/
train.py
186 lines (166 loc) · 7.55 KB
/
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
174
175
176
177
178
179
180
181
182
183
184
185
186
import os
import sys
from operator import itemgetter
import sklearn
import sklearn.metrics
import torch
import torch.autograd as autograd
import torch.nn.functional as F
from torch.autograd import Variable
import numpy as np
from eval import eval
def multiclass_metric_loss(represent, target, margin=10, class_num=2, start_idx=1):
target_list = target.data.tolist()
dim = represent.data.shape[1]
indices = []
for class_idx in range(start_idx, class_num + start_idx):
indice_i = [i for i, x in enumerate(target_list) if x == class_idx]
indices.append(indice_i)
loss_intra = Variable(torch.FloatTensor([0])).cuda()
num_intra = 0
loss_inter = Variable(torch.FloatTensor([0])).cuda()
num_inter = 0
for i in range(class_num):
# intra class loss
indice_i = indices[i]
for intra_i in range(len(indice_i)):
for intra_j in range(intra_i + 1, len(indice_i)):
r_i = represent[indice_i[intra_i]]
r_j = represent[indice_i[intra_j]]
dist_ij = (r_i - r_j).norm(2)
loss_intra += 1 / dim * (dist_ij * dist_ij)
num_intra += 1
# inter class loss
for j in range(i + 1, class_num):
indice_j = indices[j]
for inter_i in indice_i:
for inter_j in indice_j:
r_i = represent[inter_i]
r_j = represent[inter_j]
dist_ik = (r_i - r_j).norm(2)
tmp = margin - 1 / dim * (dist_ik * dist_ik)
loss_inter += torch.clamp(tmp, min=0)
num_inter += 1
if num_intra > 0:
loss_intra = loss_intra / num_intra
if num_inter > 0:
loss_inter = loss_inter / num_inter
return loss_intra, loss_inter
def metric_loss(represent, target, margin):
target_list = target.data.tolist()
dim = represent.data.shape[1]
indices = [i for i, x in enumerate(target_list) if x == 1]
other_indices = list(set(range(0, len(target_list))) - set(indices))
# no label 1 instances
if len(indices) == 0:
return Variable(torch.FloatTensor([0])).cuda(), Variable(torch.FloatTensor([0])).cuda()
loss_intra = Variable(torch.FloatTensor([0])).cuda()
num_intra = 0
for i in range(len(indices)):
for j in range(i + 1, len(indices)):
r_i = represent[indices[i]]
r_j = represent[indices[j]]
dist_ij = (r_i - r_j).norm(2)
loss_intra += 1 / dim * (dist_ij * dist_ij)
num_intra += 1
if num_intra > 0:
loss_intra = loss_intra / num_intra
loss_inter = Variable(torch.FloatTensor([0])).cuda()
num_inter = 0
for i in indices:
for k in other_indices:
r_i = represent[i]
r_k = represent[k]
dist_ik = (r_i - r_k).norm(2)
tmp = margin - 1 / dim * (dist_ik * dist_ik)
loss_inter += torch.clamp(tmp, min=0)
num_inter += 1
if num_inter > 0:
loss_inter = loss_inter / num_inter
return loss_intra, loss_inter
def train(dataset, x_train, y_train, x_val, y_val, model, args):
if args.cuda:
print("training model in cuda ...")
model.cuda()
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
steps = 0
best_acc = 0
last_step = 0
model.train()
print(model)
for epoch in range(1, args.epochs+1):
train_iter = dataset.gen_minibatch(x_train, y_train, args.batch_size, args, shuffle=True)
for batch in train_iter:
feature, target = batch[0], batch[1]
#feature.data.t_(), target.data.sub_(1) # batch first, index align
if args.cuda:
feature, target = feature.cuda(), target.cuda()
optimizer.zero_grad()
logit, represent = model(feature)
loss_target = F.cross_entropy(logit, target)
loss_metric = Variable(torch.FloatTensor([0])).cuda()
if args.metric:
class_num = dataset.get_class_num()
if class_num == 2:
#loss_intra, loss_inter = metric_loss(represent, target, margin=args.metric_margin)
loss_intra, loss_inter = multiclass_metric_loss(represent, target, margin=args.metric_margin,
class_num=class_num, start_idx=0)
elif class_num > 2:
loss_intra, loss_inter = multiclass_metric_loss(represent, target, margin=args.metric_margin, class_num=class_num)
# try:
#
# except:
# loss_intra, loss_inter = metric_loss(represent, target, margin=10)
# a = 0
loss_metric = loss_intra + loss_inter
loss = loss_target + args.metric_param * loss_metric
#print('logit vector', logit.size())
#print('target vector', target.size())
loss.backward()
optimizer.step()
steps += 1
if steps % args.log_interval == 0:
corrects = (torch.max(logit, 1)[1].view(target.size()).data == target.data).sum()
accuracy = 100.0 * corrects / args.batch_size
try:
if args.metric:
sys.stdout.write(
'\rEpoch[{}] Batch[{}] - loss: {:.4f}'
'({:.4f}/{:.4f}) acc: {:.4f}%({}/{})'.format(epoch,
steps,
loss.data[0],
loss_target.data[0],
loss_metric.data[0],
accuracy,
corrects,
args.batch_size))
else:
sys.stdout.write(
'\rEpoch[{}] Batch[{}] - loss: {:.4f} '
'acc: {:.4f}%({}/{})'.format(epoch,
steps,
loss.data[0],
accuracy,
corrects,
args.batch_size))
except:
print("Unexpected error:", sys.exc_info()[0])
if steps % args.test_interval == 0:
dev_acc = eval(dataset, x_val, y_val, model, args)
if dev_acc > best_acc:
best_acc = dev_acc
last_step = steps
if args.save_best:
save(model, args.save_dir, 'best', steps)
else:
if steps - last_step >= args.early_stop:
print('early stop by {} steps.'.format(args.early_stop))
# if steps % args.save_interval == 0:
# save(model, args.save_dir, 'snapshot', steps)
model.train()
def save(model, save_dir, save_prefix, steps):
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
save_prefix = os.path.join(save_dir, save_prefix)
save_path = '{}_steps_{}.pt'.format(save_prefix, steps)
torch.save(model.state_dict(), save_path)