-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrainer.py
254 lines (211 loc) · 10.5 KB
/
trainer.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
from __future__ import print_function
import torch
import argparse
import sys
from torch.utils.data import DataLoader,random_split
from torchvision import datasets, models, transforms
import platform
from datetime import timedelta, date, datetime
from model import vdsr
import torch.nn as nn
import numpy as np
import torch.optim as optim
import os
from math import log10
import time
from util.PrepareData import ACCESS_BARRA_v2_0,ACCESS_BARRA_v2_1,ACCESS_BARRA_vdsr
def write_log(log,args):
print(log)
if not os.path.exists("./save/"+args.train_name+"/"):
os.mkdir("./save/"+args.train_name+"/")
my_log_file=open("./save/"+args.train_name + '/train.txt', 'a')
# log="Train for batch %d,data loading time cost %f s"%(batch,start-time.time())
my_log_file.write(log + '\n')
my_log_file.close()
return
def evaluation(net,val_dataloders,loss,criterion,args):
net.eval()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
test_loss=0
avg_psnr = 0
start=time.time()
with torch.no_grad():
for batch, (pr,hr,_,_) in enumerate(val_dataloders):
pr,hr = prepare([pr, hr],device,args)
sr = net(pr)
val_loss=criterion(sr, hr)
test_loss+=val_loss.item()
psnr = 10 * log10(1000 / val_loss.item())
avg_psnr += psnr
write_log("evalutaion: time cost %f s, test_loss: %f, psnr: avg_psnr %f"%(
time.time()-start,
test_loss/(batch + 1),
avg_psnr / len(val_dataloders)
),args)
return test_loss
def prepare(l,device=False,args=0):
def _prepare(tensor):
if args.precision == 'half': tensor = tensor.half()
if args.precision == 'single': tensor = tensor.float()
return tensor.to(device)
return [_prepare(_l) for _l in l]
def train_main():
# ===========================================================
# Training settings
# ===========================================================
parser = argparse.ArgumentParser(description='PyTorch Super Res Example')
# Hardware specifications
parser.add_argument('--n_threads', type=int, default=0,
help='number of threads for data loading')
parser.add_argument('--cpu', action='store_true',help='cpu only?')
parser.add_argument('--test', action='store_true',help='cpu only?')
# hyper-parameters
parser.add_argument('--train_name', type=str, default="vdsr_pr", help='training name')
parser.add_argument('--batch_size', type=int, default=4, help='training batch size')
parser.add_argument('--testBatchSize', type=int, default=4, help='testing batch size')
parser.add_argument('--nEpochs', type=int, default=200, help='number of epochs to train for')
parser.add_argument('--lr', type=float, default=0.01, help='Learning Rate. Default=0.01')
parser.add_argument('--seed', type=int, default=123, help='random seed to use. Default=123')
# model configuration
parser.add_argument('--upscale_factor', '-uf', type=int, default=4, help="super resolution upscale factor")
parser.add_argument('--model', '-m', type=str, default='vdsr', help='choose which model is going to use')
#data
parser.add_argument('--pr', type=bool, default=True,help='add-on pr?')
parser.add_argument('--train_start_time', type=type(datetime(1990,1,25)), default=datetime(1990,1,2),help='r?')
parser.add_argument('--train_end_time', type=bool, default=datetime(1990,2,9),help='?')
parser.add_argument('--test_start_time', type=bool, default=datetime(2012,1,1),help='a?')
parser.add_argument('--test_end_time', type=bool, default=datetime(2012,12,24),help='')
parser.add_argument('--start_date', type=bool, default=datetime(1990,1,2),help='a?')
parser.add_argument('--end_date', type=bool, default=datetime(2012,12,24),help='')
parser.add_argument('--dem', action='store_true',help='add-on dem?')
parser.add_argument('--psl', action='store_true',help='add-on psl?')
parser.add_argument('--zg', action='store_true',help='add-on zg?')
parser.add_argument('--tasmax', action='store_true',help='add-on tasmax?')
parser.add_argument('--tasmin', action='store_true',help='add-on tasmin?')
parser.add_argument('--leading_time_we_use', type=int,default=1,help='add-on tasmin?')
