-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
238 lines (197 loc) · 8.35 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
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
import argparse
import os
from datetime import datetime
import torch
import torch.nn.functional as F
from torch.nn.utils import clip_grad_norm_
from torch import nn
from torchinfo import summary
from datasets import load_dataset
from torchvision.transforms import (CenterCrop, Compose,
ToTensor, RandAugment)
from timm.data.mixup import Mixup
from timm.data.random_erasing import RandomErasing
from sklearn.metrics import classification_report
from tqdm.auto import tqdm
from model import ConvNext
from utils import plot_losses, CustomLogger
def main(args):
mixup_args = {
'mixup_alpha': 0.8,
'cutmix_alpha': 1.0,
'cutmix_minmax': None,
'prob': 0.4,
'switch_prob': 0.5,
'mode': 'elem',
'label_smoothing': 0.1,
'num_classes': args.num_classes
}
mixup = Mixup(**mixup_args)
rand_erasing = RandomErasing(probability=0.25, max_area=1/4, mode="pixel")
# CIFAR10
model = ConvNext(num_channels=3,
num_classes=args.num_classes,
patch_size=4,
layer_dims=[64, 128, 256, 512],
depths=[2, 2, 2, 2],
drop_rate=0.0)
optimizer = torch.optim.AdamW(
model.parameters(),
lr=args.learning_rate,
betas=(args.adam_beta1, args.adam_beta2),
weight_decay=args.adam_weight_decay,
eps=args.adam_epsilon,
)
transforms_train = Compose([
CenterCrop(args.resolution),
RandAugment(num_ops=2),
ToTensor()
])
transforms_test = Compose([
CenterCrop(args.resolution),
ToTensor()
])
if not args.dataset_name:
raise ValueError(
"You must specify a dataset name."
)
train_data, test_data = load_dataset(args.dataset_name,
split=["train", "test"])
def transforms_train_(examples):
images = [
transforms_train(image.convert("RGB"))
for image in examples["img"]
]
labels = [l for l in examples["label"]]
return {"images": images, "labels": labels}
def transforms_test_(examples):
images = [
transforms_test(image.convert("RGB"))
for image in examples["img"]
]
labels = [l for l in examples["label"]]
return {"images": images, "labels": labels}
train_data.set_transform(transforms_train_)
test_data.set_transform(transforms_test_)
train_dataloader = torch.utils.data.DataLoader(
train_data,
batch_size=args.train_batch_size,
shuffle=True)
test_dataloader = torch.utils.data.DataLoader(
test_data, batch_size=args.eval_batch_size, shuffle=False)
lr_scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer,
max_lr=args.learning_rate,
steps_per_epoch=len(train_dataloader),
epochs=args.num_epochs)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)
loss_fnc = nn.CrossEntropyLoss()
current_date = datetime.today().strftime('%Y%m%d_%H%M%S')
logs_path = f"./training_logs/{current_date}/"
os.makedirs(logs_path, exist_ok=True)
logger = CustomLogger("simple-convnext",
file_path=f"{logs_path}/training_log.txt")
model_summary = str(summary(model, (1, 3, args.resolution, args.resolution), verbose=0))
logger.log_info(model_summary)
global_step = 0
losses = []
valid_losses = []
for epoch in range(args.num_epochs):
model.train()
progress_bar = tqdm(total=len(train_dataloader))
progress_bar.set_description(f"Epoch {epoch}")
losses_log = 0
for step, batch in enumerate(train_dataloader):
images = batch["images"].to(device)
labels = batch["labels"].to(device)
images, labels = mixup(images, labels)
images = rand_erasing(images)
preds = model(images)
loss = loss_fnc(preds, labels)
loss.backward()
if args.use_clip_grad:
clip_grad_norm_(model.parameters(), 1.0)
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
losses_log += loss.detach().item()
logs = {
"loss_avg": losses_log / (step + 1),
"loss": loss.detach().item(),
"lr": lr_scheduler.get_last_lr()[0],
"step": global_step
}
progress_bar.set_postfix(**logs)
global_step += 1
progress_bar.close()
losses.append(losses_log / (step + 1))
if epoch % args.save_model_epochs == 0:
model.eval()
valid_loss = 0
with torch.no_grad():
valid_labels = []
valid_preds = []
for step, batch in enumerate(
tqdm(test_dataloader, total=len(test_dataloader))):
images = batch["images"].to(device)
labels = batch["labels"].to(device)
preds = model(images)
loss = loss_fnc(preds, labels)
valid_loss += loss.item()
preds = preds.argmax(dim=-1)
valid_labels.extend(labels.detach().cpu().tolist())
valid_preds.extend(preds.detach().cpu().tolist())
# print for debug
print(f"Valid loss: {valid_loss / len(test_dataloader)}")
print(classification_report(valid_labels, valid_preds))
logger.log_info(f"Epoch {epoch}")
logger.log_info(logs)
logger.log_info(
f"Valid loss: {valid_loss / len(test_dataloader)}")
logger.log_info(classification_report(valid_labels,
valid_preds))
torch.save(
{
'model_state': model.state_dict(),
'optimizer_state': optimizer.state_dict(),
}, args.output_dir)
epoch_path = f"{logs_path}/{epoch}"
os.makedirs(epoch_path)
valid_losses.append(valid_loss / len(test_dataloader))
plot_losses(train_losses=losses,
valid_losses=valid_losses,
path=epoch_path)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="Simple example of a training script.")
parser.add_argument("--dataset_name", type=str, default=None)
parser.add_argument("--dataset_config_name", type=str, default=None)
parser.add_argument("--train_data_dir",
type=str,
default=None,
help="A folder containing the training data.")
parser.add_argument("--output_dir",
type=str,
default="trained_models/cifar10.pth")
parser.add_argument("--resolution", type=int, default=64)
parser.add_argument("--num_classes", type=int, default=10)
parser.add_argument("--train_batch_size", type=int, default=128)
parser.add_argument("--eval_batch_size", type=int, default=128)
parser.add_argument("--num_epochs", type=int, default=100)
parser.add_argument("--save_model_epochs", type=int, default=10)
parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
parser.add_argument("--learning_rate", type=float, default=1e-3)
parser.add_argument("--lr_warmup_steps", type=int, default=500)
parser.add_argument("--adam_beta1", type=float, default=0.9)
parser.add_argument("--adam_beta2", type=float, default=0.999)
parser.add_argument("--adam_weight_decay", type=float, default=5e-2) #1e-1
parser.add_argument("--adam_epsilon", type=float, default=1e-08)
parser.add_argument("--use_clip_grad", type=bool, default=False)
parser.add_argument("--logging_dir", type=str, default="logs")
args = parser.parse_args()
if args.dataset_name is None and args.train_data_dir is None:
raise ValueError(
"You must specify either a dataset name from the hub or a train data directory."
)
main(args)