-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_svhn2mnist.py
133 lines (116 loc) · 7.39 KB
/
train_svhn2mnist.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
from __future__ import print_function
import argparse
import data_loader
import torch
import os
import copy
import math
from utils import *
from digit_models.I2Attack import run_I2Attack
from digit_models.MDD import MDDModel
from digit_models.DANN import DANNModel
from digit_models.DAN import DANModel
# Command setting
parser = argparse.ArgumentParser(description='Domain Adaptation')
parser.add_argument('-model', type=str, default='DAN', help='model name')
parser.add_argument('-mode', type=str, default='poison', help='poison|clean')
parser.add_argument('-batch_size', type=int, default=64, help='batch size')
parser.add_argument('-test_batch_size', type=int, default=500, help='test batch size')
parser.add_argument('-cuda', type=int, default=1, help='cuda id')
parser.add_argument('-root_dir', type=str, default='E:/Codes/data/svhn2mnist/')
parser.add_argument('-source', type=str, default='svhn')
parser.add_argument('-target', type=str, default='mnist')
parser.add_argument('-epochs', type=int, default=10000)
parser.add_argument('-num_classes', type=int, default=10)
parser.add_argument('-lr', type=float, default=0.01)
parser.add_argument('-moment', type=float, default=0.9)
parser.add_argument('-l2_decay', type=float, default=5e-4)
args = parser.parse_args()
def train(src_data, tgt_data, tgt_test_data, device):
model_name = args.model
model = None
print(model_name)
if model_name == 'MDD':
model = MDDModel(num_classes=args.num_classes).to(device)
elif model_name == 'DAN':
model = DANModel(num_classes=args.num_classes).to(device)
elif model_name == 'DANN':
model = DANNModel(num_classes=args.num_classes).to(device)
optimizer = get_optimizer(model, args)
if args.mode == 'poison':
src_data = run_I2Attack(copy.deepcopy(model), src_data, tgt_data, device, eps=0.1, step_size=0.01, attack_epochs=25)
print("Poisoning source is done")
src_generator = batch_generator(src_data, batch_size=args.batch_size)
tgt_generator = batch_generator(tgt_data, batch_size=args.batch_size)
for i in range(args.epochs):
model.train()
learning_rate = args.lr / math.pow((1 + 10 * i / args.epochs), 0.75)
adjust_learning_rate(optimizer, learning_rate)
sinputs, slabels = next(src_generator)
tinputs, _ = next(tgt_generator)
sinputs = torch.tensor(sinputs, requires_grad=False, dtype=torch.float).to(device)
slabels = torch.tensor(slabels, requires_grad=False, dtype=torch.long).to(device)
tinputs = torch.tensor(tinputs, requires_grad=False, dtype=torch.float).to(device)
optimizer.zero_grad()
alpha = 2 / (1 + math.exp(-10 * i / args.epochs)) - 1
loss = model(sinputs, slabels, tinputs, alpha)
loss.backward()
optimizer.step()
# print('Epoch: [{:02d}/{:02d}], loss: {:.6f}'.format(i + 1, args.epochs, loss.item()))
if (i + 1) % 1000 == 0:
model.eval()
test_acc = 0.
with torch.no_grad():
test_len = tgt_test_data['X'].shape[0] // args.test_batch_size
for j in range(test_len):
outputs = model.inference(torch.tensor(tgt_test_data['X'][args.test_batch_size * j:args.test_batch_size * (j + 1)], requires_grad=False).to(device))
preds = torch.max(outputs, 1)[1]
test_acc += torch.sum(preds == torch.tensor(tgt_test_data['Y'][args.test_batch_size * j:args.test_batch_size * (j + 1)], requires_grad=False, dtype=torch.long).to(device))
if test_len * args.test_batch_size < tgt_test_data['X'].shape[0]:
outputs = model.inference(torch.tensor(tgt_test_data['X'][test_len * args.test_batch_size:], requires_grad=False).to(device))
preds = torch.max(outputs, 1)[1]
test_acc += torch.sum(preds == torch.tensor(tgt_test_data['Y'][test_len * args.test_batch_size:], requires_grad=False, dtype=torch.long).to(device))
test_acc = test_acc.double() / tgt_test_data['X'].shape[0]
with torch.no_grad():
train_acc = 0.
train_len = src_data['X'].shape[0] // args.test_batch_size
for j in range(train_len):
outputs = model.inference(torch.tensor(src_data['X'][args.test_batch_size * j:args.test_batch_size * (j + 1)], requires_grad=False).to(device))
preds = torch.max(outputs, 1)[1]
train_acc += torch.sum(preds == torch.tensor(src_data['Y'][args.test_batch_size * j:args.test_batch_size * (j + 1)], requires_grad=False, dtype=torch.long).to(device))
if train_len * args.test_batch_size < src_data['X'].shape[0]:
outputs = model.inference(torch.tensor(src_data['X'][train_len * args.test_batch_size:], requires_grad=False).to(device))
preds = torch.max(outputs, 1)[1]
train_acc += torch.sum(preds == torch.tensor(src_data['Y'][train_len * args.test_batch_size:], requires_grad=False, dtype=torch.long).to(device))
train_acc = train_acc.double() / src_data['X'].shape[0]
with torch.no_grad():
discrepancy = 0.
num_examples = 500
num_batchs = min(src_data['X'].shape[0], tgt_data['X'].shape[0]) // num_examples
for j in range(num_batchs):
s_val_inputs = torch.tensor(src_data['X'][num_examples * j: num_examples * (j + 1)], requires_grad=False).to(device)
t_val_inputs = torch.tensor(tgt_data['X'][num_examples * j: num_examples * (j + 1)], requires_grad=False).to(device)
discrepancy += model.get_discrepancy(s_val_inputs, t_val_inputs)
discrepancy = discrepancy / num_batchs
print('Epoch: [{:02d}/{:02d}], loss: {:.6f}, train acc: {:.4f}, discrepancy: {:.4f}, test acc: {:.4f}'.format(i + 1, args.epochs, loss.item(), train_acc, discrepancy, test_acc))
return test_acc
if __name__ == '__main__':
device = torch.device('cuda:' + str(args.cuda) if torch.cuda.is_available() else 'cpu')
torch.manual_seed(23)
torch.cuda.manual_seed(23)
if not os.path.isfile("data_preprocessed/{}.pkl".format(args.source)):
if args.source == 'svhn':
raw_src_loader = data_loader.load_training_svhn2mnist(args.root_dir, args.source + '.txt')
else:
raw_src_loader = data_loader.load_training_svhn2mnist(args.root_dir, args.source + '_train.txt')
save_data(raw_src_loader, name=args.source)
src_data = pickle.load(open("data_preprocessed/{}.pkl".format(args.source), "rb"))
if not os.path.isfile("data_preprocessed/{}.pkl".format(args.target)):
raw_tgt_loader = data_loader.load_training_svhn2mnist(args.root_dir, args.target + '_train.txt')
save_data(raw_tgt_loader, name=args.target)
tgt_data = pickle.load(open("data_preprocessed/{}.pkl".format(args.target), "rb"))
if not os.path.isfile("data_preprocessed/{}_test.pkl".format(args.target)):
raw_tgt_loader = data_loader.load_testing_svhn2mnist(args.root_dir, args.target + '_test.txt')
save_data(raw_tgt_loader, name=args.target+'_test')
tgt_test_data = pickle.load(open("data_preprocessed/{}_test.pkl".format(args.target), "rb"))
test_acc = train(src_data, tgt_data, tgt_test_data, device)