-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_ood_metaWRN.py
304 lines (268 loc) · 13.1 KB
/
train_ood_metaWRN.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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import os
# os.environ['CUDA_VISIBLE_DEVICES'] = "9"
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.utils.data import DataLoader
import numpy as np
import argparse, math, time, json, os
import random
from lib import transform, meta_module_wrn
from config import config
from datetime import datetime
# datetime object containing current date and time
now = datetime.now()
print("now =", now)
# dd/mm/YY H:M:S
dt_string = now.strftime("%d/%m/%Y %H:%M:%S")
print("date and time =", dt_string)
seed = 123
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
parser = argparse.ArgumentParser()
parser.add_argument("--alg", "-a", default="VAT", type=str, help="ssl algorithm : [supervised, PI, MT, VAT, PL, ICT]")
parser.add_argument("--em", default=0, type=float, help="coefficient of entropy minimization. If you try VAT + EM, set 0.06")
parser.add_argument("--validation", default=5000, type=int, help="validate at this interval (default 25000)")
parser.add_argument("--dataset", "-d", default="cifar10", type=str, help="dataset name : [svhn, cifar10]")
parser.add_argument("--root", "-r", default="data", type=str, help="dataset dir")
parser.add_argument("--output", "-o", default="./exp_res", type=str, help="output dir")
parser.add_argument("--ood_ratio", default=0.5, type=float, help="ood ratio for unlabled set")
parser.add_argument("--n_label", default='class6', type=str, help="number of labeled set")
parser.add_argument("--meta_val_batch", default=50, type=int, help="batch for meta val")
parser.add_argument("--L1_trade_off", default=0e-07, type=float, help="L1_trade_off param")
parser.add_argument("--w_clip", default=1e-5, type=float, help="")
parser.add_argument("--meta_lr", default=0.01, type=float, help="learning rate for w")
parser.add_argument("--cluster", default=1, type=int, help="cluster or not")
parser.add_argument("--w_initial", default=0.05, type=float, help="weight initial")
args = parser.parse_args()
print(args)
if torch.cuda.is_available():
device = "cuda"
torch.backends.cudnn.benchmark = True
else:
device = "cpu"
condition = {}
exp_name = ""
print("dataset : {}".format(args.dataset))
condition["dataset"] = args.dataset
exp_name += str(args.dataset) + "_"
dataset_cfg = config[args.dataset]
transform_fn = transform.transform(*dataset_cfg["transform"]) # transform function (flip, crop, noise)
l_train_dataset = dataset_cfg["dataset"](args.root, args.n_label, "l_train")
if args.ood_ratio == 0:
u_train_dataset = dataset_cfg["dataset"](args.root, args.n_label, "u_train_in")
else:
# u_train_dataset = dataset_cfg["dataset"](args.root, args.n_label, "u_train_new_svhn_{}".format(args.ood_ratio))
u_train_dataset = dataset_cfg["dataset"](args.root, args.n_label, "u_train_ood_{}".format(args.ood_ratio))
# u_train_dataset = dataset_cfg["dataset"](args.root, args.n_label, "u_train_ood_google_{}".format(args.ood_ratio))
val_dataset = dataset_cfg["dataset"](args.root, args.n_label, "val")
test_dataset = dataset_cfg["dataset"](args.root, args.n_label, "test")
print("labeled data : {}, unlabeled data : {}, training data : {}".format(
len(l_train_dataset), len(u_train_dataset), len(l_train_dataset)+len(u_train_dataset)))
print("validation data : {}, test data : {}".format(len(val_dataset), len(test_dataset)))
condition["number_of_data"] = {
"labeled":len(l_train_dataset), "unlabeled":len(u_train_dataset),
"validation":len(val_dataset), "test":len(test_dataset)
}
class RandomSampler(torch.utils.data.Sampler):
""" sampling without replacement """
def __init__(self, num_data, num_sample):
iterations = num_sample // num_data + 1
self.indices = torch.cat([torch.randperm(num_data) for _ in range(iterations)]).tolist()[:num_sample]
def __iter__(self):
return iter(self.indices)
def __len__(self):
return len(self.indices)
