-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsparse_gmm_spsa.py
228 lines (162 loc) · 8.26 KB
/
sparse_gmm_spsa.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
# coding=utf-8
import sys, os, pickle
from datetime import datetime
from scipy.stats import cauchy, halfcauchy
import numpy as np
from sklearn import metrics, cluster
import matplotlib.pyplot as plt
from tqdm import tqdm
import pam
from sklearn.metrics.pairwise import pairwise_distances
import spsa_clustering
import utils
def get_sparse_gmm_model(clust_num, data_shape):
sigma_mu = utils.positive_distr(cauchy.rvs, data_shape)
mu_list = [np.random.normal(0, sigma_mu, size=data_shape) for _ in range(clust_num)]
sigma_list = utils.positive_distr(halfcauchy.rvs, clust_num)
alpha_g = 10
e_zero = np.random.gamma(alpha_g, clust_num * alpha_g)
w_list = np.random.dirichlet([e_zero] * clust_num)
return mu_list, sigma_list, w_list
def get_sparse_gmm_example(data_shape):
w_list = [0.4, 0.6]
mu_list = [np.random.randint(7, size=data_shape) for _ in range(2)]
sigma_list = [1, 1]
return mu_list, sigma_list, w_list
def stat():
clust_num = 3
data_shape = 2
mu_list, sigma_list, w_list = get_sparse_gmm_model(clust_num, data_shape)
spsa_gamma = 1. / 6
spsa_alpha = lambda x: 0.25 / (x ** spsa_gamma)
spsa_beta = lambda x: 15. / (x ** (spsa_gamma / 4))
# spsa_alpha = lambda x: 0.001
# spsa_beta = lambda x: 0.001
n_run = 10
N = 3000
ari_spsa = np.zeros(n_run)
ari_kmeans = np.zeros(n_run)
ari_mb_kmeans = np.zeros(n_run)
ari_pam = np.zeros(n_run)
cent_dist = np.zeros(n_run)
cent_dist_kmeans = np.zeros(n_run)
cent_dist_mb_kmeans = np.zeros(n_run)
cent_dist_pam = np.zeros(n_run)
for i in tqdm(range(n_run)):
clustering = spsa_clustering.ClusteringSPSA(n_clusters=clust_num, data_shape=data_shape, Gammas=None,
alpha=spsa_alpha,
beta=spsa_beta, norm_init=False, verbose=False, sparse=True, eta=700,
spsa_sigma=False)
kmeans = cluster.KMeans(n_clusters=clust_num)
mb_kmeans = cluster.MiniBatchKMeans(n_clusters=clust_num, n_init=1, init='random', max_iter=1,
batch_size=1,
max_no_improvement=None)
data_set = []
true_labels = []
for _ in range(N):
mix_ind = np.random.choice(len(w_list), p=w_list)
data_point = np.random.multivariate_normal(mu_list[mix_ind], np.identity(data_shape) * sigma_list[mix_ind])
data_set.append(data_point)
true_labels.append(mix_ind)
# clustering.fit(data_point)
data_set = np.array(data_set)
# utils.order_clust_centers(np.array(mu_list), clustering)
# clustering.clusters_fill(data_set)
labels_pred_kmenas = kmeans.fit_predict(data_set)
labels_pred_mb_kmeans = mb_kmeans.fit_predict(data_set)
dist = pairwise_distances(data_set)
labels_pred_pam, pam_med = pam.cluster(dist, k=clust_num)
# ari_spsa[i] = metrics.adjusted_rand_score(true_labels, clustering.labels_)
# cent_dist[i] = utils.mean_cent_dist(np.array(mu_list), clustering)
ari_kmeans[i] = metrics.adjusted_rand_score(true_labels, labels_pred_kmenas)
ari_mb_kmeans[i] = metrics.adjusted_rand_score(true_labels, labels_pred_mb_kmeans)
ari_pam[i] = metrics.adjusted_rand_score(true_labels, labels_pred_pam)
cent_dist_kmeans[i] = utils.mean_cent_dist_(np.array(mu_list), kmeans.cluster_centers_)
cent_dist_mb_kmeans[i] = utils.mean_cent_dist_(np.array(mu_list), mb_kmeans.cluster_centers_)
cent_dist_pam[i] = utils.mean_cent_dist_(np.array(mu_list), data_set[pam_med])
print(ari_spsa.mean(), cent_dist.mean())
print('\nMean ARI k-means: {:f}, Mean L2: {:f}'.format(ari_kmeans.mean(), cent_dist_kmeans.mean()))
print('Mean ARI online k-means: {:f}, Mean L2: {:f}'.format(ari_mb_kmeans.mean(), cent_dist_mb_kmeans.mean()))
# print('Mean ARI SPSA clustering: {:f}, Mean L2: {:f}'.format(ari_spsa.mean(), cen))
print('\nMean ARI PAM: {:f}, Mean L2: {:f}'.format(ari_pam.mean(), cent_dist_pam.mean()))
