-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
164 lines (127 loc) · 6.39 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
# USAGE
# python train.py --train-config experiments/albunet_valid/train_config_part0.yaml
import argparse
import os
import torch
from pathlib import Path
from utils import helpers
import albumentations as albu
import importlib
import functools
from pneumothorax_dataset import PneumothoraxDataset, PneumoSampler
from learning import Learning
def argparser():
parser = argparse.ArgumentParser()
parser.add_argument('--train-config', type=str, help='Path to train config file path')
return vars(parser.parse_args())
def train_fold(train_config, experiment_folder, pipeline_name, log_dir, fold_id,
train_dataloader, val_dataloader, binarizer_fn, eval_fn):
fold_logger = helpers.init_logger(log_dir, f'train_log_{fold_id}.log')
best_checkpoint_folder = Path(experiment_folder, train_config['CHECKPOINTS']['BEST_FOLDER'])
best_checkpoint_folder.mkdir(parents=True, exist_ok=True)
checkpoints_history_folder = Path(
experiment_folder,
train_config['CHECKPOINTS']['FULL_FOLDER'],
f'fold_{fold_id}'
)
checkpoints_history_folder.mkdir(parents=True, exist_ok=True)
checkpoints_topk = train_config['CHECKPOINTS']['TOPK']
calculation_name = f'{pipeline_name}_fold_{fold_id}'
device = train_config['DEVICE']
module = importlib.import_module(train_config['MODEL']['PY'])
model_class = getattr(module, train_config['MODEL']['CLASS'])
model = model_class(**train_config['MODEL']['ARGS'])
pretrained_model_config = train_config['MODEL'].get('PRETRAINED', False)
if pretrained_model_config:
loaded_pipeline_name = pretrained_model_config['PIPELINE_NAME']
pretrained_model_path = Path(
pretrained_model_config['PIPELINE_PATH'],
pretrained_model_config['CHECKPOINTS_FOLDER'],
f'{loaded_pipeline_name}_fold_{fold_id}.pth'
)
if pretrained_model_path.is_file():
model.load_state_dict(torch.load(pretrained_model_path))
fold_logger.info(f'Load model from {pretrained_model_path}')
if len(train_config['DEVICE_LIST']) > 1:
model = torch.nn.DataParallel(model)
module = importlib.import_module(train_config['CRITERION']['PY'])
loss_class = getattr(module, train_config['CRITERION']['CLASS'])
loss_fn = loss_class(**train_config['CRITERION']['ARGS'])
optimizer_class = getattr(torch.optim, train_config['OPTIMIZER']['CLASS'])
optimizer = optimizer_class(model.parameters(), **train_config['OPTIMIZER']['ARGS'])
scheduler_class = getattr(torch.optim.lr_scheduler, train_config['SCHEDULER']['CLASS'])
scheduler = scheduler_class(optimizer, **train_config['SCHEDULER']['ARGS'])
n_epochs = train_config['EPOCHS']
grad_clip = train_config['GRADIENT_CLIPPING']
grad_accum = train_config['GRADIENT_ACCUMULATION_STEPS']
early_stopping = train_config['EARLY_STOPPING']
validation_frequency = train_config.get('VALIDATION_FREQUENCY', 1)
freeze_model = train_config['MODEL']['FREEZE']
Learning(
optimizer, binarizer_fn, loss_fn, eval_fn, device, n_epochs, scheduler,
freeze_model, grad_clip, grad_accum, early_stopping, validation_frequency,
calculation_name, best_checkpoint_folder, checkpoints_history_folder,
checkpoints_topk, fold_logger
).run_train(model, train_dataloader, val_dataloader)
if __name__ == '__main__':
args = argparser()
config_file = Path(args['train_config'].strip('/'))
experiment_folder = config_file.parents[0]
train_config = helpers.load_yaml(config_file)
log_dir = Path(experiment_folder, train_config['LOGGER_DIR'])
log_dir.mkdir(parents=True, exist_ok=True)
main_logger = helpers.init_logger(log_dir, 'train_main.log')
seed = train_config['SEED']
helpers.init_seed(seed)
main_logger.info(train_config)
if "DEVICE_LIST" in train_config:
os.environ['CUDA_VISIBLE_DEVICES'] = ','.join(map(str, train_config['DEVICE_LIST']))
pipeline_name = train_config['PIPELINE_NAME']
dataset_folder = train_config['DATA_DIRECTORY']
train_transform = albu.load(train_config['TRAIN_TRANSFORMS'])
val_transform = albu.load(train_config['VAL_TRANSFORMS'])
# ? non_empty_mask_prob/ use_sampler is not sure
non_empty_mask_prob = train_config.get('NON_EMPTY_MASK_PROB', 0)
use_sampler = train_config['USE_SAMPLER']
dataset_folder = train_config['DATA_DIRECTORY']
folds_distr_path = train_config['FOLD']['FILE']
num_workers = train_config['WORKERS']
batch_size = train_config['BATCH_SIZE']
n_folds = train_config['FOLD']['NUMBER']
usefolds = map(str, train_config['FOLD']['USEFOLDS'])
binarizer_module = importlib.import_module(train_config['MASK_BINARIZER']['PY'])
binarizer_class = getattr(binarizer_module, train_config['MASK_BINARIZER']['CLASS'])
binarizer_fn = binarizer_class(**train_config['MASK_BINARIZER']['ARGS'])
eval_module = importlib.import_module(train_config['EVALUATION_METRIC']['PY'])
eval_fn = getattr(eval_module, train_config['EVALUATION_METRIC']['CLASS'])
eval_fn = functools.partial(eval_fn, **train_config['EVALUATION_METRIC']['ARGS'])
for fold_id in usefolds:
main_logger.info(f'Start training of {fold_id} fold...')
train_dataset = PneumothoraxDataset(
data_folder=dataset_folder, mode='train',
transform=train_transform, folder_index=fold_id,
folds_distr_path=folds_distr_path
)
train_sampler = PneumoSampler(folds_distr_path, fold_id, non_empty_mask_prob)
if use_sampler:
train_dataloader = torch.utils.data.DataLoader(
dataset=train_dataset, batch_size=batch_size,
sampler=train_sampler, num_workers=num_workers
)
else:
train_dataloader = torch.utils.data.DataLoader(
dataset=train_dataset, batch_size=batch_size,
sampler=train_sampler, shuffle=True
)
val_dataset = PneumothoraxDataset(
data_folder=dataset_folder, mode='val',
transform=val_transform, folder_index=fold_id,
folds_distr_path=folds_distr_path
)
val_dataloader = torch.utils.data.DataLoader(
val_dataset, batch_size=batch_size, num_workers=num_workers
)
train_fold(
train_config, experiment_folder, pipeline_name, log_dir, fold_id,
train_dataloader, val_dataloader, binarizer_fn, eval_fn
)