shared_cfg = config["shared"]
if args.alg != "supervised":
# batch size = 0.5 x batch size
l_loader = DataLoader(
l_train_dataset, shared_cfg["batch_size"]//2, drop_last=True,
sampler=RandomSampler(len(l_train_dataset), shared_cfg["iteration"] * shared_cfg["batch_size"]//2)
)
else:
l_loader = DataLoader(
l_train_dataset, shared_cfg["batch_size"], drop_last=True,
sampler=RandomSampler(len(l_train_dataset), shared_cfg["iteration"] * shared_cfg["batch_size"])
)
print("algorithm : {}".format(args.alg))
condition["algorithm"] = args.alg
exp_name += str(args.alg) + "_"
u_loader = DataLoader(
u_train_dataset, shared_cfg["batch_size"]//2, drop_last=True,
sampler=RandomSampler(len(u_train_dataset), shared_cfg["iteration"] * shared_cfg["batch_size"]//2)
)
val_loader_in_train = DataLoader(val_dataset, args.meta_val_batch, drop_last=True,
sampler=RandomSampler(len(val_dataset), shared_cfg["iteration"] * args.meta_val_batch))
val_loader = DataLoader(val_dataset, 64, shuffle=False, drop_last=False)
test_loader = DataLoader(test_dataset, 64, shuffle=False, drop_last=False)
print("maximum iteration : {}".format(min(len(l_loader), len(u_loader))))
alg_cfg = config[args.alg]
print("parameters : ", alg_cfg)
condition["h_parameters"] = alg_cfg
if args.em > 0:
print("entropy minimization : {}".format(args.em))
exp_name += "em_"
condition["entropy_maximization"] = args.em
model = meta_module_wrn.WideResNet(28, dataset_cfg["num_classes"], 2, transform_fn).to(device)
optimizer = optim.Adam(model.params(), lr=alg_cfg["lr"])
trainable_paramters = sum([p.data.nelement() for p in model.params()])
print("trainable parameters : {}".format(trainable_paramters))
if args.alg == "VAT": # virtual adversarial training
from lib.algs.vat import VAT_base
ssl_obj = VAT_base(alg_cfg["eps"][args.dataset], alg_cfg["xi"], 1)
elif args.alg == "PL": # pseudo label
from lib.algs.pseudo_label import PL2
ssl_obj = PL2(alg_cfg["threashold"])
elif args.alg == "MT": # mean teacher
from lib.algs.mean_teacher import MT2
t_model = meta_module_wrn.WideResNet(28, dataset_cfg["num_classes"], 2, transform_fn).to(device)
t_model.load_state_dict(model.state_dict())
ssl_obj = MT2(t_model, alg_cfg["ema_factor"])
elif args.alg == "PI": # PI Model
from lib.algs.pimodel import PiModel2
ssl_obj = PiModel2()
elif args.alg == "ICT": # interpolation consistency training
from lib.algs.ict import ICT
t_model = meta_module_wrn.WideResNet(28, dataset_cfg["num_classes"], 2, transform_fn).to(device)
t_model.load_state_dict(model.state_dict())
ssl_obj = ICT(alg_cfg["alpha"], t_model, alg_cfg["ema_factor"])
elif args.alg == "MM": # MixMatch
from lib.algs.mixmatch import MixMatch
ssl_obj = MixMatch(alg_cfg["T"], alg_cfg["K"], alg_cfg["alpha"])
elif args.alg == "supervised":
pass
else:
raise ValueError("{} is unknown algorithm".format(args.alg))
if args.ood_ratio == 0.25:
ood_idx = [1,2,9,4,3,8,0]
in_idx = [6,5,7]
elif args.ood_ratio == 0.5:
ood_idx = [2, 4, 3, 9, 1, 6]
in_idx = [5, 0, 7, 8]
else:
ood_idx = [2, 7, 1]
in_idx = [0, 3, 4, 5, 6, 8, 9]
emd_label = np.load('output/emd/cifar10/{}_{}_kmean_{}_new_label_{}_ood_{}.npy'.format(args.dataset, 'vgg19', 10, args.n_label, args.ood_ratio))
u_label = u_train_dataset.dataset['labels']
if args.cluster:
weight = np.ones([20, 1])
else:
weight = np.ones([len(u_label), 1])
weight = torch.tensor(weight* args.w_initial, dtype=torch.float32, device="cuda", requires_grad=True)
opt_w = optim.Adam([weight], lr=args.meta_lr)
iteration = 0
maximum_val_acc = 0
s = time.time()
start_time = time.time()
for l_data, u_data, v_data in zip(l_loader, u_loader, val_loader_in_train):
iteration += 1
l_input, target, _ = l_data
l_input, target = l_input.to(device).float(), target.to(device).long()
u_input, dummy_target, idx = u_data
u_input, dummy_target = u_input.to(device).float(), dummy_target.to(device).long()
val_input, val_target, _ = v_data
val_input, val_target = val_input.to(device).float(), val_target.to(device).long()
coef = alg_cfg["consis_coef"] * math.exp(-5 * (1 - min(iteration / shared_cfg["warmup"], 1)) ** 2)