parser.add_argument('--ensemble', type=int, default=11,help='total ensambles is 11')
parser.add_argument('--channels', type=float, default=0,help='channel of data_input must')
#[111.85, 155.875, -44.35, -9.975]
parser.add_argument('--domain', type=list, default=[112.9, 154.25, -43.7425, -9.0],help='dataset directory')
parser.add_argument('--file_ACCESS_dir', type=str, default="../data/",help='dataset directory')
parser.add_argument('--file_BARRA_dir', type=str, default="../data/barra_aus/",help='dataset directory')
parser.add_argument('--file_DEM_dir', type=str, default="../DEM/",help='dataset directory')
parser.add_argument('--precision', type=str, default='single',choices=('single', 'half','double'),help='FP precision for test (single | half)')
args = parser.parse_args()
# init_date=date(1970, 1, 1)
# start_date=date(1990, 1, 2)
# end_date=date(2011,12,25)
sys = platform.system()
if sys == "Windows":
init_date=date(1970, 1, 1)
args.start_date=date(1990, 1, 2)
args.end_date=date(1990,2,9) #if 929 is true we should substract 1 day
# args.file_ACCESS_dir="E:/climate/access-s1/"
# args.file_BARRA_dir="C:/Users/JIA059/barra/"
args.file_DEM_dir="../DEM/"
args.channels=0
if args.pr:
args.channels+=1
if args.zg:
args.channels+=1
if args.psl:
args.channels+=1
if args.tasmax:
args.channels+=1
if args.tasmin:
args.channels+=1
if args.dem:
args.channels+=1
print("training statistics:")
print(" ------------------------------")
print(" trainning name | %s"%args.train_name)
print(" ------------------------------")
print(" num of channels | %5d"%args.channels)
print(" ------------------------------")
print(" num of threads | %5d"%args.n_threads)
print(" ------------------------------")
print(" batch_size | %5d"%args.batch_size)
print(" ------------------------------")
print(" using cpu only | %5d"%args.cpu)
lr_transforms = transforms.Compose([
transforms.Resize((316, 376)),
# transforms.RandomResizedCrop(IMG_SIZE),
# transforms.RandomHorizontalFlip(),
# transforms.RandomRotation(30),
transforms.ToTensor()
# transforms.Normalize(IMG_MEAN, IMG_STD)
])
hr_transforms = transforms.Compose([
# transforms.Resize((316, 376)),
# transforms.RandomResizedCrop(IMG_SIZE),
# transforms.RandomHorizontalFlip(),
# transforms.RandomRotation(30),
transforms.ToTensor()
# transforms.Normalize(IMG_MEAN, IMG_STD)
])
data_set=ACCESS_BARRA_vdsr(args.start_date,args.end_date,lr_transform=lr_transforms,hr_transform=hr_transforms,args=args)
if args.test:
print(1)
else:
train_data,val_data=random_split(data_set,[int(len(data_set)*0.8),len(data_set)-int(len(data_set)*0.8)])
#######################################################################
train_dataloders =DataLoader(train_data,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.n_threads)
val_dataloders =DataLoader(val_data,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.n_threads)
#######################################################################
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net=vdsr()
# criterion = nn.MSELoss(size_average=False)
criterion=nn.L1Loss()
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=0.9, weight_decay=1e-4)
# optimizer_my = optim.SGD(net.parameters(), lr=args.lr, momentum=0.9)
scheduler = optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.99)
if torch.cuda.device_count() > 1:
write_log("!!!!!!!!!!!!!Let's use"+str(torch.cuda.device_count())+"GPUs!",args)
net = nn.DataParallel(net,range(torch.cuda.device_count()))
else:
write_log("Let's use"+str(torch.cuda.device_count())+"GPUs!",args)
net.to(device)
##############################################
write_log("start",args)
max_error=np.inf
val_max_error=np.inf
for e in range(args.nEpochs):
loss=0
start=time.time()
for batch, (pr,hr,_,_) in enumerate(train_dataloders):
pr,hr= prepare([pr,hr],device,args)
optimizer.zero_grad()
with torch.set_grad_enabled(True):
sr = net(pr)
running_loss =criterion(sr, hr)
running_loss.backward()
optimizer.step()
loss+=running_loss.item()
if batch%10==0:
state = {'model': net, 'optimizer': optimizer.state_dict(), 'epoch': e}
torch.save(state, "./save/"+args.train_name+"/last.pth")
write_log("epoche: %d,lr: %f,time cost %f s, train_loss: %f "%(
e,
optimizer.state_dict()['param_groups'][0]['lr'],
time.time()-start,
loss/(batch + 1),
),args)
scheduler.step()
if e%10==0:
test_loss=evaluation(net,val_dataloders,loss,criterion,args)
if test_loss<val_max_error:
write_log("saveing model for best test",args)
state = {'model': net, 'optimizer': optimizer.state_dict(), 'epoch': e}
torch.save(state, "./save/"+args.train_name+"/best_test_"+str(e)+".pth")
if max_error>loss:
write_log("saveing model for best train",args)
state = {'model': net, 'optimizer': optimizer.state_dict(), 'epoch': e}
torch.save(state, "./save/"+args.train_name+"/best_train_"+str(e)+".pth")
if __name__=='__main__':
train_main()