-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathmain.py
204 lines (180 loc) · 8.58 KB
/
main.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
import time
import random
import torch
torch.cuda.current_device()
import logging
import argparse
import os
import numpy as np
import warnings
import torch.multiprocessing
torch.multiprocessing.set_sharing_strategy('file_system')
localtime = time.localtime(time.time())
str_time = f'{str(localtime.tm_mon)}-{str(localtime.tm_mday)}-{str(localtime.tm_hour)}-{str(localtime.tm_min)}'
from model.model import MMC
from src.train_food101 import train_food101
from src.config import Config
from src.functions import dict_to_str
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.deterministic = True
def run(args):
if not os.path.exists(args.model_save_dir):
os.makedirs(args.model_save_dir)
if not os.path.exists(args.fig_save_dir):
os.makedirs(args.fig_save_dir)
args.name_seed = args.name + '_' + str(args.seed)
args.model_save_path = os.path.join(args.model_save_dir, f'{args.dataset}-{args.name_seed}-{str_time}.pth')
args.best_model_save_path = os.path.join(args.model_save_dir, f'{args.dataset}-{args.name_seed}-best.pth')
setup_seed(args.seed)
if args.dataset in ['Food101', 'N24News']:
results = train_food101(args)
else:
results = train_medical(args)
return results
def set_log(args):
if not os.path.exists(args.logs_dir):
os.makedirs(args.logs_dir)
log_file_path = os.path.join(args.logs_dir, f'{args.dataset}-{args.name}-{str_time}.log')
# set logging
logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
# logger.setLevel(logging.WARNING)
logging.getLogger('PIL').setLevel(logging.WARNING)
for ph in logger.handlers:
logger.removeHandler(ph)
# add FileHandler to log file
formatter_file = logging.Formatter('%(asctime)s:%(levelname)s:%(message)s', datefmt='%Y-%m-%d %H:%M:%S')
fh = logging.FileHandler(log_file_path)
fh.setLevel(logging.DEBUG)
fh.setFormatter(formatter_file)
logger.addHandler(fh)
# add StreamHandler to terminal outputs
formatter_stream = logging.Formatter('%(message)s')
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
ch.setFormatter(formatter_stream)
logger.addHandler(ch)
return logger
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--name', type=str, default='MMC',
help='project name')
parser.add_argument('--dataset', type=str, default='Food101',
help='support N24News/Food101')
parser.add_argument('--text_type', type=str, default='headline',
help='support headline/caption/abstract')
parser.add_argument('--mmc', type=str, default='UniSMMC',
help='support UniSMMC/UnSupMMC/SupMMC')
parser.add_argument('--mmc_tao', type=float, default=0.07,
help='use supervised contrastive loss or not')
parser.add_argument('--batch_size', type=int, default=32,
help='batch_size')
parser.add_argument('--lr_mm', type=float, default=1e-3,
help='--lr_mm')
parser.add_argument('--min_epoch', type=int, default=1,
help='min_epoch')
parser.add_argument('--valid_step', type=int, default=50,
help='valid_step')
parser.add_argument('--max_length', type=int, default=512,
help='max_length')
parser.add_argument('--text_encoder', type=str, default='bert_base',
help='bert_base/roberta_base/bert_large')
parser.add_argument('--image_encoder', type=str, default='vit_base',
help='vit_base/vit_large')
parser.add_argument('--text_out', type=int, default=768,
help='text_out')
parser.add_argument('--img_out', type=int, default=768,
help='img_out')
parser.add_argument('--lr_mm_cls', type=float, default=1e-3,
help='--lr_mm_cls')
parser.add_argument('--mm_dropout', type=float, default=0.0,
help='--mm_dropout')
parser.add_argument('--lr_text_tfm', type=float, default=2e-5,
help='--lr_text_tfm')
parser.add_argument('--lr_img_tfm', type=float, default=5e-5,
help='--lr_img_tfm')
parser.add_argument('--lr_img_cls', type=float, default=1e-4,
help='--lr_img_cls')
parser.add_argument('--lr_text_cls', type=float, default=5e-5,
help='--lr_text_cls')
parser.add_argument('--text_dropout', type=float, default=0.0,
help='--text_dropout')
parser.add_argument('--img_dropout', type=float, default=0.1,
help='--img_dropout')
parser.add_argument('--nplot', type=str, default='',
help='MTAV')
parser.add_argument('--data_dir', type=str, default='Path/To/Dataset_Home_Directory/',
help='support wmsa')
parser.add_argument('--test_only', type=bool, default=False,
help='train+test or test only')
parser.add_argument('--pretrained_dir', type=str, default='Path/To/Pretrained',
help='path to pretrained models from Hugging Face.')
parser.add_argument('--model_save_dir', type=str, default='Path/To/results/models',
help='path to save model parameters.')
parser.add_argument('--res_save_dir', type=str, default='Path/To/results/results',
help='path to save training results.')
parser.add_argument('--fig_save_dir', type=str, default='Path/To/results/imgs',
help='path to save figures.')
parser.add_argument('--logs_dir', type=str, default='Path/To/results/logs',
help='path to log results.') # NO
parser.add_argument('--local_rank', default=-1, type=int,
help='node rank for distributed training')
parser.add_argument('--seeds', nargs='+', type=int,
help='set seeds for multiple runs!')
return parser.parse_args()
if __name__ == '__main__':
start = time.time()
torch.autograd.set_detect_anomaly(True)
torch.cuda.empty_cache()
warnings.filterwarnings("ignore")
args = parse_args()
logger = set_log(args)
config = Config(args)
args = config.get_config()
if args.local_rank == -1:
device = torch.device("cuda")
else:
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend="nccl")
args.device = device
args.data_dir = os.path.join(args.data_dir, args.dataset)
if args.local_rank in [-1, 0]:
logger.info("Pytorch version: " + torch.__version__)
logger.info("CUDA version: " + torch.version.cuda)
logger.info(f"CUDA device: + {torch.cuda.current_device()}")
logger.info(f"CUDNN version: {torch.backends.cudnn.version()}")
logger.info("GPU name: " + torch.cuda.get_device_name())
logger.info("Current Hyper-Parameters:")
logger.info(args)
final_results = {}
final_std_results = {}
temp_results = {}
# final_std_results
for seed in args.seeds:
args.seed = seed
temp_results = run(args)
if len(final_results.keys()):
for key in temp_results.keys():
final_results[key] += temp_results[key]
final_std_results[key].append(temp_results[key])
else:
final_results = temp_results
final_std_results = {key: [] for key in temp_results.keys()}
for key in temp_results.keys():
final_std_results[key].append(temp_results[key])
if args.local_rank in [-1, 0]:
logger.info(f"Run {len(args.seeds)} times！Final test results:")
for key in final_results.keys():
print(key, ": ", final_std_results[key])
final_std_results[key] = np.std(final_std_results[key])
final_results[key] = final_results[key] / len(args.seeds)
logger.info(f"{args.dataset}-{args.name}")
logger.info("Average: " + dict_to_str(final_results))
logger.info("Standard deviation: " + dict_to_str(final_std_results))
end = time.time()
logger.info(f"Run {end - start} seconds in total！")