-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathmain_tva_1.py
153 lines (147 loc) · 7.41 KB
/
main_tva_1.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
import torch
import numpy as np
import argparse
from data_utils import *
from torch.utils.data import DataLoader
from torchnet.dataset import TensorDataset
import train_tva_1
import random
if __name__ == '__main__':
# get arguments
p = argparse.ArgumentParser()
p.add_argument('--seed', type=int, default=1)
p.add_argument('--data_path', type=str, default='/processed_Yoon/IEMOCAP/seven_category_120/folds/fold01')
p.add_argument('--batch_size', type=int, default=32)
p.add_argument('--lr', type=float, default=1e-3)
p.add_argument('--rnntype', type=str, default='gru')
p.add_argument('--rnndir', type=str, default=True,
help='Uni (False) or Bi (True) directional')
p.add_argument('--rnnsize', type=int, default=60)#30)#200
# video params
p.add_argument('--vid_rnnnum', type=int, default=1)#1)#3
p.add_argument('--vid_rnndp', type=int, default=0.3)#0.3
p.add_argument('--vid_rnnsize', type=int, default=60)
p.add_argument('--vid_nh', type=int, default=6,
help='number of attention heads for mha')#4
p.add_argument('--vid_dp', type=int, default=0.1,
help='dropout rate for mha')#0.1
# text params
p.add_argument('--txt_rnnnum', type=int, default=1)
p.add_argument('--txt_rnndp', type=int, default=0.3)#0.3
p.add_argument('--txt_rnnsize', type=int, default=60)
p.add_argument('--txt_nh', type=int, default=6,
help='number of attention heads for mha')#4
p.add_argument('--txt_dp', type=int, default=0.1,
help='dropout rate for mha')#0.1
# audio params
p.add_argument('--aud_rnnnum', type=int, default=1)
p.add_argument('--aud_rnndp', type=int, default=0.3) # 0.3
p.add_argument('--aud_rnnsize', type=int, default=60)
p.add_argument('--aud_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--aud_dp', type=int, default=0.1,
help='dropout rate for mha') # 0.1
# tv params
p.add_argument('--tv_nh', type=int, default=6,
help='number of attention heads for mha')#4
p.add_argument('--tv_dp', type=int, default=0.1,
help='dropout rate for mha')#0.1
# ta params
p.add_argument('--ta_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--ta_dp', type=int, default=0.1,
help='dropout rate for mha') # 0.1
# vt params
p.add_argument('--vt_nh', type=int, default=6,
help='number of attention heads for mha')#4
p.add_argument('--vt_dp', type=int, default=0.1,
help='dropout rate for mha')
# va params
p.add_argument('--va_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--va_dp', type=int, default=0.1,
help='dropout rate for mha')
# at params
p.add_argument('--at_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--at_dp', type=int, default=0.1,
help='dropout rate for mha')
# av params
p.add_argument('--av_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--av_dp', type=int, default=0.1,
help='dropout rate for mha')
# tf params
p.add_argument('--tf_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--tf_dp', type=int, default=0.1,
help='dropout rate for mha')
# vf params
p.add_argument('--vf_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--vf_dp', type=int, default=0.1,
help='dropout rate for mha')
# af params
p.add_argument('--af_nh', type=int, default=6,
help='number of attention heads for mha') # 4
p.add_argument('--af_dp', type=int, default=0.1,
help='dropout rate for mha')
p.add_argument('--output_dim', type=int, default=7,
help='number of classes')
p.add_argument('--optim', type=str, default='Adam',
help='optimizer to use (default: Adam)')
params = p.parse_args()
#seed = 123
torch.manual_seed(params.seed)
torch.cuda.manual_seed(params.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
random.seed(params.seed)
# get train data
from sklearn.preprocessing import StandardScaler
scaler_mfcc = StandardScaler()
x_text, x_vid, vid_seq, x_mfcc, x_pros, aud_seq, labels = get_text_video_audio_data(params.data_path, 'train')
s1 = x_mfcc.shape[1]
s2 = x_mfcc.shape[2]
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], -1])
scaler_mfcc.fit(x_mfcc)
scaler_mfcc.transform(x_mfcc)
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], s1, s2])
train_dataset = TensorDataset([torch.Tensor(x_text).float().to('cuda'), torch.Tensor(x_vid).float().to('cuda'),
torch.Tensor(vid_seq).int().to('cuda'), torch.Tensor(x_mfcc).float().to('cuda'),
torch.Tensor(x_pros).float().to('cuda'), torch.Tensor(aud_seq).int().to('cuda'),
torch.Tensor(labels).long().to('cuda')])
train_loader = DataLoader(train_dataset, batch_size=params.batch_size, shuffle=True)
params.n_train = len(x_text)
# get dev data
x_text, x_vid, vid_seq, x_mfcc, x_pros, aud_seq, labels = get_text_video_audio_data(params.data_path, 'dev')
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], -1])
scaler_mfcc.transform(x_mfcc)
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], s1, s2])
dev_dataset = TensorDataset([torch.Tensor(x_text).float().to('cuda'), torch.Tensor(x_vid).float().to('cuda'),
torch.Tensor(vid_seq).int().to('cuda'), torch.Tensor(x_mfcc).float().to('cuda'),
torch.Tensor(x_pros).float().to('cuda'), torch.Tensor(aud_seq).int().to('cuda'),
torch.Tensor(labels).long().to('cuda')])
dev_loader = DataLoader(dev_dataset, batch_size=params.batch_size, shuffle=False)
params.n_dev = len(x_text)
# get test data
x_text, x_vid, vid_seq, x_mfcc, x_pros, aud_seq, labels = get_text_video_audio_data(params.data_path, 'test')
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], -1])
scaler_mfcc.transform(x_mfcc)
x_mfcc = np.reshape(x_mfcc, [x_mfcc.shape[0], s1, s2])
test_dataset = TensorDataset([torch.Tensor(x_text).float().to('cuda'), torch.Tensor(x_vid).float().to('cuda'),
torch.Tensor(vid_seq).int().to('cuda'), torch.Tensor(x_mfcc).float().to('cuda'),
torch.Tensor(x_pros).float().to('cuda'), torch.Tensor(aud_seq).int().to('cuda'),
torch.Tensor(labels).long().to('cuda')])
test_loader = DataLoader(test_dataset, batch_size=params.batch_size, shuffle=False)
params.n_test = len(x_text)
# train
params.num_epochs = 20000 # give a random big number
params.when = 10 # reduce LR patience
params.txt_dim = 300
params.vid_dim = 2048
params.aud_dim = 120
params.pros_dim = 35
count = 0
import sys
test_loss = train_tva_1.initiate(params, train_loader, dev_loader, test_loader)