if args.cluster:
w_u = weight[emd_label[idx]]
else:
w_u = weight[idx].squeeze()
meta_model = meta_module_wrn.WideResNet(28, dataset_cfg["num_classes"], 2, transform_fn).to(device)
meta_model.load_state_dict(model.state_dict())
outputs_hat = meta_model(u_input)
ssl_loss_meta = ssl_obj(u_input, outputs_hat.detach(), meta_model)
cls_loss_meta = F.cross_entropy(meta_model(l_input), target, reduction="none", ignore_index=-1).mean()
loss_meta = cls_loss_meta + (ssl_loss_meta*w_u).mean() * coef
meta_model.zero_grad()
grads = torch.autograd.grad(loss_meta, meta_model.fc.params(), create_graph=True)
meta_model.update_last_params(lr_inner=alg_cfg["lr"], source_params=grads)
del grads
val_loss_meta = F.cross_entropy(meta_model(val_input), val_target, reduction="none", ignore_index=-1).mean()
val_loss_meta += args.L1_trade_off * torch.sum(torch.abs(weight))
opt_w.zero_grad()
val_loss_meta.backward()
torch.nn.utils.clip_grad_norm_([weight], args.w_clip)
opt_w.step()
weight.data = torch.clamp(weight, min=0.0000000001, max=1)
if args.cluster:
w_up = weight[emd_label[idx]]
else:
w_up = weight[idx].squeeze()
w = w_up.clone().detach()
outputs_u = model(u_input)
ssl_loss = ssl_obj(u_input, outputs_u.detach(), model) * coef
# supervised loss
outputs = model(l_input)
cls_loss = F.cross_entropy(outputs, target, reduction="none", ignore_index=-1).mean()
ssl_loss = (ssl_loss*w).mean()
loss = cls_loss + ssl_loss * coef
optimizer.zero_grad()
loss.backward()
optimizer.step()
if args.alg == "MT" or args.alg == "ICT":
# parameter update with exponential moving average
ssl_obj.moving_average(model.parameters())
# display
if iteration == 1 or (iteration % 100) == 0:
wasted_time = time.time() - s
rest = (shared_cfg["iteration"] - iteration)/100 * wasted_time / 60
print("iteration [{}/{}] cls loss : {:.6e}, SSL loss : {:.6e}, coef : {:.5e}, time : {:.3f} iter/sec, rest : {:.3f} min, lr : {}".format(
iteration, shared_cfg["iteration"], cls_loss.item(), ssl_loss.item(), coef, 100 / wasted_time, rest, optimizer.param_groups[0]["lr"]))
s = time.time()
# validation
if (iteration % args.validation) == 0 or iteration == shared_cfg["iteration"]:
with torch.no_grad():
model.eval()
print("### validation ###")
sum_acc = 0.
s = time.time()
for j, data in enumerate(val_loader):
input, target, _ = data
input, target = input.to(device).float(), target.to(device).long()
output = model(input)
pred_label = output.max(1)[1]
sum_acc += (pred_label == target).float().sum()
if ((j+1) % 10) == 0:
d_p_s = 10/(time.time()-s)
print("[{}/{}] time : {:.1f} data/sec, rest : {:.2f} sec".format(j+1, len(val_loader), d_p_s, (len(val_loader) - j-1)/d_p_s))
s = time.time()
acc = sum_acc/float(len(val_dataset))
print("varidation accuracy : {}".format(acc))
# test
if maximum_val_acc < acc:
print("### test ###")
maximum_val_acc = acc
sum_acc = 0.
s = time.time()
for j, data in enumerate(test_loader):
input, target, _ = data
input, target = input.to(device).float(), target.to(device).long()
output = model(input)
pred_label = output.max(1)[1]
sum_acc += (pred_label == target).float().sum()
if ((j+1) % 10) == 0:
d_p_s = 100/(time.time()-s)
print("[{}/{}] time : {:.1f} data/sec, rest : {:.2f} sec".format(j+1, len(test_loader), d_p_s, (len(test_loader) - j-1)/d_p_s))
s = time.time()
test_acc = sum_acc / float(len(test_dataset))
print("test accuracy : {}".format(test_acc))
# torch.save(model.state_dict(), os.path.join(args.output, "best_model.pth"))
model.train()
s = time.time()
# lr decay
if iteration == shared_cfg["lr_decay_iter"]:
optimizer.param_groups[0]["lr"] *= shared_cfg["lr_decay_factor"]
print("test acc : {}".format(test_acc))
condition["test_acc"] = test_acc.item()
end_time = time.time()
print("running time: ", (start_time-end_time)/60.0, " min")
exp_name += str(int(time.time())) # unique ID
if not os.path.exists(args.output):
os.mkdir(args.output)
with open(os.path.join(args.output, exp_name + ".json"), "w") as f:
json.dump(condition, f)