-
Notifications
You must be signed in to change notification settings - Fork 0
/
Prior_train.py
179 lines (160 loc) · 7.07 KB
/
Prior_train.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
"""
Created on Thu Dec 14 10:12:13 2023
@author: Shulei Ji
"""
import numpy as np
import argparse
import torch
import torch.nn as nn
import time
import os
import pickle
from models.MusER_TRANS_CA_GE import VQ_VAE
from models.Prior import VQ_prior
from torch.utils.data import DataLoader,TensorDataset
from torch.nn.utils import clip_grad_norm_
from utils import timeSince,setup_seed
device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
def get_from_pretrained_encoder(model,data_loader):
index=[]
emotion=[]
index_dict={}
for i, prior_data in enumerate(data_loader):
train_x, train_y=prior_data
indices=model.prior(train_y)
indices=indices.view(train_y.shape[0],-1)
index.append(indices)
emo=train_x[:,0,:].narrow(1,7,1)
emotion.append(emo)
index=torch.cat(index,dim=0)
for i in range(index.shape[0]):
for j in range(index.shape[1]):
index[i][j]=VQ_dict[0][index[i][j].item()]
padding=torch.zeros((index.shape[0],1)).to(device)
emotion=torch.cat(emotion, dim=0)+len(VQ_dict[0])
index_dict['x'] = torch.cat((emotion,index), dim=1)
index_dict['y'] = torch.cat((index,padding), dim=1)
return index_dict
def train(input_x,input_y,is_train):
if is_train=="train":
model.train()
else:
model.eval()
output=model(input_x,temperature=0.3)
topk,topi=output.topk(1)
topi=topi.squeeze(-1)
loss=0;acc=0
for i in range(args.batch_size):
loss += criterion(output[i], input_y[i])
acc_temp=0
for j in range(len(input_y[i])):
if topi[i][j]==input_y[i][j]:
acc_temp+=1
acc += acc_temp / len(input_y[i])
acc = acc/args.batch_size
loss = loss/args.batch_size
if is_train == "train":
optimizer.zero_grad()
loss.backward()
clip_grad_norm_(model.parameters(), 3)
optimizer.step()
return loss.item(),acc
def trainIter():
max_test_loss=10
for epoch in range(epoch_already+1,args.Epoch):
f = open(args.log_path, 'a')
print("-----------------------------epoch ", epoch, "------------------------------")
f.write('-----------------------------epoch %d------------------------------\n' % (epoch))
total_loss = 0
train_total_loss = 0
total_acc=0
for i, data in enumerate(train_loader):
input_x,input_y = data
loss,acc = train(input_x,input_y,"train")
total_loss += loss
train_total_loss += loss
total_acc += acc
print('epoch: %d, time: %s, train loss: %.6f, train acc: %.6f' %
(epoch, timeSince(start_time),train_total_loss / (i+1), total_acc/(i+1)))
f.write('epoch: %d, time: %s, train loss: %.6f, train acc: %.6f\n' %
(epoch, timeSince(start_time),train_total_loss / (i+1), total_acc/(i+1)))
loss_average=train_total_loss/(i+1)
if loss_average < max_test_loss:
print("epoch: %d save min test loss model-->test loss: %.6f" % (epoch, loss_average))
f.write('epoch: %d save min test loss model-->test loss: %.6f\n' % (epoch, loss_average))
model_save_path = args.model_path + f"{epoch}_best.pth"
state = {'model': model.state_dict(), 'optimizer': optimizer.state_dict(), 'epoch': epoch}
torch.save(state, model_save_path)
max_test_loss = loss_average
f.close()
if __name__=='__main__':
setup_seed(13)
parser = argparse.ArgumentParser()
parser.add_argument("--Epoch", type=int, default=1300)
parser.add_argument("--batch_size", type=int, default=16)
parser.add_argument("--encoder_width", type=int, default=256)
parser.add_argument("--learning_rate", type=float, default=1e-4)
parser.add_argument("--transformer_N", type=int, default=8)
parser.add_argument("--multihead_N", type=int, default=8)
parser.add_argument("--patience", type=int, default=100)
parser.add_argument("--print_every", type=int, default=240)
parser.add_argument("--dropout", type=float, default=0.1)
parser.add_argument("--encoder_attention_type", type=str, default='causal-linear')
parser.add_argument("--activate", type=str, default='gelu')
parser.add_argument("--dataset", type=str, default='emopia')
parser.add_argument("--data_path", type=str, default='./datafile/co-representation/emopia_data.npz')
parser.add_argument("--VQ_VAE", type=str,default='MusER_TRANS_CA_GE_emopia')
parser.add_argument("--load_VQ_prior", type=str,default='')
parser.add_argument("--model_path", type=str, default='./saved_models/Prior_MusER_TRANS_CA_GE_emopia/')
parser.add_argument("--log_path", type=str, default='./logs/Prior_MusER_TRANS_CA_GE_emopia.txt')
args = parser.parse_args()
# print params
f = open(args.log_path, 'a')
f.write("-----------------------------------------\n")
for arg in vars(args):
f.write(str(arg)+"="+str(getattr(args,arg))+"\n")
f.close()
# prepare data
VQ_dict_file = open(f'data/{args.VQ_VAE}_VQ_dict.data', 'rb')
VQ_dict = pickle.load(VQ_dict_file)
print(VQ_dict)
with torch.no_grad():
data = np.load(args.data_path)
train_x = data['x']
train_y = data['y']
train_data_x = torch.LongTensor(train_x).to(device)
train_data_y = torch.LongTensor(train_y).to(device)
train_dataset = TensorDataset(train_data_x,train_data_y)
train_loader = DataLoader(dataset=train_dataset, batch_size=args.batch_size, shuffle=False, num_workers=0,
drop_last=False)
VQ_VAE_model=VQ_VAE(8, 8, 128, 256, 512, 112, 0.1, 'gelu', 'linear', 'causal-linear').to(device)
if args.VQ_VAE!='':
VQ_VAE_path=f"./saved_models/{args.VQ_VAE}/best.pt"
model_dict=torch.load(VQ_VAE_path,map_location=device)
VQ_VAE_model.load_state_dict(model_dict['model'])
VQ_VAE_model=VQ_VAE_model.eval()
data = get_from_pretrained_encoder(VQ_VAE_model, train_loader)
train_x=data['x'].long()
train_y=data['y'].long()
train_data_num=train_x.shape[0]
train_dataset = TensorDataset(train_x,train_y)
train_loader = DataLoader(dataset=train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=0,
drop_last=True)
# load model
model=VQ_prior(args.transformer_N, args.multihead_N, len(VQ_dict[0]), args.encoder_width,
args.dropout, args.activate,args.encoder_attention_type).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate)
criterion = nn.NLLLoss().to(device)
epoch_already=-1
if args.load_VQ_Prior!='':
VQ_prior_path={args.model_path}+{args.VQ_Prior}
model_dict = torch.load(VQ_prior_path, map_location=device)
model.load_state_dict(model_dict['model'])
model = model.to(device)
optimizer.load_state_dict(model_dict['optimizer'])
epoch_already = model_dict['epoch']
# begin training
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
start_time = time.time()
trainIter()