Repository navigation
Expand file tree
/
Copy pathtrain.py
More file actions
144 lines (125 loc) · 5.29 KB
/
Copy pathtrain.py
File metadata and controls
144 lines (125 loc) · 5.29 KB
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
import os
import pickle
import argparse
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from utils.models import *
import warnings
warnings.filterwarnings("ignore")
def save_model(model, dataset, index_to_remove, output_dir):
a = []
for i in model.parameters():
a.append(np.transpose(i.cpu().detach().numpy()))
print("---------------", index_to_remove, "-----------------")
for i in a:
print(i.shape)
if index_to_remove >= 0:
model_path = "./"+output_dir+"/"+dataset+"_"+str(index_to_remove)
else:
model_path = "./" + output_dir + "/" + dataset
pickle.dump(a, open(model_path + ".p", "wb"))
torch.save(model.state_dict(), model_path + '.pth')
def load_model(model_arch, dim):
if model_arch == "2x10":
model = FC_2x10(dim)
elif model_arch == "2x50":
model = FC_2x50(dim)
elif model_arch == "2x100":
model = FC_2x100(dim)
elif model_arch == "4x30":
model = FC_4x30(dim)
elif model_arch == "CNN":
model = CNN(dim)
else:
assert ("New model arch has been detected, please expand models.py and this if condition.")
return model
def load_dataset(dataset):
if dataset == "adult":
dim = 21
elif dataset == "credit":
dim = 23
elif dataset == "crypto":
dim = 7
elif dataset == "twitter":
dim = 15
else:
assert False, "New dataset has been detected, please expand the code to support it."
train_set = torch.load('./datasets/' + dataset + '/train.pth')
test_set = torch.load('./datasets/' + dataset + '/test.pth')
return train_set, test_set, dim
def train(index_to_remove, dataset, model_arch, seed, output_dir, device='cpu', batch_size=1024, num_epochs=30, num_of_threads = 1, log=False):
torch.set_num_threads(num_of_threads)
train_set, test_set, dim = load_dataset(dataset)
np.random.seed(seed)
torch.manual_seed(seed)
model = load_model(model_arch, dim)
model = model.to(device)
if index_to_remove >= len(train_set):
return
elif index_to_remove >= 0:
x, _ = train_set[index_to_remove]
train_set[index_to_remove] = (x, torch.tensor(-1, dtype=torch.int64))
train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True)
test_loader = DataLoader(test_set, batch_size=batch_size, shuffle=False)
loss = nn.CrossEntropyLoss().to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=0.1, weight_decay=1e-5)
for epoch in range(num_epochs):
total_batch = len(train_loader) // batch_size
for i, (samples, labels) in enumerate(train_loader):
mask = labels != -1
X = samples[mask].to(device)
Y = labels[mask].to(device)
pre = model(X)
cost = loss(pre, Y.long())
optimizer.zero_grad()
cost.backward()
optimizer.step()
if (i + 1) % 30 == 0 and log:
print('Epoch [%d/%d], lter [%d/%d], Loss: %.4f' % (
epoch + 1, num_epochs, i + 1, total_batch, cost.item()))
with torch.no_grad():
model.eval()
correct = 0
total = 0
for images, labels in test_loader:
images = images.to(device)
outputs = model(images).to(device)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels.to(device)).sum()
print('Test accuracy: %.2f %%' % (100 * float(correct) / total))
try:
os.mkdir(output_dir)
except:
print("Folder exists.")
save_model(model, dataset, index_to_remove, output_dir)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--model_arch', type=str, default="2x50", help='2x50, 2x100, 4x30, or CNN')
parser.add_argument('--output_dir', type=str, default="./model/", help='output directory to save the models')
parser.add_argument('--dataset', type=str, default="adult", help='dataset: adult, credit, crypto, or twitter')
parser.add_argument('--seed', type=int, default=666, help='random seed 42 or 666')
parser.add_argument('--start', type=int, default=0, help='start model')
parser.add_argument('--end', type=int, default=1, help='end model')
parser.add_argument('--device', type=str, default="cpu", help='device to train with')
parser.add_argument('--batch_size', type=int, default=1024, help='batch size')
parser.add_argument('--epochs', type=int, default=30, help='number of epochs')
parser.add_argument('--num_of_threads', type=int, default=1, help='number of threads')
args = parser.parse_args()
batch_size = args.batch_size
num_epochs = args.epochs
output_dir = args.output_dir
model_arch = args.model_arch
dataset = args.dataset
seed = args.seed
start = args.start
end = args.end
device = args.device
num_of_threads = args.num_of_threads
if start == 0:
train(-1, dataset, model_arch, seed, output_dir, device, batch_size, num_epochs, num_of_threads)
for j in range(start, end):
index_to_remove = j
train(index_to_remove, dataset, model_arch, seed, output_dir, device, batch_size, num_epochs, num_of_threads)