-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_base.py
136 lines (107 loc) · 3.16 KB
/
main_base.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
import os
from typing import Tuple, Union, List, Callable, Optional
from tqdm import tqdm
from itertools import islice
import dataclasses
import numpy as np
import pandas as pd
import torch
import torch.nn.functional as F
from torch import nn
from torch import distributions
from torch.utils.data import DataLoader, Dataset, WeightedRandomSampler
import torchaudio
from IPython import display as display_
from collections import defaultdict
from IPython.display import clear_output
from matplotlib import pyplot as plt
import matplotlib.pyplot as plt
from thop import profile
import tempfile
import wandb
from data.data import get_dataloaders
from melspecs import get_melspecs
from base_config import BaseConfig
from models import CRNN
import utils
from train_distil import train_distillation
from train import train_base
import argparse
from pathlib import Path
from datetime import datetime
WANDB_GROUP = 'Base'
PARAMS = {
'num_epochs': 200,
'learning_rate': 3e-4,
'alpha': 0.9,
'temp': 20,
'hidden_size': 12,
'cnn_out_channels': 3,
'gru_num_layers': 2,
'kernel_size': (5, 24),
'stride': (3, 15),
'dropout': 0.,
}
def main(args, params={}):
utils.fix_seed(args.seed)
save_dir = Path(f'save/{WANDB_GROUP}') / datetime.now().strftime(r"%m%d_%H%M%S")
if not os.path.isdir(save_dir):
os.makedirs(save_dir, exist_ok=True)
config = BaseConfig(**params)
config.use_wandb = args.use_wandb
train_loader, val_loader = get_dataloaders(config.keyword)
melspec_train, melspec_val = get_melspecs(config)
base_model = CRNN(config).to(config.device)
print(base_model)
opt = torch.optim.Adam(
base_model.parameters(),
lr=config.learning_rate,
weight_decay=config.weight_decay
)
scheduler = (
torch.optim.lr_scheduler.CosineAnnealingLR(opt, config.num_epochs, eta_min=5e-5)
if args.use_sched
else None
)
exp_name = f'{WANDB_GROUP}_sched={args.use_sched}'
if config.use_wandb:
wandb.init(project='dla-kws', config=dataclasses.asdict(config), name=exp_name, group='base')
wandb.watch(base_model)
history = train_base(base_model, train_loader, val_loader, melspec_train, melspec_val, opt, config)
print('FINAL_METRIC', history[-1])
save_path = str(save_dir / (exp_name + '.tar'))
utils.save_model(base_model, save_path)
wandb.finish()
return history
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="KWS")
parser.add_argument(
"-s",
"--seed",
default=911,
type=int,
help="Seed (default: 911)",
)
parser.add_argument(
"-sc",
"--use_sched",
default=False,
type=bool,
help="Use scheduler (default: False)",
)
parser.add_argument(
"-w",
"--use_wandb",
default=True,
type=bool,
help="Use wandb to log data",
)
args = parser.parse_args()
params = {
'num_epochs' : 30
}
for lr in [3e-4, 5e-4]:
for seed in [99, 911, 119, 9, 123, 42]:
params['learning_rate'] = lr
args.seed = seed
main(args, params)