_measures=True)
ls_tim[t] = time.time() - start_t
# print time.time() - start_t
if not os.path.exists(dir_name):
os.makedirs(dir_name)
# Save data --------------------------------------------------------------------------------------------------------
experiment_id = 0
while os.path.exists(dir_name + '/' + str(experiment_id)):
experiment_id += 1
dirpath = dir_name + '/' + str(experiment_id)
os.makedirs(dirpath)
dv.plot_data(P, k, ground_truth, 2, show_legend=False, show_data=False, save_to_file=True,
file_name=dirpath + '/GT',
title='Ground Truth (N = %d, K = %d)' % (n*k, k))
np.savetxt(str(dirpath) + '/P.dat', P)
np.savetxt(str(dirpath) + '/gt.dat', ground_truth, fmt='%d')
if not use_D31 and use_previous_dataset == -1:
save_params(params, dirpath)
if not random_init:
np.savetxt(str(dirpath) + '/ab_sequence.dat', ab_sequence)
print("ALPHA-BETA SWAP (id %d):" % experiment_id)
print("> Mean time: %.4f + %.4f" % (np.mean(ab_tim), np.std(ab_tim)))
print("> Mean min pur: %.4f + %.4f" % (np.mean(ab_min_pur), np.std(ab_min_pur)))
print("> Mean SI: %.4f" % np.mean(ab_si))
props = {"purities": ab_pur, "min_purities": ab_min_pur, "energies": ab_ene, "percentages_energy": ab_per,
"CH": ab_ch, "SI": ab_si, "DB": ab_db, "DU": ab_du, "times": ab_tim, "iterations": ab_it}
save_data(props, num_trials, dirpath, filename="ab_results")
# print("ALPHA EXPANSION (id %d):" % experiment_id)
# print("> Mean time: %.4f + %.4f" % (np.mean(ae_tim), np.std(ae_tim)))
# print("> Mean min pur: %.4f + %.4f" % (np.mean(ae_min_pur), np.std(ae_min_pur)))
# print("> Mean SI: %.4f" % np.mean(ae_si))
# props = {"purities": ae_pur, "min_purities": ae_min_pur, "energies": ae_ene, "percentages_energy": ae_per,
# "CH": ae_ch, "SI": ae_si, "DB": ae_db, "DU": ae_du, "times": ae_tim, "iterations": ae_it}
# save_data(props, num_trials, filename="ae_results")
#
# print("ALPHA EXPANSION-BETA SHRINK (id %d):" % experiment_id)
# print("> Mean time: %.4f + %.4f" % (np.mean(aebs_tim), np.std(aebs_tim)))
# print("> Mean min pur: %.4f + %.4f" % (np.mean(aebs_min_pur), np.std(aebs_min_pur)))
# print("> Mean SI: %.4f" % np.mean(aebs_si))
# props = {"purities": aebs_pur, "min_purities": aebs_min_pur, "energies": aebs_ene, "percentages_energy": aebs_per,
# "CH": aebs_ch, "SI": aebs_si, "DB": aebs_db, "DU": aebs_du, "times": aebs_tim, "iterations": aebs_it}
# save_data(props, num_trials, dirpath, filename="aebs_results")
print("LOCAL SEARCH (id %d):" % experiment_id)
print("> Mean time: %.4f + %.4f" % (np.mean(ls_tim), np.std(ls_tim)))
print("> Mean min pur: %.4f + %.4f" % (np.mean(ls_min_pur), np.std(ls_min_pur)))
print("> Mean SI: %.4f" % np.mean(ls_si))
props = {"purities": ls_pur, "min_purities": ls_min_pur, "energies": ls_ene, "percentages_energy": ls_per,
"CH": ls_ch, "SI": ls_si, "DB": ls_db, "DU": ls_du, "times": ls_tim, "iterations": ls_it}
save_data(props, num_trials, dirpath, filename="ls_results")
# gt_pur, gt_min_pur, gt_ene, gt_per, gt_ch, gt_si, gt_db, gt_du \
# = cu.stats_clustering_pairwise(C, ground_truth, ground_truth, P)
# props = {"purities": gt_pur, "min_purities": gt_min_pur, "energies": gt_ene, "percentages_energy": gt_per,
# "CH": gt_ch, "SI": gt_si, "DB": gt_db, "DU": gt_du, "times": 0, "iterations": 0}
# save_data(props, num_trials, dirpath, filename="gt_results")
print("Total time (id %d): %.4f s\n" % (experiment_id, time.time() - time_start))
if __name__ == "__main__":
random_init = True
dir_name = 'test_ls_ab'
num_trials = 3
K = [4]
points_per_cluster = [4]
sigmas = [0.3, 0.5, 0.6]
min_dists = [0.01]
exp_number = 1
for min_dist in min_dists:
for k in K:
for sigma in sigmas:
for n in points_per_cluster:
print('EXP. %d - 1 ============================================================' % (exp_number + 1))
print('K: %d, n: %d, min_dist: %.4f, sigma %.4f' % (k, n, min_dist, sigma))
exp_number += 1
run_test(n, k, sigma, min_dist, num_trials, random_init, dir_name=dir_name)
print("SET FINISHED ========================================================================== \n")
# print('EXP - D31 ============================================================')
# run_test(0, 0, 0, 0, num_trials, random_init, use_D31=True, dir_name=dir_name)
# print("SET FINISHED ========================================================================== \n")