-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathmain.py
196 lines (176 loc) · 8.2 KB
/
main.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
from Load_data import load_data_conv
from tensorflow.keras.optimizers import SGD, Adam
import numpy as np
from sklearn.manifold import TSNE
import os
from time import time
import Nmetrics
from SDMVC import MvDEC
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")
def _make_data_and_model(args):
# prepare dataset
x, y = load_data_conv(args.dataset)
view = len(x)
view_shapes = []
Loss = []
Loss_weights = []
for v in range(view):
view_shapes.append(x[v].shape[1:])
Loss.append('kld')
Loss.append('mse')
Loss_weights.append(args.lc)
Loss_weights.append(args.Idec)
print(view_shapes)
print(Loss)
print(Loss_weights)
# prepare optimizer
optimizer = Adam(lr=args.lr)
# prepare the model
n_clusters = len(np.unique(y))
# n_clusters = 40 # over clustering
print("n_clusters:" + str(n_clusters))
# lc = 0.1
model = MvDEC(filters=[32, 64, 128, 10], n_clusters=n_clusters, view_shape=view_shapes)
model.compile(optimizer=optimizer, loss=Loss, loss_weights=Loss_weights)
return x, y, model
def train(args):
# get data and mode
x, y, model = _make_data_and_model(args)
model.model.summary()
# pretraining
t0 = time()
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
if args.pretrain_dir is not None and os.path.exists(args.pretrain_dir): # load pretrained weights
model.autoencoder.load_weights(args.pretrain_dir)
# model.load_weights(args.pretrain_dir)
else: # train
optimizer = Adam(lr=args.lr)
model.pretrain(x, y, optimizer=optimizer, epochs=args.pretrain_epochs,
batch_size=args.batch_size, save_dir=args.save_dir, verbose=args.verbose)
args.pretrain_dir = args.save_dir + '/ae_weights.h5'
t1 = time()
print("Time for pretraining: %ds" % (t1 - t0))
# clustering
# DEMVC, IDEC, DEC
# y_pred, y_mean_pred = model.fit(arg=args, x=x, y=y, maxiter=args.maxiter,
# batch_size=args.batch_size, UpdateCoo=args.UpdateCoo,
# save_dir=args.save_dir)
# SDMVC
y_pred, y_mean_pred = model.new_fit(arg=args, x=x, y=y, maxiter=args.maxiter,
batch_size=args.batch_size, UpdateCoo=args.UpdateCoo,
save_dir=args.save_dir)
if y is not None:
for view in range(len(x)):
print('Final: acc=%.4f, nmi=%.4f, ari=%.4f' %
(Nmetrics.acc(y, y_pred[view]), Nmetrics.nmi(y, y_pred[view]), Nmetrics.ari(y, y_pred[view])))
print('Final: acc=%.4f, nmi=%.4f, ari=%.4f' %
(Nmetrics.acc(y, y_mean_pred), Nmetrics.nmi(y, y_mean_pred), Nmetrics.ari(y, y_mean_pred)))
t2 = time()
print("Time for pretaining, clustering and total: (%ds, %ds, %ds)" % (t1 - t0, t2 - t1, t2 - t0))
print('='*60)
def test(args):
assert args.weights is not None
x, y, model = _make_data_and_model(args)
model.model.summary()
print('Begin testing:', '-' * 60)
model.load_weights(args.weights)
y_pred, y_mean_pred = model.predict_label(x=x)
if y is not None:
for view in range(len(x)):
print('Final: acc=%.4f, nmi=%.4f, ari=%.4f' %
(Nmetrics.acc(y, y_pred[view]), Nmetrics.nmi(y, y_pred[view]), Nmetrics.ari(y, y_pred[view])))
print('Final: acc=%.4f, nmi=%.4f, ari=%.4f' %
(Nmetrics.acc(y, y_mean_pred), Nmetrics.nmi(y, y_mean_pred), Nmetrics.ari(y, y_mean_pred)))
print('End testing:', '-' * 60)
if __name__ == "__main__":
