-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathreverse_testing.py
154 lines (132 loc) · 5.21 KB
/
reverse_testing.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
import os
import time
import logging
from logging import handlers
from definitions import WEIGHT_DIR
from utils import dataset
# from utils.plot import plot_spectrogram
import numpy as np
import tensorflow as tf
from models.SCNN18 import SCNN18
LOG = logging.getLogger(__name__)
def initLog(debug=False):
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s %(levelname)s %(message)s',
datefmt='%Y-%m-%d %H:%M',
handlers=[logging.StreamHandler(), handlers.RotatingFileHandler('output.log', "w", 1024 * 1024 * 100, 3, "utf-8")]
)
LOG.setLevel(logging.DEBUG if debug else logging.INFO)
tf.get_logger().setLevel('ERROR')
initLog()
dataset_list = [
'SCNN-Jamendo-train.h5',
'SCNN-FMA-C-1-fixed-train.h5',
'SCNN-FMA-C-2-fixed-train.h5',
'SCNN-KTV-train.h5',
'SCNN-Taiwanese-CD-train.h5',
'SCNN-Taiwanese-stream-train.h5',
'SCNN-Chinese-CD-train.h5',
'SCNN-Classical-train.h5',
# ---- Test
'SCNN-Jamendo-test.h5',
'SCNN-FMA-C-1-fixed-test.h5',
'SCNN-FMA-C-2-fixed-test.h5',
'SCNN-KTV-test.h5',
'SCNN-Taiwanese-CD-test.h5',
'SCNN-Taiwanese-stream-test.h5',
'SCNN-Chinese-CD-test.h5',
'SCNN-Classical-test.h5',
# ---- Only have one
'SCNN-MIR-1k-train.h5',
'SCNN-Instrumental-non-vocal.h5',
'SCNN-A-Cappella-vocal.h5',
'SCNN-test-hard.h5',
]
dataset_train_list = [
# 'SCNN-Jamendo-train.h5',
# 'SCNN-FMA-C-1-fixed-train.h5',
# 'SCNN-FMA-C-2-fixed-train.h5',
# 'SCNN-KTV-train.h5',
# 'SCNN-Taiwanese-CD-train.h5',
# 'SCNN-Taiwanese-stream-train.h5',
# 'SCNN-Chinese-CD-train.h5',
# 'SCNN-Classical-train.h5',
# ---- Test
# 'SCNN-Jamendo-test.h5',
# 'SCNN-FMA-C-1-fixed-test.h5',
# 'SCNN-FMA-C-2-fixed-test.h5',
# 'SCNN-KTV-test.h5',
# 'SCNN-Taiwanese-CD-test.h5',
# 'SCNN-Taiwanese-stream-test.h5',
# 'SCNN-Chinese-CD-test.h5',
# 'SCNN-Classical-test.h5',
# ---- Only have one
# 'SCNN-MIR-1k-train.h5',
# 'SCNN-Instrumental-non-vocal.h5',
# 'SCNN-A-Cappella-vocal.h5',
# 'SCNN-test-hard.h5',
'SCNN-RWC.h5',
]
times = 10
# Setting visible of gpus
batch_size = 150
nb_classes = 2
nb_epoch = 160
sample_size = 32000
input_shape = (sample_size, 1)
# Calculat time
start = time.time()
training = True
total_acc = 0
acc_list = []
# Config gpus
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
try:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
except RuntimeError as e:
LOG.error(e)
input_shape = (32000, 1)
classes = 2
lr = 1.0
logging = []
reversed_accs = {key: [] for key in dataset_train_list}
origin_accs = {key: [] for key in dataset_train_list}
def run_test(train_dataset, origin_dataset, times):
for i in range(times):
strategy = tf.distribute.MirroredStrategy(devices=[f'/gpu:{_i}' for _i in range(3)])
with strategy.scope():
model = SCNN18(input_shape, classes).model()
model.compile(loss='categorical_crossentropy', optimizer=tf.optimizers.Adadelta(lr), metrics=['accuracy'])
model.load_weights(os.path.join(WEIGHT_DIR, "FMA-C-1", f"2021-02-13_11_SCNN18_SCNN-FMA-C-1-fixed-train_h5_2GPU-{i}.h5"))
X = dataset.get_dataset_without_label(train_dataset)
Y = np.where(model.predict(dataset.load(train_dataset).batch(batch_size)) >= 0.5, 1, 0)
real_train_ds = dataset.load(train_dataset).batch(batch_size)
train_ds = tf.data.Dataset.from_tensor_slices((X, Y)).batch(batch_size)
# val_ds = dataset.load(validation_dataset).batch(batch_size)
origin_ds = dataset.load(origin_dataset).batch(batch_size)
strategy = tf.distribute.MirroredStrategy(devices=[f'/gpu:{i}' for i in range(3)])
with strategy.scope():
model_retrain = SCNN18(input_shape, classes).model()
model_retrain.compile(loss='categorical_crossentropy', optimizer=tf.optimizers.Adadelta(lr), metrics=['accuracy'])
model_retrain.fit(train_ds, epochs=nb_epoch, validation_data=None)
_reversed_acc = model_retrain.evaluate(origin_ds)[1]
_origin_acc = model.evaluate(real_train_ds)[1]
LOG.info(
f"{origin_dataset.replace('.h5', '')} retrain with pseudo lable {train_dataset.replace('.h5', '')} {i}: {_reversed_acc:.5f}, {_origin_acc:.5f}"
)
reversed_accs[train_dataset].append(_reversed_acc)
origin_accs[train_dataset].append(_origin_acc)
model_retrain.save_weights(
os.path.join(
WEIGHT_DIR, 'reverse_testing_RWC_FMA-C-1',
f"SCNN_{origin_dataset.replace('.h5', '')}_with_pseudo_{train_dataset.replace('.h5', '')}_{i}.h5"
)
)
for train_dataset in dataset_train_list:
run_test(train_dataset, 'SCNN-FMA-C-1-fixed-train.h5', times)
for train_dataset in dataset_train_list:
LOG.info(f"{train_dataset} reversed retrain avg acc: {np.sum(reversed_accs[train_dataset])/len(reversed_accs[train_dataset])}")
LOG.info(f"{train_dataset} avg acc: {np.sum(origin_accs[train_dataset])/len(origin_accs[train_dataset])}")