-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun.py
201 lines (175 loc) · 6.94 KB
/
run.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
import dataset
import pandas as pd
import random
import numpy as np
import os
import torch
import models
import augmentation
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss
from transformers import get_linear_schedule_with_warmup
import functions
class args:
DEBUG = False
amp = False
wandb = False
exp_name = "resnest50_5fold_base"
network = "AudioClassifier"
pretrain_weights = None
model_param = {
'encoder' : 'resnest50d',
'sample_rate': 48000,
'window_size' : 1024 * 2,
'hop_size' : 512 * 2,
'classes_num' : 24
}
losses = "BCEWithLogitsLoss"
lr = 1e-3
step_scheduler = True
epoch_scheduler = False
period = 60
seed = 1988
start_epoch = 0
epochs = 50
batch_size = 8
num_workers = 4
early_stop = 10
device = ('cuda' if torch.cuda.is_available() else 'cpu')
train_csv = "train_folds.csv"
test_csv = "test_df.csv"
sub_csv = "D:\\Kaggle_Bird_Frog\\sample_submission.csv"
output_dir = "weights"
def main(fold):
# Setting seed
seed = args.seed
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
args.fold = fold
args.save_path = os.path.join(args.output_dir, args.exp_name)
os.makedirs(args.save_path, exist_ok=True)
train_df = pd.read_csv(args.train_csv)
#test_df = pd.read_csv(args.test_csv)
sub_df = pd.read_csv(args.sub_csv)
if args.DEBUG:
train_df = train_df.sample(1000)
train_fold = train_df[train_df.kfold != fold]
valid_fold = train_df[train_df.kfold == fold]
train_dataset = dataset.AudioDataset(
df=train_fold,
period=args.period,
transforms=augmentation.augmenter,
train=True,
data_path="D:\\Kaggle_Bird_Frog\\train"
)
valid_dataset = dataset.AudioDataset(
df=valid_fold,
period=args.period,
transforms=None,
train=True,
data_path="D:\\Kaggle_Bird_Frog\\train"
)
test_dataset = dataset.AudioDataset(
df=sub_df,
period=60,
transforms=None,
train=False,
data_path="D:\\Kaggle_Bird_Frog\\test"
)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=args.batch_size,
shuffle=True,
drop_last=True,
num_workers=args.num_workers
)
valid_loader = torch.utils.data.DataLoader(
valid_dataset,
batch_size=args.batch_size,
shuffle=False,
drop_last=False,
num_workers=args.num_workers
)
test_loader = torch.utils.data.DataLoader(
test_dataset,
batch_size=args.batch_size//2,
shuffle=False,
drop_last=False,
num_workers=args.num_workers
)
model = models.__dict__[args.network](**args.model_param)
model = model.to(args.device)
if args.pretrain_weights:
print("---------------------Loading pretrain weights")
model.load_state_dict(torch.load(args.pretrain_weights,
map_location=args.device)["model"],
strict=False)
model = model.to(args.device)
criterion = BCEWithLogitsLoss()
optimizer = torch.optim.AdamW(model.parameters(), lr=args.lr)
num_train_steps = int(len(train_loader) * args.epochs)
num_warmup_steps = int(0.1 * args.epochs * len(train_loader))
scheduler = get_linear_schedule_with_warmup(optimizer,
num_warmup_steps=num_warmup_steps,
num_training_steps=num_train_steps)
best_lwlrap = -np.inf
for epoch in range(args.start_epoch, args.epochs):
train_avg, train_loss = functions.train_epoch(args,
model,
train_loader,
criterion,
optimizer,
scheduler,
epoch)
valid_avg, valid_loss = functions.valid_epoch(args,
model,
valid_loader,
criterion,
epoch)
if args.epoch_scheduler:
scheduler.step()
content = f"""
{time.ctime()} \n
Fold:{args.fold}, Epoch:{epoch}, lr:{optimizer.param_groups[0]['lr']:.7}\n
Train Loss:{train_loss:0.4f} - LWLRAP:{train_avg['lwlrap']:0.4f}\n
Valid Loss:{valid_loss:0.4f} - LWLRAP:{valid_avg['lwlrap']:0.4f}\n
"""
print(content)
with open(f'{args.save_path}/log_{args.exp_name}.txt', 'a') as appender:
appender.write(content+'\n')
if valid_avg['lwlrap'] > best_lwlrap:
print(f"########## >>>>>>>> Model Improved From {best_lwlrap} ----> {valid_avg['lwlrap']}")
torch.save(model.state_dict(), os.path.join(args.save_path,
f'fold-{args.fold}.bin'))
best_lwlrap = valid_avg['lwlrap']
#torch.save(model.state_dict(), os.path.join(args.save_path, f'fold-{args.fold}_last.bin'))
model.load_state_dict(torch.load(os.path.join(args.save_path,
f'fold-{args.fold}.bin'),
map_location=args.device))
model = model.to(args.device)
target_cols = sub_df.columns[1:].values.tolist()
test_pred, ids = functions.test_epoch(args, model, test_loader)
print(np.array(test_pred).shape)
test_pred_df = pd.DataFrame({
"recording_id" : sub_df.recording_id.values
})
test_pred_df[target_cols] = test_pred
test_pred_df.to_csv(os.path.join(args.save_path,
f"fold-{args.fold}-submission.csv"),
index=False)
print(os.path.join(args.save_path, f"fold-{args.fold}-submission.csv"))
oof_pred, ids = functions.test_epoch(args, model, valid_loader)
oof_pred_df = pd.DataFrame({
"recording_id" : ids
})
oof_pred_df[target_cols] = oof_pred
oof_pred_df.to_csv(os.path.join(args.save_path,
f"oof-fold-{args.fold}.csv"),
index=False)
if __name__ == "__main__":
for fold in range(5):
if fold == 0:
main(fold)