Repository navigation
Expand file tree
/
Copy pathtrain.py
More file actions
207 lines (168 loc) · 7.95 KB
/
Copy pathtrain.py
File metadata and controls
207 lines (168 loc) · 7.95 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
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
import argparse
import sys
import time
import torch
import wandb
import pandas as pd
import os
import torch.nn as nn
import torch.nn.functional as F
import torchvision
from torchvision import transforms
import wandb
from utils import *
from utils import print_args, str2bool, get_grad_norm_squared
from models import ResNet18, vgg11_bn
def main():
arguments = sys.argv
print("Command-line arguments:", arguments)
gpu = [0]
parser = argparse.ArgumentParser()
parser.add_argument('--project', default='test_results', type=str)
parser.add_argument('--use_wandb', type=str2bool, nargs='?',const=True)
parser.add_argument('--uid', type=str)
parser.add_argument('--dataset', type=str)
parser.add_argument('--model', type=str)
parser.add_argument('--epochs', type=int)
parser.add_argument('--optimizer', type=str)
parser.add_argument('--seed', type=int)
parser.add_argument('--bs', type=int)
parser.add_argument('--lr', type=float)
parser.add_argument("--lr_decay", type=str2bool, nargs='?',const=True)
parser.add_argument('--wd', type=float)
parser.add_argument('--beta1', default=0, type=float)
parser.add_argument('--beta2', default=0.999, type=float)
args = parser.parse_args()
print_args(args)
########### Setting Up Seed ###########
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
########### Setting Up GPU ###########
torch.cuda.set_device(gpu[0])
device = 'cuda'
gpu_index = torch.cuda.current_device()
gpu_name = torch.cuda.get_device_name(gpu_index)
print(f"Using GPU {gpu_index}: {gpu_name}")
########### Setting Up wandb ###########
if args.use_wandb:
run=wandb.init(project=args.project,config=vars(args), dir="/wandb_tmp")
print(vars(args))
########### Setup Data and Model ###########
if args.dataset=="cifar10": #only option!
#data transforms
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
#data
train_dataset = torchvision.datasets.CIFAR10(root='./data', download=True, train=True, transform=transform_train)
validation_dataset = torchvision.datasets.CIFAR10(root='./data', download=True, train=False, transform=transform_test)
#trainloader subset
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=int(args.bs), shuffle=True, num_workers=2, pin_memory=True, persistent_workers=True)
#testloader
validation_loader = torch.utils.data.DataLoader(dataset=validation_dataset, batch_size=int(args.bs), shuffle=False, num_workers=8, pin_memory=True, persistent_workers=True)
#models
if args.model == "vgg11":
model = torch.nn.DataParallel(vgg11_bn(),gpu).cuda()
elif args.model == "res18":
model = torch.nn.DataParallel(ResNet18(),gpu).cuda()
else:
raise NotImplementedError("Model not defined")
criterion = nn.CrossEntropyLoss()
else:
raise NotImplementedError("Dataset not defined")
##### iteration counter
iteration = 0
total_steps = int(len(train_loader)*args.epochs)
########### Getting number of layers ###########
n_groups = 0
dim_model = 0
with torch.no_grad():
for param in model.parameters():
n_groups = n_groups + 1
dim_model = dim_model + torch.numel(param)
print('Model dimension: ' + str(dim_model))
print('Number of groups: ' + str(n_groups))
print('Number of iterations: ' + str(total_steps))
print(f'Steps per epoch: {len(train_loader)}')
########### Init of Optimizers ###########
if args.optimizer == 'sgd':
opt = torch.optim.SGD(model.parameters(), lr=args.lr, momentum = args.beta1, weight_decay= args.wd)
if args.lr_decay:
scheduler = torch.optim.lr_scheduler.StepLR(opt, step_size=int(args.epochs*0.75), gamma=0.1)
if args.optimizer == 'adam':
opt = torch.optim.AdamW(model.parameters(), lr=args.lr, betas = (args.beta1, args.beta2), weight_decay= args.wd, eps = 1e-8 )
if args.lr_decay:
scheduler = torch.optim.lr_scheduler.StepLR(opt, step_size=int(args.epochs*0.75), gamma=0.1)
df = [] #stats saved here
filename = args.model+'_'+args.dataset+'_'+args.optimizer+'_s'+str(args.seed)+'_lr'+str(args.lr)+'_decay'+str(args.lr_decay)+'_uid'+str(args.uid)+'.csv'
print('saving in '+str(filename))
########### Training ###########
for epoch in range(args.epochs):
start_time = time.time()
########### Saving stats every few epochs ###########
model.eval()
#computing stats: train loss
train_loss, correct = 0, 0
for d in train_loader:
data, target = d[0].to(device, non_blocking=True),d[1].to(device, non_blocking=True)
output = model(data)
train_loss += criterion(output, target).data.item()/len(train_loader)
pred = output.data.max(1)[1]
correct += pred.eq(target.data).cpu().sum()
accuracy_train = 100. * correct.to(torch.float32) / len(train_loader.dataset)
#computing stats: test loss
test_loss, correct, total = 0, 0, 0
for d in validation_loader:
data, target = d[0].to(device, non_blocking=True),d[1].to(device, non_blocking=True)
output = model(data)
test_loss += criterion(output, target).data.item()/len(validation_loader)
pred = output.data.max(1)[1]
correct += pred.eq(target.data).cpu().sum()
accuracy_test = 100. * correct.to(torch.float32) / len(validation_loader.dataset)
#saving to wandb
if args.use_wandb:
wandb.log({"train_loss":train_loss, "train_acc":accuracy_train,"test_loss":test_loss,"test_acc":accuracy_test}, commit=False)
print('Epoch {}: Train L: {:.4f}, TrainAcc: {:.2f}, Test L: {:.4f}, TestAcc: {:.2f} \n'.format(epoch, train_loss, accuracy_train, test_loss, accuracy_test))
########### Training Loop ###########
model.train()
for _, batch in enumerate(train_loader):
opt.zero_grad()
model.zero_grad()
########### Backprop ###########
data, target = batch[0].to(device),batch[1].to(device)
output = model(data)
loss = criterion(output, target)
loss.backward()
# Getting gradient norm squared
with torch.no_grad():
grad_norm_squared = get_grad_norm_squared(model)
if args.use_wandb:
wandb.log({"grad_norm_squared":grad_norm_squared}, commit=False)
########### Optimizer update for standard methods ###########
effective_lr = opt.param_groups[0]["lr"]
if args.use_wandb:
wandb.log({"effective_lr":effective_lr}, commit=False)
df.append({'model': args.model, 'data': args.dataset, 'opt':args.optimizer, 'seed': args.seed, 'base_lr': args.lr, 'decay_lr': args.lr_decay, 'loss': loss.item(), 'epoch': epoch, 'lr': effective_lr})
opt.step()
iteration = iteration +1
if args.lr_decay:
scheduler.step()
epoch_time = time.time()-start_time
if args.use_wandb:
wandb.log({"epoch_time":epoch_time})
pd.DataFrame(df).to_csv(os.path.join('results', filename)) #backup
########### Closing Writer ###########
if args.use_wandb:
run.finish()
wandb.finish()
return None
if __name__ == '__main__':
main()