# -------------------------------------------------------
# Dataset settings
# 'MNIST_USPS' # 2 views, 10 clusters, 5000 examples
# 'Fashion_MV' # 3 views, 10 clusters, 10000 examples
# 'BDGP' # 2 views, 5 clusters, 2500 examples
# 'Caltech101_20' # 6 views, 20 clusters, 2386 examples
# -------------------------------------------------------
data = 'MNIST_USPS'
TEST = True # Test the clustering performance of the trained models
# TEST = False
# Train_ae = True # The stability of AE’s pre-training and K-means might be the bottleneck for AE/K-means based MVC
Train_ae = False # The reported results are the average values after pre-training
AR = 0.90 # Aligned Ratio, e.g., 90%, the threshold that determines the stop condition
Coo = 1 # Unified P
View = 1 # View_first SetC for DEMVC
# K123q = View # DEMVC
K123q = 0 # SDMVC
if Coo == 0:
K123q = 0 # K-means 1：k1 , 2：k2, 3：k3, 0: k-means, >view number: no settings centers
epochs = 500 # 500 epochs for pre-training AEs
Update_Coo = 1000 # Iterations to update self-supervised objective
Maxiter = 30000 # Max iterations for DEC, IDEC or DEMVC, not for SDMVC
Batch = 256 # Batch size
Idec = 1.0 # Dec 0.0 , Idec 1.0 -------- Reconstruction loss 1.0
lc = 0.1 # Clustering loss = 0.1
lrate = 0.001 # Learning rate = 0.001 ---- keras defult
import argparse
parser = argparse.ArgumentParser(description='main')
parser.add_argument('--dataset', default=data,
help="Dataset name to train on")
PATH = './results/'
path = PATH + data
if Train_ae:
load = None
else:
load = path + '/ae_weights.h5'
if TEST:
load_test = path + '/model_final.h5'
else:
load_test = None
parser.add_argument('-d', '--save-dir', default=path,
help="Dir to save the results")
# Parameters for pretraining
parser.add_argument('--pretrain_dir', default=load, type=str,
help="Pretrained weights of the autoencoder")
parser.add_argument('--pretrain-epochs', default=epochs, type=int, # 500
help="Number of epochs for pretraining")
parser.add_argument('-v', '--verbose', default=1, type=int,
help="Verbose for pretraining")
# Parameters for clustering
parser.add_argument('--testing', default=TEST, type=bool,
help="Testing the clustering performance with provided weights")
parser.add_argument('--weights', default=load_test, type=str,
help="Model weights, used for testing")
# pretrain_optimizer = 'adam' # adam, sgd
# parser.add_argument('--optimizer', default=pretrain_optimizer, type=str,
# help="Optimizer for clustering phase")
parser.add_argument('--lr', default=lrate, type=float,
help="learning rate during clustering")
parser.add_argument('--batch-size', default=Batch, type=int, # 256
help="Batch size")
parser.add_argument('--maxiter', default=Maxiter, type=int, # 2e4
help="Maximum number of iterations")
parser.add_argument('-uc', '--UpdateCoo', default=Update_Coo, type=int, # 200
help="Number of iterations to update the target distribution")
parser.add_argument('--view_first', default=View, type=int,
help="view-first")
parser.add_argument('--Coo', default=Coo, type=int,
help="Coo?")
parser.add_argument('--K12q', default=K123q, type=int,
help="Kmeans")
parser.add_argument('--Idec', default=Idec, type=float,
help="dec?")
parser.add_argument('--lc', default=lc, type=float,
help="Idec?")
parser.add_argument('--AR', default=AR, type=float,
help="aligned rate?")
parser.add_argument('--ARtime', default=1, type=float,
help="over aligned rate times?")
args = parser.parse_args()
print('+' * 30, ' Parameters ', '+' * 30)
print(args)
print('+' * 75)
# testing
if args.testing:
test(args)
else:
train(args)