# print('\nMean ARI DBSCAN: {:f}'.format(ari_dbscan.mean()))
def main():
clust_num = 3
data_shape = 2
mu_list, sigma_list, w_list = get_sparse_gmm_model(clust_num, data_shape)
spsa_gamma = 1. / 6
spsa_alpha = lambda x: 0.25 / (x ** spsa_gamma)
spsa_beta = lambda x: 15. / (x ** (spsa_gamma / 4))
# spsa_alpha = lambda x: 0.001
# spsa_beta = lambda x: 0.001
clustering = spsa_clustering.ClusteringSPSA(n_clusters=clust_num, data_shape=data_shape, Gammas=None, alpha=spsa_alpha,
beta=spsa_beta, norm_init=False, verbose=False, sparse=False, eta=None)
N = 3000
data_set = []
true_labels = []
for _ in range(N):
mix_ind = np.random.choice(len(w_list), p=w_list)
data_point = np.random.multivariate_normal(mu_list[mix_ind], np.identity(data_shape) * sigma_list[mix_ind])
data_set.append(data_point)
true_labels.append(mix_ind)
clustering.fit(data_point)
data_set = np.array(data_set)
dataset_name = 'good'
dataset_dir = os.path.join('datasets', dataset_name)
if not os.path.isdir(dataset_dir):
os.mkdir(dataset_dir)
np.save(os.path.join(dataset_dir, 'data.npy'), data_set)
np.save(os.path.join(dataset_dir, 'true.npy'), np.array(true_labels))
param = {'mu': mu_list, 'sigma': sigma_list, 'w': w_list}
with open(os.path.join(dataset_dir, 'param.pickle'), 'wb') as f:
pickle.dump(param, f)
utils.order_clust_centers(np.array(mu_list), clustering)
clustering.clusters_fill(data_set)
ari_spsa = metrics.adjusted_rand_score(true_labels, clustering.labels_)
print('ARI: {}'.format(ari_spsa))
print('Mean centers dist: {}'.format(utils.mean_cent_dist(np.array(mu_list), clustering)))
utils.plot_centers(np.array(mu_list), clustering)
# utils.plot_centers_converg(np.array(mu_list), clustering)
utils.plot_clustering(data_set, clustering.labels_, 'SPSA clustering partition')
utils.plot_clustering(data_set, true_labels, 'True partition')
plt.show()
def load_experiment(name='bad'):
dataset_dir = os.path.join('datasets', name)
data_set = np.load(os.path.join(dataset_dir, 'data.npy'))
true_labels = np.load(os.path.join(dataset_dir, 'true.npy'))
with open(os.path.join(dataset_dir, 'param.pickle'), 'rb') as f:
param = pickle.load(f)
mu_list, sigma_list, w_list = param['mu'], param['sigma'], param['w']
clust_num = len(mu_list)
data_shape = data_set[0].shape[0]
spsa_gamma = 1. / 6
spsa_alpha = lambda x: 0.25 / (x ** spsa_gamma)
spsa_beta = lambda x: 15. / (x ** (spsa_gamma / 4))
clustering = spsa_clustering.ClusteringSPSA(n_clusters=clust_num, data_shape=data_shape, Gammas=None,
alpha=spsa_alpha,
beta=spsa_beta, norm_init=False, verbose=False, sparse=False, eta=None,
spsa_sigma=False)
rand_ind = np.random.permutation(data_set.shape[0])
for i in rand_ind:
clustering.fit(data_set[i])
# utils.order_clust_centers(np.array(mu_list), clustering)
clustering.clusters_fill(data_set[rand_ind])
ari_spsa = metrics.adjusted_rand_score(true_labels[rand_ind], clustering.labels_)
print('ARI: {}'.format(ari_spsa))
print('Mean centers dist: {}'.format(utils.mean_cent_dist(np.array(mu_list), clustering)))
utils.plot_centers(np.array(mu_list), clustering)
# utils.plot_centers_converg(np.array(mu_list), clustering)
# utils.plot_clustering(data_set[rand_ind], clustering.labels_, 'SPSA clustering partition')
# utils.plot_clustering(data_set[rand_ind], true_labels[rand_ind], 'True partition')
# for Gamma in clustering.Gammas:
# print(Gamma)
# for center in clustering.cluster_centers_:
# print(center)
# utils.plot_clustering_cov(data_set, clustering.labels_, 'SPSA clustering partition', clustering.cluster_centers_,
# clustering.Gammas)
plt.show()
if __name__ == '__main__':
# sys.exit(main())
# sys.exit(stat())
# plt.style.use('grayscale')
sys.exit(load_experiment('ugly'))